The Structure and Grounding of Epistemic Justification

Home , Barry Smith (ontologist), Coherentism, Evidentialism, George Pappas, Louise Antony, William Lycan

THE STRUCTURE AND GROUNDING OF

EPISTEMIC JUSTIFICATION

DISSERTATION

Presented in Partial Fulfillment of the Requirements for

The Degree Doctor of Philosophy in the Graduate

School of the Ohio State University

William A. Roche, B.S., M.S.

* * * * *

The Ohio State University

2006

Dissertation Committee: Approved by

Professor George Pappas, Advisor

Professor Louise Antony ______

Professor William Taschek Advisor

Graduate Program in Philosophy

William A. Roche

2006

ABSTRACT

I articulate and defend a new version of the coherence theory of epistemic justification. It is new, in part, because, unlike traditional varieties of coherentism, it is externalist—viz., it has the consequence that justification supervenes, in part, on things that are neither mental nor supervenient on the mental. In other words, my theory entails that there could be mental duplicates (i.e., cognizers with the same beliefs, the same experiences, etc.) whose beliefs differ in justification.

The theory, overall, has three distinguishing components. First, there is an explanationist component, which says, in short, that an inductive inference (or inferential relation) is cogent only if it is explanatorily virtuous—where one way to be explanatorily virtuous is to be an instance of inference to the best explanation. In this respect, my account is in the spirit of both

William Lycan’s brand of coherentism, and the brand oft attributed to Gilbert Harman and

Wilfrid Sellars. Second, there is a meta-perspectivalist component, which says (among other things) that S has good reason for thinking that p obtains only if from S’s perspective it is likely that he is connected (e.g., causally) to p. This, when fully spelled out, has the result that S’s belief system is coherent only if S has a view as to how he (or, better, his belief system) is connected to the outside world, and according to which the mechanisms involved (e.g., vision) are reliable. And third, there is a veridicality component, which requires that S’s reasons be true, and that S’s reasons for his reasons be true, and so on. This, together with the meta- perspectivalist component, requires that S be correct as to how he is reliably connected to the outside world. This is what makes my position externalist. I argue positively for each of these

three distinguishing components, and, in doing so, refute my theory’s chief rivals in the coherentist camp.

It is widely thought that coherence theories are hopeless, because of objections such as the Alternative-Systems Objection (which starts with the claim that there are lots of incompatible yet fully coherent belief systems such that any belief whatsoever is a member of at least one such system) and the Isolation Objection (which focuses on the alleged fact that coherence neither involves nor requires any sort of connection, such as a causal connection, to the outside world). I argue against these objections on three fronts. I argue that the coherentist can be an externalist, and that the objections fail against externalist coherentism. I argue that internalist foundationalism is faced with similar objections—so that were internalist coherentism to fall, internalist foundationalism too would fall. And, last, I argue against the objections directly, arguing that each version thereof, when fully fleshed out, has either a false premise or a fallacious sub-argument. The upshot is that, contrary to orthodox epistemology, there is nothing to be learned—vis-à-vis the question of whether justification is skyscraper-like in structure (as the foundationalist supposes)—from either the Alternative-Systems Objection or the Isolation

Objection.

It is also widely thought that, even setting aside these stock objections, foundationalist

theories are superior to coherentist theories because, unlike coherentist theories, they allow a role

for experiences in justification. With coherentist theories, the space of reasons (to borrow a

phrase from Sellars) is restricted to the space of beliefs—so that only beliefs can serve as reasons.

With foundationalist theories, in contrast, the space of reasons includes not just the space of

beliefs, but also the space of experiences. The view, in brief, is that some beliefs are justified

non-inferentially by experiences (e.g., visual experiences), and that the beliefs so justified serve

as the foundation of justification. I argue that, initial appearances notwithstanding, experiences

are unfit to serve as reasons—either for or against beliefs. I argue in addition, however, that there

is a contingent respect in which experiences are nonetheless required for justification, in that,

iii

though they cannot themselves serve as reasons, they can (and contingently do) help in enabling beliefs to serve as reasons.

The ultimate payoff is philosophical understanding of the structure and grounding of justification—an understanding, that is, that justification is coherentist in structure and externalist

in grounding.

To my parents,

Kent and Brenda

ACKNOWLEDGMENTS

I am greatly, if not immeasurably, indebted to George Pappas, my advisor. He has thoughtfully read and commented on several versions of nearly every chapter, and—both in person and in writing—has always served for me as a model of how to do philosophy properly.

This project has benefited immensely from his guidance (though, as he is a reliabilist and this project is a defense of coherentism, probably not as much as he would have liked).

I owe a special thanks to Louise Antony and William Taschek (my other two readers),

Robert Batterman, and to the rest of the faculty in the philosophy department here at the Ohio

State University. The encouragement and critical feedback that I have received from them have helped tremendously.

I have discussed nearly every aspect of this dissertation with William Melanson, Joshua

Smith, and my brother, Michael. This has aided not just in clarifying and strengthening my

overall position, but also in keeping what turned out to be a rather cumbersome task relatively

enjoyable.

I should also mention T. M. Reed, whose mentoring (while I was an undergraduate in

philosophy at the University of Utah) I will always remember, and will surely never be able to

repay.

To all these people (and countless others), I give my sincere thanks.

VITA

BACKGROUND AND EDUCATION

July 4, 1975, Born, Pocatello, Idaho

1997, B.S., University of Utah, Philosophy

1998, M.S., University of Utah, Philosophy

1999 - 2002, Teaching Assistant, The Ohio State University, Philosophy

2002 - Present, Graduate Teaching Associate, The Ohio State University, Philosophy

FIELDS OF STUDY

Major Field: Philosophy

Areas of Specialization: Epistemology, Philosophy of Science

vii

TABLE OF CONTENTS

ABSTRACT...... II

ACKNOWLEDGMENTS ...... VI

VITA...... VII

1 THE VIEW—ROUGHLY SKETCHED...... 1 1.1 The Question ...... 1 1.2 The Foundationalism/Coherentism Debate ...... 6 1.3 Some Examples...... 11 1.4 Looking Ahead...... 14 1.4.1 Moderate Explanationism...... 16 1.4.2 Meta-Perspectivalism ...... 22 1.4.3 Veridicalism...... 23 1.5 Conclusion...... 30 2 THE ALTERNATIVE-SYSTEMS AND ISOLATION OBJECTIONS...... 33 2.1 The Objections ...... 34 2.1.1 The Alternative-Systems Objection (ASO) ...... 35 2.1.2 The Isolation Objection (IO) ...... 38 2.2 The First Rebuttal...... 40 2.3 The Second Rebuttal ...... 44 2.4 The Third Rebuttal ...... 50 2.4.1 Against the ASO 1...... 50 2.4.2 Against the ASO 2...... 52 2.4.3 Against the ASO 3...... 53 2.4.4 Against the IO 1...... 58 2.4.5 Against the IO 2...... 59 2.5 Conclusion...... 60 3 THE DATABASE OBJECTION...... 63 3.1 The Objection...... 64 3.1.1 Lycan ...... 65 3.1.2 Cornman and Pappas ...... 66 3.1.3 Lehrer...... 75 3.1.4 Summary...... 77 3.2 The First Version Rebutted ...... 78 3.3 The Second Version Rebutted...... 84 3.4 The Third Version Rebutted...... 85

viii

3.5 Conclusion...... 86 4 THE EXPERIENCE OBJECTION...... 89 4.1 The Objection...... 89 4.2 The Trilemma Challenge...... 91 4.2.1 Accounts on which the Target Beliefs are about the External World...... 93 4.2.2 Accounts on which the Target Beliefs are about Experiences...... 96 4.3 Accounts that Ignore the How-Question...... 100 4.4 Acquaintance Accounts...... 103 4.4.1 Chalmers...... 104 4.4.2 Fumerton...... 108 4.4.3 Fales...... 109 4.4.4 Peacocke ...... 114 4.4.5 BonJour...... 119 4.4.6 Brewer ...... 128 4.4.7 The Lesson...... 131 4.5 Inference Accounts...... 132 4.5.1 Peacocke ...... 132 4.5.2 Moser...... 134 4.5.3 Heck...... 138 4.6 Reliabilist Accounts ...... 141 4.7 Conclusion...... 145 5 MODERATE EXPLANATIONISM ...... 148 5.1 Bestness, Loveliness, and Likeliness ...... 150 5.2 Deductivism and Potential Explainers ...... 155 5.3 Explanation...... 158 5.4 Inductive Generalization ...... 159 5.4.1 Lycan’s Improvements ...... 161 5.4.2 Two Objections...... 163 5.4.3 Fumerton’s Objection ...... 164 5.4.4 Kitcher-Unification...... 167 5.4.5 The Solution ...... 170 5.5 Other Inductive Argument Forms ...... 172 5.5.1 Argument From Authority...... 173 5.5.2 Randomized Experimental Studies...... 176 5.5.3 Mill’s Methods ...... 180 5.5.4 Argument From Analogy...... 182 5.5.5 Statistical Syllogism ...... 185 5.6 Conclusion...... 196 6 META-PERSPECTIVALISM...... 199 6.1 In Search of a Rationale ...... 200 6.1.1 BonJour...... 200 6.1.2 Lehrer...... 202 6.1.3 Lycan ...... 204 6.1.4 The Rationale...... 205 6.1.5 A Possible Reply ...... 208 6.2 In Search of the Right Variety...... 210 6.2.1 Accounts Requiring Justified Meta-Beliefs and Justified Reliability Beliefs...... 210 6.2.2 Accounts Requiring Meta-Beliefs and Justified Reliability Beliefs...... 211

6.2.3 Accounts Requiring Justified Reliability Beliefs...... 214 6.2.4 Accounts Requiring Justified Connection Beliefs ...... 216 6.2.5 The Account ...... 216 6.3 The Over-Demandingness Charge ...... 223 6.4 Conclusion...... 228 7 VERIDICALISM...... 231 7.1 A Potential Objection...... 232 7.2 An Analogy to Logic and Practical Reason ...... 235 7.3 Justification and Truth...... 239 7.4 Non-Epistemic Justification ...... 242 7.5 The Veridicality Requirement...... 243 7.6 Is Experience Essential After All? ...... 247 7.7 Is Truth Required for Justification?...... 248 7.8 Is Just One False Belief Enough to Undermine Justification? ...... 249 7.9 Conclusion...... 251 BIBLIOGRAPHY...... 253

CHAPTER 1

THE VIEW—ROUGHLY SKETCHED

1.1 The Question

There is a bevy of well-known cases from Laurence BonJour and Keith Lehrer suggesting that there is an intimate connection between justification and having good reason, such that S’s belief that p is justified only if S has good reason for thinking that p. Each case is one in which externalist conditions such as reliability are present, in which S does not have good reason for thinking that p, and in which—intuitively—S’s belief that p is unjustified.

BonJour starts with a case in which, though S’s belief that p is the result of a reliable

belief-forming process, S has good (but misleading) reason for thinking that not-p:

Samantha believes herself to have the power of clairvoyance, though she has no reasons for or against this belief. One day she comes to believe, for no apparent reason, that the President is in New York City. She maintains this belief, appealing to her alleged clairvoyant power, even though she is at the same time aware of a massive amount of apparently cogent evidence, consisting of news reports, press releases, allegedly live television pictures, and so on, indicating that the President is at that time in Washington, D.C. Now the President is in fact in New York City, the evidence to the contrary being part of a massive official hoax mounted in the face of an assassination threat. Moreover, Samantha does in fact have completely reliable clairvoyant power under the conditions which were then satisfied, and her belief about the President did result from the operation of that power. (BonJour 1985, 38)

The intuition is supposed to be that, though it is the result of her reliable clairvoyance,

Samantha’s belief about the President is unjustified—given that she does not have good reason for thinking that the President is in New York City.

BonJour then gives two slightly modified cases, in which S’s conflicting evidence is itself slightly modified. In the first modified case, the subject, Casper, has good reason for thinking

that his clairvoyance is unreliable. In the second, the subject, Maud, has good reason for thinking that reliable clairvoyance is nomologically impossible. The intuition, again, is supposed to be that, though it is the result of a reliable belief-forming process, the subject’s belief about the

President’s whereabouts is unjustified—given that the subject does not have good reason for thinking that the President is in New York City.

And last, BonJour gives a case in which S is slightly better off, in not have any conflicting evidence:

Norman, under certain conditions which usually obtain, is a completely reliable clairvoyant with respect to certain kinds of subject matter. He possesses no evidence or reasons of any kind for or against the general possibility of such a cognitive power or for or against the thesis that he possesses it. One day Norman comes to believe that the President is in New York City, though he has no evidence either for or against his belief. In fact the belief is true and results from his clairvoyant power under circumstances in which it is completely reliable. (Ibid., 41)

Though perhaps slightly less so than Samantha, Casper, and Maud, Norman is being epistemically irrational, or irresponsible, in believing that the President is in New York City—or so thinks BonJour. The intuition is supposed to be that, because of this, Norman’s belief is unjustified.

Lehrer, in turn, gives a case similar in structure to BonJour’s Norman case, but with the

advantage of being more readily imaginable:

Suppose a person, Mr. Truetemp, undergoes brain surgery by an experimental surgeon who invents a small device that is both a very accurate thermometer and a computational device capable of generating thoughts. The device, call it a tempucomp, is implanted in Truetemp’s head so that the very tip of the device, no larger than the head of a pin, sits unnoticed on his scalp and acts as a sensor to transmit information about the temperature to the computational system in his brain. This device, in turn, sends a message to his brain causing him to think of the temperature recorded by the external sensor. Assume that the tempucomp is very reliable, and so his thoughts are correct temperature thoughts. All told, this is a reliable belief-forming process and a properly functioning cognitive faculty. Now imagine, finally, that Mr. Truetemp has no idea that the tempucomp has been inserted in his brain and is only slightly puzzled about why he thinks so obsessively about the temperature; but he never checks a thermometer to determine whether these thoughts about the temperature are correct. He accepts them unreflectively, another effect of the tempucomp. Thus, he thinks and accepts that the temperature is 104 degrees. It is. Does he know that it is? Surely not. He has no idea whether he or his thoughts about the temperature are reliable. (Lehrer 2000, 187)

The supposed problem, again, is that, though it is the result of a reliable belief-forming process,

Truetemp’s belief that the temperature is 104 degrees is unjustified (or not a piece of knowledge)—sinceTruetemp has no reason whatsoever for thinking that it is 104 degrees.

The suggested hypothesis, from these cases, is that justification requires having good reason. This, if true, would account for each case, since in each case there is neither justification

nor good reason.

But, of course, there are other hypotheses, and ones that do not require having good reason. Alvin Goldman, for instance, couples a reliability condition with an availability condition, requiring that S not have a reliable process such that if he had used it, he would not have believed that p:

S’s belief that p is justified only if (1) it is the result of a reliable belief-forming process and (2) S does not have an alternative process such that if he had used it, he would not have believed that p.

Goldman argues that using one’s evidence (or the lack thereof) properly is a reliable belief- forming process, and that, thus, his account implies that cases in which, though S’s belief that p is the result of a reliable process, S does not have good reason for thinking that p are cases of unjustified belief. Here is what he says, in “What is Justified Belief?” (1979b), about a case in

which S has good reason for thinking that not-p:

Jones has strong evidence against certain propositions concerning his past. He doesn’t use this evidence, but if he were to use it properly, he would stop believing these propositions. Now the proper use of evidence would be an instance of a (conditionally) reliable process. So what we can say about Jones is that he fails to use a certain (conditionally) reliable process that he could and should have used. (Goldman 1992, 123)

And here is what he says, in Epistemology and Cognition (1986), about BonJour’s Norman case, in which S, Norman, has neither evidence for p nor evidence against p:

BonJour describes this case as one in which Norman possesses no evidence or reasons of any kind for or against the general possibility of clairvoyance, or for or against the thesis that he possesses it. But it is hard to envisage this description holding. Norman ought to reason along the following lines: ‘If I had a clairvoyant power, I would surely find some evidence for this. I would find myself believing things in otherwise inexplicable ways, and when these things were checked by other reliable processes, they would usually check out positively. Since I lack any such signs, I apparently do not possess reliable

clairvoyant processes.’ Since Norman ought to reason this way, he is ex ante justified in believing that he does not possess reliable clairvoyant processes. This undermines her belief in N [i.e., that the President is in New York City]. (Goldman 1986, 112)

The point, then, is that the hypothesis according to which justification requires having good reason is not the only hypothesis that can account for cases such as BonJour’s clairvoyance cases and Lehrer’s tempucomp case.

The proponent of the having-good-reason hypothesis could, at this point, appeal to evil- demon cases in which S’s belief that p is not the result of a reliable belief-forming process, in which S has good reason for thinking that p, and in which—intuitively—S’s belief that p is justified. The having-good-reason hypothesis can explain such cases, but, it would seem, the reliability hypothesis cannot.

In turn, the proponent of the reliability hypothesis could, for instance, relativize the relevant reliability facts to the actual world—so that whether a belief produced by process r in world w is justified hinges on whether r is reliable in the actual world.1 The reliability hypothesis

would then be able to account for evil-demon cases, since S, in such cases, has the same belief-

forming processes as do we, and since (presumably) our belief-forming processes are reliable.

The proponent of the reliability hypothesis might even take the offensive at this point, pointing, first, to children and animals, and, second, to lost-evidence cases. Neither children nor animals, the idea goes, are sophisticated enough to have good reason, so that the having-good- reason hypothesis incorrectly says that neither children nor animals have justified beliefs. And since S, in lost-evidence cases, no longer has his original evidence for p, he no longer has good reason for thinking that p—so that, again, the having-good-reason hypothesis gives the wrong result. The reliability hypothesis, in contrast, can account for each such kind of case, in that each such kind of case is a case of reliably-produced belief.

And so it goes, back and forth between the proponent of the having-good-reason hypothesis and the proponent of the reliability hypothesis. The one starts with his favorite cases, and then either modifies his account or bites the bullet in response to the other’s favorite cases.

It is important to see that though I shall be defending a version of coherentism and though it requires having good reason, I mean to stay neutral on the reliability hypothesis. It seems to me that there is a rather ordinary sense of the word “justification” that is synonymous with the words

“good reason”, so that the sentence “S’s belief that p is justified.” is equivalent in meaning to the sentence “S has good reason for thinking that p.”. I have in mind everyday scenarios in which someone asks for a person’s justification for a claim, where what is being asked for is the person’s (allegedly good) reason for the claim. But I am not concerned to argue that this is so, and am especially not concerned to argue that there is no sense of justification on which having good reason is not required, and which is best explained by the reliability hypothesis. The aim, in defending a version of coherentism, is to explain, or make intelligible, what it would be to have good reason. That is, the aim is to make sense of the difference between beliefs for which the subject has good reason and beliefs for which the subject does not have good reason. Thus though I shall speak in terms of justification and though, in the dialectic sketched above, the proponent of the having-good-reason hypothesis is diametrically opposed to the proponent of the reliability hypothesis, I shall remain agnostic on the reliability hypothesis—allowing for the possibility that there is a sense of justification on which having good reason is not required, and

which is best explained by the reliability hypothesis.

But what is a coherentist theory of justification, in contrast to a foundationalist theory of justification? Or, equivalently, what is a coherentist theory of what it would be to have good reason, in contrast to a foundationalist theory of what it would be to have good reason? I shall say a few things in answer to this question in the next section, and shall give a few examples in the section after the next. In the final section, I shall give a sketch of the version of coherentism to be defended in this dissertation.

1.2 The Foundationalism/Coherentism Debate

Much has been said on the foundationalism/coherentism debate, both as to what the debate is about and as to who the winner is.2 I shall thus make just a few comments on it, highlighting three questions on which the foundationalist and the coherentist disagree. This section will be important later, especially in the next chapter; the stock anti-coherentism objections are premised on an inaccurate understanding of coherentism.

The first question on which the foundationalist and the coherentist disagree is the question of whether there is non-inferential justification, which is justification independent of support from other beliefs. The foundationalist answers in the affirmative, oftentimes arguing that the key to non-inferential justification is sensory experience. The coherentist, in contrast, answers in the negative—so that a belief is justified only if it is supported by other beliefs.

The point of the foundationalist’s regress argument, of course, is to establish just such a claim (viz., that there is non-inferential justification). Suppose, the argument goes, that there were no non-inferential justification. Given this, S’s belief b1 would be justified only if it were inferentially justified. That is, b1 would be justified only if there were a belief b2 such that b2 was already justified and b2 stood in a good inferential, or logical, relation to b1. b2, in turn, would be justified only if there were a belief b3 such that b3 was already justified and b3 stood in a good

inferential relation to b2. And so on. This would either stop with an unjustified belief bn, or circle

3 back to b1, or continue on without end. Either way, b1 would not be justified. So if there were no non-inferential justification, there would be no justification at all. But obviously, there is justification. Thus, there is non-inferential justification.

You might think that, in addition to the regress argument, the foundationalist could establish the claim that there is non-inferential justification by simply pointing to a justified perceptual belief—since a perceptual belief, of course, is non-inferential (i.e., not inferred from other beliefs). But this would be to misunderstand the claim that there is non-inferential justification. The claim is not that there is a justified belief that is not inferred from other beliefs.

Even the coherentist thinks that there is non-inferential justification in that sense. Rather, the claim, to oversimplify a bit, is that there is a justified belief for which the subject has no doxastic reasons—viz., other beliefs that logically (i.e., deductively or inductively) support it.4 Hence what would work, in establishing the claim that there is non-inferential justification, is not a justified belief that is not inferred from other beliefs, but a justified belief for which the subject has no doxastic reasons.

The second question on which the foundationalist and the coherentist disagree is the

question of how inferential justification works. The foundationalist, on one hand, says that

inferential justification is part-to-part and uni-directional. What makes it part-to-part is that it

moves between beliefs—the parts of the belief system. b1 is justified inferentially by b2, in virtue of b2’s already being justified and b2’s standing in a good inferential relation to b1. b2, in turn, is

justified inferentially by b3, in virtue of b3’s already being justified and b3’s standing in a good inferential relation to b2. b1, thus, gets its justification from b2, which, in turn, gets its justification

5 from b3. What makes it uni-directional is that it moves in just one direction. b2 is prior to b1 and

hence gives justification to, but does not receive justification from, b1. b3, in turn, is prior to b2,

so that it gives justification to, but does not receive justification from, b2. The coherentist, on the

other hand, says that inferential justification is whole-to-part, not part-to-part. b2 stands in a good

inferential relation to b1, just as the foundationalist supposes. b3 stands in a good inferential

relation to b2—again, just as the foundationalist supposes. However, it is D (the belief system), not b2, that gives justification to b1, and it is D, not b3, that gives justification to b2. There is justification transfer, but instead of moving from part to part (i.e., from belief to belief), it moves from whole to part (i.e., from system to belief). The difference, in short, is that inferential justification is linear for the foundationalist, but holistic for the coherentist.6

Let me stress that this talk of a belief system’s being justified could simply be replaced by talk of a belief system’s being justification-conferring. The point, either way, is that

inferential justification, on the whole-to-part conception, is a function of playing an inferential role in a system of the right kind.

The dialectic runs as follows. The foundationalist gives the regress argument, and in doing so assumes (quite commonsensically) that inferential justification is part-to-part and uni- directional. The coherentist agrees with the second part of the argument, that if the regress (once started) were to stop with an unjustified belief bn, or circle back to b1, or continue on without end,

then b1 would not be justified. He further agrees with the third part, that there is justification.

The fourth part, however, he finds entirely mysterious. How could it be, he queries, that a belief

is non-inferentially justified? He thus rejects the first part, moving from the part-to-part-and-uni-

directional conception of inferential justification to the whole-to-part conception—which keeps

the regress from ever starting.

The third, and final, question on which the foundationalist and the coherentist disagree is the question of what grounds inferential justification. The foundationalist points to non- inferential justification, saying that all inferentially justified beliefs get their justification, in the end, from non-inferentially justified beliefs.7,8 The beliefs in the foundation, which are justified

non-inferentially, inferentially justify the beliefs on the first floor (so to speak), which, in turn,

inferentially justify the beliefs on the second floor, and so on. The coherentist, in contrast, points

(at least in part) to coherence.9 The view is that an inferentially justified belief gets its justification from the subject’s belief system, which, in turn, gets (at least part of) its justification from its being coherent.

The parentheticals “at least in part” and “at least part of” are significant, in that they allow for coherentists who require for justification much more than just coherence. Consider, for instance, the view that a belief system is justified just in case (1) it is coherent and (2) each of its beliefs is the result of a reliable belief-forming process. Though it says that inferential justification is partially grounded in reliability, it also says that inferential justification is partially grounded in coherence—and thus is a kind of coherentism.

This, it seems to me, is the right result, since any view on which there is no non- inferential justification, on which inferential justification is whole-to-part, and on which inferential justification is partially grounded in coherence (and not even partially grounded in non-inferential justification) is diametrically opposed to the three core tenets of foundationalism.

Consider, again, the hybrid view sketched in the previous paragraph. It is in conflict with foundationalism on whether there is non-inferential justification, since it says that justification requires having a reason, and that only beliefs can serve as reasons. It is in conflict with foundationalism on how inferential justification works, given that it says that inferential justification is whole-to-part (not part-to-part-and-uni-directional). Third, it is in conflict with foundationalism on what grounds inferential justification, pointing not to non-inferential justification (i.e., justified beliefs the justification of which is independent of any support from other beliefs) but to coherence and reliability. Hence it is coherentist in a very natural sense— viz., in being diametrically opposed to foundationalism, and in making heavy use of coherence.

But this, of course, is just a taxonomic issue, and, thus, is of no serious theoretical consequence. The point is the same either way: there are non-foundationalist views on which inferential justification is holistic, on which inferential justification is partially grounded in coherence, and on which inferential justification is partially grounded in things other than coherence.

I should pause, before moving on, to note two very different readings of BonJour’s discussion of non-linear, or holistic, justification, so as to further clarify the way in which the coherentist is thinking of inferential justification. Here he seems to articulate and endorse the whole-to-part conception of inferential justification, as I delineated it above:

The epistemic issue on a particular occasion will usually be merely the justification of a single empirical belief, or small set of such beliefs, within the context of a cognitive system whose overall justification is (more or less) taken for granted; we may call this the local level of justification. But it is also possible, at least in principle, to raise the issue of the overall justification of the entire system of empirical beliefs; we may call this the global level of justification. For the sort of coherence theory which will be developed here—and indeed, I would argue, for any comprehensive, nonskeptical epistemology—it

is the issue of justification as it arises at the latter, global, level which is in the final analysis decisive for the determination of empirical justification in general. (BonJour 1985, 91)

And here he seems to say that whether a belief system is justified hinges on whether it is coherent:

According to the envisaged coherence theory, the relation between the various particular beliefs is correctly to be conceived, not as one of linear dependence, but rather as one of mutual or reciprocal support. There is no ultimate relation of epistemic priority among the members of such a system and consequently no basis for a true regress. Rather the component beliefs of such a coherent system will ideally be so related that each can be justified in terms of the others, with the direction of argument on a particular occasion of local justification depending on which belief (or set of beliefs) has actually been challenged in that particular occasion. And hence, a coherence theory will claim, the apparent circle of justification is not in fact vicious because it is not genuinely a circle: the justification of a particular empirical belief finally depends, not on other particular beliefs as the linear conception of justification would have it, but instead on the overall system and its coherence. (Ibid., 91-2)

But here he seems to say something quite different:

According to this conception, the fully explicit justification of a particular empirical belief would involve four distinct main steps or stages of argument, as follows: (1) The inferability of that particular belief from other particular beliefs and further relations among particular empirical beliefs. (2) The coherence of the overall system of empirical beliefs. (3) The justification of the overall system of empirical beliefs. (4) The justification of the particular belief in question, by virtue of its membership in the system. The claim of a coherence theory of empirical justification is that each of these steps depends on the ones which precede it. (Ibid., 92)

In speaking of (1)-(4) as steps, or stages, of argument, he seems to be speaking of (1)-(4) as claims that S needs to believe, and for which S needs to have good reason. Working backwards, the picture looks like this. What explains why S has good reason for thinking that p is S’s belief, for which he has good reason, that his belief that p is a member of a justified belief system. What explains why S has good reason for thinking this, in turn, is S’s belief, for which he has good reason, that his belief that p is a member of a coherent belief system. And finally, what explains why S has good reason for thinking this is both (a) S’s belief, for which he has good reason, that his belief that p is logically (i.e., deductively or inductively) supported by other beliefs in his system and (b) S’s beliefs, for which he has good reason, about the support relations between the

other beliefs in his system. This is quite different from the picture suggested in the first two passages—in which (a) what explains why S has good reason for thinking that p is that his belief system is justified, (b) what explains this is (at least in part) that his belief system is coherent, and

(c) what explains this is the set of support relations between the beliefs in the system (where at least one such relation involves the belief that p).

I am not concerned, now, with settling the question of which reading is correct, but am instead concerned with stressing that the picture involved in the second reading is drastically at odds with the picture I am attributing to the coherentist. The coherentist, in pushing the whole-to- part conception of inferential justification, is not saying that S needs to think that his belief system is justified. Nor is he saying that S needs to think that his belief system is coherent.

Rather, the coherentist (or at least the kind who says that coherence is all that matters for justification) is saying that the mere fact that S’s belief system is coherent is sufficient for the justification of S’s belief system, and that the mere fact that S’s belief system is justified is sufficient for the inferential justification of S’s beliefs.

The foundationalism/coherentism debate, in essence, is about the structure of justification. The foundationalist—arguing that there is non-inferential justification, that inferential justification is part-to-part and uni-directional, and that inferential justification is grounded in non-inferential justification—argues that justification is skyscraper-like in structure.

The coherentist—arguing that there is no non-inferential justification, that inferential justification is whole-to-part, and that inferential justification is grounded (at least partially) in coherence— argues that the system, not some special subset of it, is the primary seat of justification.

1.3 Some Examples

It will prove helpful to look at a few paradigm instances of foundationalism, as well as a paradigm instance of coherentism; I will use Paul Moser’s theory in Knowledge and Evidence and

Alvin Goldman’s theory in Epistemology and Cognition as paradigm instances of

foundationalism, and Keith Lehrer’s theory in Theory of Knowledge as a paradigm instance of coherentism. This will help both in illustrating the rather abstract claims from the previous section, and in setting up later discussions (in this chapter and others).

Moser’s account says that the key to justification is experience. The beliefs in the superstructure are justified in virtue of logical relations—either deductive, or inductive but explanatory—tracing back, in the end, to the beliefs in the foundation. The beliefs in the foundation, in turn, are justified in virtue of explanatory inferential relations to experiences:

S’s belief that p (bp) is non-inferentially justified just in case (1) S has an experience e with subjective non-conceptual contents c, (2) p is the best explanation of why e has c, (3) p’s explaining e’s having c is uncontravened (all told), (4) S has associated, or is associating, p and c, and (5) bp is based on c.

Suppose, to illustrate, that S is having a visual experience the non-conceptual contents of which

consist of an apparent blue book. Suppose further that, relative to S’s other experiences, the best

explanation of why S is having a visual experience of an apparent blue book is that, in fact, there

is a blue book before him. Then (assuming the association condition and the basing condition are

met) S’s belief that there is a blue book before him is justified, and non-inferentially so.

Goldman’s account too is a kind of foundationalism, though not a kind on which

experience is central to non-inferential justification. The key for Goldman is reliability. The

beliefs in the foundation are justified in virtue of coming from reliable non-inferential (in the

psychological sense) belief-forming processes (e.g., vision)—viz., processes that involve no

reasoning, and thus no premise beliefs. The beliefs in the superstructure, in contrast, are justified

in virtue of coming from reliable inferential belief-forming processes (e.g., inductive

generalization) the premise beliefs of which are themselves justified.

The important point is that, though they are in significant disagreement on the nature of

justification (with Moser stressing experience and explanation and Goldman stressing reliability),

Moser and Goldman are in complete agreement on the structure of justification. For both

theorists, there is non-inferential justification. Moser says that S’s belief that p is justified—even

in the absence of any support from other beliefs—if S has an experience e with non-conceptual contents c, and p is the best explanation of why e has c. In the same way, Goldman says that S’s belief that p is justified—even in the absence of any support from other beliefs—if it is the result of a reliable non-inferential belief-forming process. And for both theorists, inferential justification is grounded—via linear support relations—in non-inferential justification. Moser

appeals to deductive relations and explanatory relations, so that an inferentially justified belief is

justified in virtue of a chain of deductive and/or explanatory relations ending in at least one non-

inferentially justified belief. Goldman appeals not (explicitly) to deductive relations and

explanatory relations, but to reliability—saying that a chain of inferential (in the psychological

sense) relations is justification-yielding only if each link in the chain, or each process involved, is

reliable.

Lehrer, of course, is in full disagreement with Moser and Goldman on the structure of justification, arguing that coherence, not experience or reliability, is the ultimate source of justification:

S’s belief that p is justified just in case p coheres with the rest of S’s belief system D.

p coheres with the rest of D just in case all objections to p on D are either answered on D or neutralized on D.

o is an objection to p on D just in case the degree to which it is reasonable for S to accept that p on D on the assumption that o is true is lower than the degree to which it is reasonable for S to accept that p on D on the assumption that o is false.

o is answered on D just in case the degree to which it is reasonable for S to accept that p on D is higher than the degree to which it is reasonable for S to accept that o on D.

o is neutralized on D just in case there is a proposition n such that (a) S accepts that n, (b) o&n is not an objection to p on D, and (c) the degree to which it is reasonable for S to accept that o&n on D is higher than or equal to the degree to which it is reasonable for S to accept that o on D.

The view, in effect, is that S’s belief that p is justified just in case S, just given his other beliefs, can answer or neutralize every possible attack on p.10

There is no explicit mention in any of this of the whole-to-part conception of justification, or even of coherence as a property of a belief system. Instead, there is mention only of coherence as a relation between beliefs. But notice two things. First, there is no room for non- inferential justification. Whether every objection to p on D is answered on D or neutralized on D hinges on whether D includes beliefs that logically support p—since, to focus just on answering objections, this is what makes it more reasonable to accept p on D than to accept the objections.

Hence, there is no cohering with the rest of D—and thus no justification—in the absence of logical support from other beliefs in D. Second, there is no room for justification without system- wide coherence. It would not be reasonable, for instance, for S to accept that it has rained

recently—even given his belief that the street is wet and his belief that the only thing that could

explain this is that it has rained recently—if it were not reasonable for S to accept that the street is

wet, or if it were not reasonable for S to accept that the only thing that could explain the street’s

being wet is that it has rained recently. The supporting beliefs too would need logical support

from other beliefs in S’s belief system, as would the supporting beliefs for the supporting beliefs,

and so on. So though there is no explicit mention of coherence as a property of a belief system, it

is nonetheless required for justification—given that coherence as a relation between beliefs,

which is explicitly mentioned, requires coherence as a property of a belief system.

1.4 Looking Ahead

The kind of coherentism to be defended in this dissertation has three distinguishing components. First, it has an explanationist component, according to which an inductive (or probabilistic) inferential relation is cogent only if it is explanatorily virtuous—as in inference to the best explanation. In this respect, my account is in the spirit of both William Lycan’s brand of coherentism, and the brand oft attributed to Gilbert Harman and Wilfrid Sellars. Second, it has a meta-perspectivalist component, which says, in part, that S has good reason for thinking that p obtains only if from S’s perspective it is likely that he is connected (e.g., causally) to p—that is,

only if relative to the rest of S’s beliefs it is likely that he is connected to p. This, when fully spelled out, has the result that S’s belief system is coherent only if S has a view as to how he (or, better, his belief system) is connected to the outside world, and according to which the mechanisms involved (e.g., vision) are reliable. And third, it has a veridicality component, which requires that S’s reasons be true, and that S’s reasons for his reasons be true, and so on. This component, when coupled with the meta-perspectivalist component, ensures (among other things) that S is right about the ways in which he is reliably connected to the world. The overall theory

(hereafter “EMV”), thus, has both an internalist side and an externalist side. Things need to be right on the inside, in that S’s beliefs need to hang together in the right way—where this (i.e., the right way) is specified by the explanationist component and the meta-perspectivalist component.

But things also need to be right on the outside, which is where the veridicality component comes into play.

EMV, in being explanationist in the way that it is, in being meta-perspectivalist in the way that it is, and in being veridicalist in the way that it is, is a novel version of coherentism.

Hence the arguments to be given in favor of EMV’s three distinguishing components will tell simultaneously against EMV’s three main coherentist competitors: Lehrer’s account, BonJour’s account, and Lycan’s account.

The aim now is to further flesh out EMV’s distinguishing components, and, while doing so, to highlight some, but not all, of the respects in which EMV differs from its three main coherentist rivals—thus going part, but not all, of the way to establishing the claim that if the arguments to be given (in chapters 5, 6, and 7) in favor of EMV’s distinguishing components go through, then EMV’s most formidable rivals in the coherentist camp are false. I shall start with the explanationist component, moving next to the meta-perspectivalist component, and then, last, to the veridicality component.

1.4.1 Moderate Explanationism

The brand of explanationism in EMV sanctions just two basic inductive inferential- relation forms:

Inference to the Best Explainer (1) q obtains. (2) p’s obtaining is a potential explainer for q’s obtaining. (3) With respect to the available potential explainers for q’s obtaining, p’s obtaining is the best. (4) p’s obtaining is good enough as a potential explainer for q’s obtaining. ------(5) p obtains.

Inference to an Explainee (1) q obtains. (2) q’s obtaining is a potential explainer for p’s obtaining; in other words, p’s obtaining is a potential explainee for q’s obtaining. (3) There are no interfering factors. ------(4) p obtains.

In each form, there is a premise saying that such and such phenomenon obtains. In each form, there is a premise saying that there is such and such potential explanatory relation between the

first phenomenon’s obtaining and a second phenomenon’s obtaining. And in each form, the

conclusion says that the second phenomenon obtains. The main difference is that whereas in the

first such form the inferential relation moves from explainee to explainer, in the second such form

the inferential relation moves from explainer to explainee.

It might be helpful to give a few examples of inference to an explainee, as it is much less

discussed in philosophy than is inference to the best explainer. Here are two from Harman, who

(as far as I know) was the first to write about inference to an explainee (which he calls “inference

from an explanation”):

For example, one infers that a person will be at a certain place at a certain time, given that he now intends to be there then. One infers that his present intention will explain his being there. Competing alternatives are that something will prevent his appearance or that he will change his mind or forget. One’s inference is warranted to the extent that its conclusion is sufficiently more plausible than these competing alternatives. Again, one infers on a particular occasion that sugar will dissolve when stirred into one’s tea. That is, one infers that the solubility of sugar in hot tea will explain the

dissolving of this sugar in this tea. The conclusion is acceptable to the extent that it is sufficiently more plausible than competing alternatives such as that some factor or other will prevent the dissolving this time. (Harman 1975, 268)

And here is one from Peter Achinstein:

That Jones has a severe chest wound can be veridical evidence that he will die, even though the hypothesis that he will die does not explain the fact that he has a severe chest wound. Rather the reverse explanation is correct: he will die because he has a severe chest wound. (Achinstein 1983, 324)

In each of the three inferences, the move is from explainer to explainee—from the person’s intending to be there to the person’s being there, from the sugar’s solubility to the sugar’s dissolving, from Jones’s chest wound to Jones’s dying.

I said at the start of this section that the brand of explanationism in EMV sanctions just two basic inductive inferential-relation forms. The qualifier “basic” is meant to allow for inferential-relation forms such as the following, where q’s obtaining neither potentially explains nor is potentially explained by p’s obtaining, but where, nonetheless, the claim that q obtains makes it likely that p obtains:

(1) q obtains. (2) r’s obtaining is a potential explainer for q’s obtaining. (3) With respect to the available potential explainers for q’s obtaining, r’s obtaining is the best. (4) r’s obtaining is good enough as a potential explainer for q’s obtaining. ------(5) r obtains. (6) r’s obtaining is a potential explainer for p’s obtaining. (7) There are no interfering factors. ------(8) p obtains.

The inference from the claim that the street is wet to the claim that the roof is wet, for instance, has this non-basic form. The potential explanatory relation between the street’s being wet and the roof’s being wet is not direct, in that the street’s being wet neither potentially explains nor is potentially explained by the roof’s being wet. Rather, it is indirect: its just now raining both potentially explains the street’s being wet, and potentially explains the roof’s being wet. And it is this, according to the brand of explanationism in EMV, that makes it such that the claim about the

street is a good, though less than fully conclusive, (provisional) reason for thinking that the roof is wet.11

The brand of explanationism in EMV is weaker than at least some brands of explanationism, in that, in contrast to at least some brands of explanationism, it allows for good but non-explanatory deductive inferential relations. The claim is not that an inferential relation is

good only if it is explanatorily virtuous, but that an inductive inferential relation is good only if it

is explanatorily virtuous. Let the weaker brand be “moderate explanationism”, and the stronger

brand (according to which all good inferential relations are explanatorily virtuous) be “strong

explanationism”.12

Much more needs to be said, of course—both by way of clarification, and by way of argument. What makes it such p’s obtaining is a potential explainer for q’s obtaining? What makes it such that p’s obtaining is the best of the available potential explainers for q’s obtaining?

Why think that all inductive inferential relations are instances of either a basic or a non-basic explanatory inferential-relation form? And so on. The point for now is just that EMV is moderate explanationist, and so says that all good inductive inferential relations (and some, but not all, good deductive inferential relations) are explanatorily virtuous, meaning that each such inferential relation is an instance of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee.

The coherentist says that a belief is justified only if it is logically (i.e., deductively or inductively) supported by the subject’s other beliefs, and only if the belief system of which it is a member is coherent. But what would it be, on EMV, for a belief to be logically supported by the subject’s other beliefs? And what would it be, on EMV, for a belief system to be coherent? This is where, on EMV, moderate explanationism comes into play. With respect to the first question, it entails that a belief is inductively supported by the subject’s other beliefs only if there is a virtuous explanatory inferential relation moving to it from the subject’s other beliefs—viz., an

inferential relation that is an instance of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee. With respect to the second question, it entails that an inductive inferential relation is coherence-enhancing—and thus figures in the set of reciprocal support relations (i.e., the hanging together) on which coherence depends—only if it is an instance of either inference to the best explainer, or inference to an explainee, or an explanatory

inferential-relation form built up out of inference to the best explainer and/or inference to an

explainee.

This (i.e., the sense in which moderate explanationism figures in spelling out what it

would be, on EMV, for a belief system to be coherent) bears further elucidation, as the nature of

coherence is the key issue on which rival coherentists disagree. I shall draw on some material

from BonJour, in which he gives a schema for coherence—a schema to which all coherentists,

both explanationist coherentists and non-explanationist coherentists alike, can, and should,

subscribe. It frames nicely the sense in which moderate explanationism helps in spelling out what

it would be, on EMV, for a belief system to be coherent.

It would seem that, for any coherentist, coherence comes in degrees, and that, (again) for

any coherentist, the degree to which a belief system is coherent hinges on at least five things.

First, there is whether the system is logically consistent:

If D is logically consistent, if E is logically inconsistent, and if other things are equal, then D is more coherent than E.

The beliefs in D would better hang together than would the beliefs in E—given that, because of the inconsistency, the beliefs in E would be more at odds with each other than would be the beliefs in D. Second, there is the degree to which the system is probabilistically consistent:

If D is more probabilistically consistent than E and if other things are equal, then D is more coherent than E—where D is more probabilistically consistent than E if the number of probabilistically inconsistent pairs of beliefs (e.g., the belief that p and the belief that p is unlikely) in D is less than the number of probabilistically inconsistent pairs of beliefs in E.13

The beliefs in D would hurt, or repel, each other less so than would the beliefs in E, and thus in that sense the beliefs in D would better hang together than would the beliefs in E. Third, there is the ratio of supported beliefs (i.e., beliefs for which the subject has doxastic reasons) to total beliefs:

If the ratio of supported beliefs in D to total beliefs in D is higher than the ratio of supported beliefs in E to total beliefs in E and if other things are equal, then D is more coherent than E.

There would be more togetherness in D than in E, since more (percentage-wise) would be included in the reciprocal support in D than in E; there would be less (percentage-wise) without support in D than in E. Fourth, there is the strength of the inferential relations:

If the inferential relations standing between the beliefs in D are stronger (with deductive entailment being the strongest) on average than are the inferential relations standing between the beliefs in E and if other things are equal, then D is more coherent than E.14

The beliefs in D would better hang together than would the beliefs in E, in that, though the ratio of supported beliefs in E to total beliefs in E would be equal to the ratio of supported beliefs in D to total beliefs in D, the hanging in D would be tighter on average than would be the hanging in

E. And fifth, there is the number of sub-systems into which the system is divided:

If D has less sub-systems than E and if other things are equal, then D is more coherent than E—where s and t are sub-systems just in case there are no good inferential relations standing between the beliefs in s and the beliefs in t.

There would be less estrangement in D than there would be in E, and so, again, the beliefs in D

(taken as a group) would better hang together than would the beliefs in E (taken as a group).

It is with respect to the third and fourth factors that, on EMV, moderate explanationism is most central. It entails that whether an inductive inferential relation contributes to the third factor—viz., the ratio of supported beliefs to total beliefs—hinges on whether it is explanatorily virtuous. It also entails that the degree to which an inductive inferential relation, if good, contributes to the fourth factor—viz., the strength of the inferential relations—hinges on the degree to which it is explanatorily virtuous (e.g., the degree to which, in an instance of inference to the best explainer, the best potential explainer is good enough as a potential explainer).

In contrast to moderate explanationist coherentist accounts such as EMV, there are strong explanationist coherentist accounts, and non-explanationist coherentist accounts. For strong explanationist coherentist accounts, there are two kinds of coherence-enhancing inferential relations:

(1) deductive + explanatorily virtuous (2) inductive + explanatorily virtuous

For (at least some) non-explanationist coherentist accounts, there are four kinds of coherence- enhancing inferential relations:

(1) deductive + explanatorily virtuous (2) deductive + non-explanatory (3) inductive + explanatorily virtuous (4) inductive + non-explanatory

The key difference between these three sorts of coherentism is not on the schematic features of coherence, but on how best to fill in the schematic features in question.

Lycan’s brand of coherentism is probably closer to EMV on this issue (i.e.,

explanationism) than is any other extant brand of coherentism. It, like EMV, is explanationist.

And it, like EMV, is moderate explanationist, allowing for good but not explanatorily virtuous

deductive inferential relations.

There is a significant difference, though, between the way in which EMV is moderate explanationist and the way in which Lycan’s brand of coherentism is moderate explanationist.

EMV sanctions two basic inductive inferential-relation forms: inference to the best explainer, and

inference to an explainee. Lycan’s brand of coherentism sanctions just one: inference to the best

explainer. It says that an inductive inferential relation is good only if it is an instance of either

inference to the best explainer, or an inferential-relation form built up out of inference to the best

explainer and valid deductive inferential-relation forms.

You might be a bit surprised that I said that Lycan is probably closer to EMV on this issue (i.e., explanationism) than is any other extant brand of coherentism. “What about Gilbert

Harman and Wilfrid Sellars?”, you might be wondering. Would not their views be as close, if not

closer, to EMV on this issue as is Lycan’s view? Would not their views be paradigm examples of explanationist coherentism? I argue to the contrary elsewhere. Specifically, I argue that neither

Harman nor Sellars is an explanationist coherentist—Harman because he is not a coherentist, and

Sellars because he is not an explanationist.

1.4.2 Meta-Perspectivalism

It will prove best to wait until Chapter 6 to more fully articulate EMV’s meta- perspectivalist component. The rough idea, as I said at the start of this section, is that (in order for S to have good reason for thinking that p obtains) it needs to be the case that from S’s perspective it is likely that he is connected to p, and that, in order for this to be the case, it needs to be the case that S has a view as to how he is connected to the outside world, and according to which the mechanisms involved are reliable.

Like with the moderate explanationist component, the meta-perspectivalist component figures in EMV’s account of what it would be for a belief system to be coherent. The view, to be defended at length in Chapter 6, is that without a view as to how he is reliably connected to the outside world, S’s beliefs would not hang together in the right way. Imagine, for example, that S believes that there is a blue Great Dane before him. Imagine also that S does not have a view as to how he is reliably connected to the outside world. Then, according to EMV, it would not be the case that relative to the rest of his belief system it is likely that there is a blue Great Dane before him—so that, in that sense, it would not be the case that his beliefs hang together, or support each other, in the right way.

I have adopted the term “meta-perspectivalism” from Ernest Sosa’s discussion of coherentism in “The Coherence of Virtue and the Virtue of Coherence”, in which Sosa coins the term “perspectival coherentism”. Sosa raises an objection to coherentism (i.e., a version of what I

call the “Alternative-Systems Objection”), and then argues that the coherentist needs perspectivalism, which he describes below, to answer it:

To avoid refutation by the thought experiment, coherentism is now requiring that the subject of the experiment must be self-aware enough to both grasp and sort his beliefs, and to catalogue his belief sorts by degree of reliability. It is only this requirement that yields the incoherence of the set resulting from the tranformation in our thought experiment. (Sosa 1991, 205-6)

I prefer “meta-perspectivalism” to “perspectivalism” because the requirement, at least generally speaking, is that S have a perspective, or view, of kind K on his perspective, or view, as to what the world is like. That is, the requirement is that S have a meta-perspective of kind K—where K is specified differently on different versions of meta-perspectivalism—on his first-level perspective on the world.

The chief division in the meta-perspectivalist camp is between accounts, such as the one

Sosa has in mind, that require meta-beliefs of the form “I believe that p.” and accounts, such as the one in EMV, that do not. The neat thing about the meta-perspectivalist component in my account is that, unlike the meta-perspectivalist components in my rivals’ accounts, it does not start a meta-level regress, and does not run counter to the coherentist tenet that a belief is justified only if it is supported by a reason. The reason why, I shall argue, is that, unlike the meta- perspectivalist components in my rivals’ accounts, it does not require meta-beliefs (though it does require beliefs about belief-forming processes).

1.4.3 Veridicalism

I said at the start of this section that EMV has a veridicality component, which requires that S’s reasons be true, and that S’s reasons for his reasons be true, and so on. I also said that this component, when coupled with the meta-perspectivalist component, ensures (among other things) that S is right about the ways in which he is reliably connected to the world, and that this, in turn, makes EMV a kind of externalism. This might be a bit puzzling to you, both because of

the multifarious uses of the term “externalism” in the literature, and because of the long-standing tradition of putting the coherentist in the internalist camp. But, as I shall now explain, there is a straightforward sense of externalism on which there is nothing, qua coherentist, precluding a

coherentist from being an externalist, and on which, in addition, EMV (in virtue of its veridicality

component) is externalist.

The internalism/externalism distinction I have in mind is about whether the things on which the justification facts supervene are either mental or supervenient on the mental. The internalist says that the justification facts supervene, in whole, on things that are either mental or supervenient on the mental—so that Smith and Jones, for example, are identical justificationally if they are identical mentally. In contrast, the externalist says that the justification facts supervene, in whole or in part, on things that are neither mental nor supervenient on the mental.

Let this distinction mark the mind-internalism/mind-externalism distinction.

I am following Earl Conee and Richard Feldman in drawing the distinction in this manner:

Somewhat more precisely, internalism as we characterize it is committed to the following two theses. The first asserts the strong supervenience of epistemic justification on the mental: S The justificatory status of a person’s doxastic attitudes strongly supervenes on the person’s occurrent and dispositional mental states, events, and conditions. The second thesis spells out a principal implication of S: M If any two possible individuals are exactly alike mentally, then they are alike justificationally, e.g., the same beliefs are justified for them to the same extent.11 (M) implies that mental duplicates in different possible worlds have the same attitudes justified for them. This cross world comparison follows from the strong supervenience condition in (S).12 Externalists characteristically hold that differences in justification can result from contingent non-mental differences, such as differing causal connections or reliability. Theories that appeal to such factors clearly deny (S) and (M). (Feldman and Conee 2001, 234)

But whereas I call the internalist “the mind-internalist” and the externalist “the mind-externalist”,

Conee and Feldman call the internalist “the mentalist” and the externalist “the non-mentalist”.

The simple coherentist, according to whom a belief is justified just in case it plays an inferential role in a coherent belief system, makes for a nice example on the internalist side. For if Smith and Jones are identical mentally, then they are identical in terms of beliefs and thus in terms of coherence.

The evidentialist, in the manner of Conee and Feldman, too is a mind-internalist.

Whether a belief is justified hinges (solely) on whether it fits the subject’s evidence, which consists of his beliefs and experiences.15 Hence if Smith and Jones are identical mentally, then they are identical in terms of beliefs and experiences, and thus in terms of fittingness.16

For an instance on the externalist side, consider David Armstrong’s account (construed as

an account of justification instead of as an account of knowledge).17 Roughly put, the view is that S’s belief that p is justified just in case S has a property H such that it is a law of nature that if a subject x has H, then x’s believing that p guarantees that p. Neither the property nor the law (as construed by Armstrong) are either mental or supervenient on the mental, and so Armstrong’s account is mind-externalist.

But what about past mental facts? Need a mind-internalist say that the justification facts at t are fixed by the mental facts at t? Or can he allow for mental facts prior to t? On the flip side, is a theorist a mind-externalist just in virtue of making the justification facts at t depend, in part, on certain mental facts prior to t? Or need he require a fact that is neither mental nor supervenient on the mental?18 I shall opt for the latter, in each case—so that a mind-internalist

can allow for mental facts prior to t, and so that a mind-externalist needs to require a fact that is

neither mental nor supervenient on the mental.19

The evil demon scenario, notice, is a perfect test for whether a theory of justification is mind-internalist or mind-externalist. Stewart Cohen, for example, uses it to expose reliabilism as mind-externalist (my words, not his) and thus as false:

this a tenable result? I maintain not. . . . Now part of what the hypothesis entails is that our experience is just as it would be if our cognitive processes were reliable. Thus, on the demon hypothesis, we should have every reason for holding our beliefs, that we have in the actual world. Moreover since we actually have reason to believe that our cognitive processes are reliable, it follows that in the demon world we would have every reason to believe that our cognitive processes were in fact reliable. We might even imagine that a brilliant philosopher had seemingly demonstrated (a la Descartes of the later meditations) the falsity of the demon hypothesis, to the extent that anyone who could follow the reasoning was (intuitively) justified in accepting the conclusion. It strikes me as clearly false to deny that under these circumstances our beliefs could be justified. (Cohen 1984, 281-2)

The victim and his counterpart in the actual world are identical mentally—in beliefs, in experiences, and so on. But according to reliabilism, they are not identical justificationally; for the victim’s processes, unlike his counterpart’s, are unreliable, and so the victim’s beliefs, unlike his counterpart’s, are unjustified. The reliabilist, therefore, fails the test, and so is a mind- externalist.20

The claim, then, in saying that EMV is externalist is that EMV is mind-externalist. It

implies that the justification facts supervene, at least in part, on things that are neither mental nor

supervenient on the mental, so that Smith and Jones, for example, could be identical mentally but

different—even drastically different—justificationally. Imagine, to illustrate, that Smith and

Jones are identical mentally—viz., in terms of beliefs, experiences, desires, fears, etc. Imagine

also that whereas Smith is right about the ways in which he is reliably connected to the world,

Jones, who is a brain in a vat, is wrong about the ways in which he is reliably connected to the

world. Jones is wrong, for instance, in thinking that he is visually connected to the world. Then

(assuming the satisfaction of the other conditions for justification) Smith’s beliefs, but not Jones’s

beliefs, would be justified. Smith, but not Jones, would have good reason for his beliefs.

The reason why the coupling of coherentism and externalism is possible is that the foundationalism/coherentism debate and the mind-internalism/mind-externalism debate are about two very different things. The foundationalism/coherentism debate is about whether justification is skyscraper-like in structure. The coherentist, in his part, says that the structure of justification is holistic, and that, structurally, the parts of a justified (or justification-conferring) belief system

are related not as the parts of a skyscraper are related, but as the parts of, say, a floating piece of paper are related. No part is prior to any other part. Each part contributes to the sustenance of the whole. The mind-internalism/mind-externalism debate, in contrast, is about another thing entirely—i.e., whether justification is wholly mental in grounding. Moser and Goldman, for instance, agree on the foundationalism/coherentism question (siding with foundationalism), but disagree on the mind-internalism/mind-externalism question—with Moser siding with mind- internalism and Goldman siding with mind-externalism. That is, they agree on whether there is non-inferential justification and whether inferential justification is dependent on non-inferential justification, but disagree—drastically—on whether the source of justification is mental (or supervenient on the mental).

Let me stress, as I did in section 2, that I would be fine with not using the term

“coherentism” to describe views such as EMV—using instead, say, the term “non- foundationalism on which justification requires coherence”. The point, then, would be that there there is nothing, qua non-foundationalist who thinks that justification requires coherence, precluding a non-foundationalist who thinks that justification requires coherence from being a mind-externalist, and that EMV (in virtue of its veridicality component) is a non-foundationalist account which both says that justification requires coherence, and says that justification is grounded partially in things that are neither mental nor supervenient on the mental.

I should note that the long-standing tradition of putting the coherentist in the internalist camp is quite understandable, and perhaps even excusable. Though the various extant coherentist accounts deviate in various ways from one another and/or from the view that justification is solely a matter of coherence, each such account entails that sameness in past and present mental facts is sufficient for sameness in justification—so that each such account is mind-internalist.

Take BonJour’s account, for example. Whether a belief is justified at t hinges on (a) whether it plays an inferential role at t, (b) whether the subject’s belief system is coherent at t, (c) whether the subject’s belief system satisfies the observation requirement at t (and so, at t, has both (1) lots

of non-inferential beliefs and (2) reliablility beliefs saying that the belief-forming processes involved are reliable), (d) whether the doxastic presumption (which says that at least most of the subject’s meta-beliefs of the form “I hold the belief that p.” are true) is true at t, (e) whether the subject’s belief system has been coherent, in accord with the observation requirement, and relatively stable over a relatively long stretch of time, and (f) whether the subject has access to, or a reflective grasp of, the facts referred to in (a)-(e). Each such fact—i.e., each fact referred to in

(a)-(f)—is determined, in the end, by past and/or present doxastic facts, so that if Smith and Jones are identical in past and present mental facts, then the facts referred to in (a)-(f) hold of Smith just in case they hold of Jones, and so that, thus, Smith’s beliefs are justified just in case Jones’s beliefs are justified. Hence, the account is mind-internalist.

But what about Lehrer’s account of ultra justification, which restricts the beliefs to which

S can appeal in answering and neutralizing objections to S’s true beliefs? It would seem that, on this account, the demon-victim’s beliefs, in contrast to his counterpart’s beliefs, would not be justified—since the demon-victim’s beliefs would be false, and thus could not be used in answering and neutralizing objections. It would seem, then, that Lehrer’s overall account is mind-externalist. I am reluctant to classify Lehrer’s account as mind-externalist, though, because his account of ultra justification is really an account of knowledge, not justification. He defines undefeated (or irrefutable) justification as ultra justification, and then argues that knowledge reduces to undefeated justification:

D9. S is justified in accepting p in a way that is undefeated at t (S is irrefutably justified in accepting that p at t) if and only if S is justified in accepting p at t on the ultrasystem of S at t. (Lehrer 2000, 171)

Needless to say, the attempt to analyze justification and undefeated justification in terms of acceptance, reasonableness, and truth has yielded a complicated analysis. As is often the case, however, thorough analysis enables us to find the underlying simplicity. We are now in a position to provide an elegant reduction of the original analysis of knowledge (DK). Knowledge reduces to undefeated justification, a just reward for our arduous analytical efforts. (Ibid., 171)

Lehrer is not a mind-externalist on justification. Rather, Lehrer is a mind-internalist on justification, and, like everyone else, a mind-externalist on knowledge.

I should also note that the internalist-coherentism/externalist-coherentism distinction discussed in this section differs significantly from the internalist-coherentism/externalist- coherentism distinction at play in the passage below:

It would be possible, of course, to adopt an externalist version of coherentism. Such a view would hold that the person whose belief is justified need himself have no cognitive access to the fact of coherence, that his belief is justified if it in fact coheres with his system of beliefs, whether or not such coherence is cognitively accessible to him (or, presumably, to anyone). But such a view is unacceptable for essentially the same reasons which were offered against foundationalist versions of externalism and, as discussed earlier, seems to run counter to the whole rationale for coherence theories. (BonJour 1985, 101-2)

BonJour is here invoking not the mind-internalism/mind-externalism distinction, but the quite familiar access-internalism/access-externalism distinction. The issue here is not about whether the things on which the justification facts supervene are mental (or supervenient on the mental), but about whether, to put it roughly (and a bit paradoxically), the subject needs to have access to the things on which the justification facts supervene.21

Much would need to be said to make the access-internalism/access-externalism distinction, and therefore the access-internalist-coherentism/access-externalist-coherentism distinction, clear. For instance, what is it to have access to something? Is it, as BonJour seems to think, to have a belief about it, or, even more demanding, to have a justified belief about it? Or is it something weaker? And if a belief (whether actual or not, and whether justified or not), is it a belief of a certain kind (e.g., non-inferential)?

Luckily, this is not territory into which I need to venture, given that EMV has no access

requirement. It is enough for justification that the moderate explanationist requirement, the meta-

perspectivalist requirement, and the veridicality requirement be satisfied. It is not further

required that the subject have any special access to any of this.

1.5 Conclusion

The way is now clear to start EMV’s defense. I shall start with the standard, or stock, objections to coherentism, and, in rebutting the third such objection (i.e., the Experience

Objection), shall refute experientialist foundationalist accounts such as Moser’s. Then, in chapters 5, 6 and 7, I shall refine and argue for EMV’s three distinguishing components, thus at the same time arguing against EMV’s chief rivals in the coherentist camp.

1 Goldman makes a slightly different move, relativizing the relevant reliability facts to normal worlds— viz., worlds consistent with our general beliefs about the actual world. See Goldman 1986, 107-9.

2 For discussion of what the debate is about, see Cornman 1977 and Haack 1993.

3 The reasoning runs roughly as follows. Suppose, first, that the regress stops with an unjustified belief bn. Then bn would have no justification to give to bn-1, which then would have no justification to give to bn-2, . . ., which then would have no justification to give to b2, which then would have no justification to give to b1. b1, thus, would not be justified. Suppose, second, that the regress circles back to b1. Then b1, to be justified, would have to give justification to itself, in that it would give justification to bn, which then would give justification to bn-1, which then would give justification to bn-2, . . ., which then would give justification to b2, which then would give justification to b1. But this is impossible, and so b1 (were the regress to circle back to b1) would not be justified. And suppose, third, that the regress continues on without end. Then b1 would be justified if b2 were justified, b2 would be justified if b3 were justified, b3 would be justified if b4 were justified, . . .. With the regress continuing on without end, the conditionals too would continue on without end. Hence, b1 would not be justified.

4 This is a bit over-simplified in that the foundationalist could allow for a non-inferentially justified belief for which the subject has doxastic reasons, as long as such reasons were inessential to the justification.

5 The foundationalist, of course, also allows for scenarios in which a belief is inferentially justified by two (or more) beliefs together, as well as for scenarios in which a belief inferentially justifies two (or more) beliefs.

6 Some theorists distinguish between linear coherentism and holistic coherentism, thus making room for the coupling of coherentism and the part-to-part conception of inferential justification. I ignore this possibility—construing the coherentist as pushing the whole-to-part conception—since linear coherentism, which is more aptly dubbed “part-to-part coherentism”, is patently false. On part-to-part coherentism, b1, to be justified, would have to give justification to itself—in that, since inferential justification (on this view) is circular, b1 would have to give justification to bn, which would have to give justification to bn-1, which would have to give justification to bn-2, . . ., which would have to give justification to b2, which would have to give justification to b1. But this is impossible, as a belief cannot be antecedent in justification to itself (or dependent for justification on itself). See Pollock and Cruz 1999 for more on the distinction. See Steup 1996 for more on the objection.

7 That the question of what grounds inferential justification differs from the question of how inferential justification works follows from the fact that the foundationalist and the infinitist agree on the latter but disagree on the former. They agree on how inferential justification works, agreeing that it is part-to-part and uni-directional. They disagree, however, on whether there is non-inferential justification and thus on what grounds inferential justification.

8 The point could be put slightly differently, so that the foundationalist is saying not that inferential justification is grounded in non-inferential justification, but that inferential justification is grounded (at the end of the day) in, for instance, experience (which, in turn, explains non-inferential justification). But either way, the position is that inferential justification is dependent on non-inferential justification.

9 That the coherentist says that inferential justification is grounded (at least in part) in coherence is what makes the coherentist a coherentist, and not some other kind of whole-to-part-ist. The consistentist, who thinks that what makes a belief system justified is consistency, would be an example of a non-coherentist proponent of the whole-to-part conception.

10 Lehrer says that the idea of answering and neutralizing objections can be thought of in terms of a justification game in which S uses his other beliefs in defending his belief that p against objections—so that objections are answered or neutralized not by the other beliefs in S’s belief system, but by S himself. Thus I shall sometimes speak as though, according to Lehrer, it is S, not the other beliefs in S’s belief system, that does the answering and neutralizing.

11 The qualifier “provisional” is meant to highlight the fact that the claim that the street is wet gives us good reason for thinking that the roof is wet only if we have good reason for thinking that the street is wet. Whether it gives us good reason is provisional, hinging on whether we have good reason for thinking that it itself is true.

12 Lycan distinguishes between four kinds, or grades, of explanationism: weak, sturdy, ferocious, and holocaust. See Lycan 2002.

13 D could be more probabilistically consistent than E even if the number of probabilistically inconsistent pairs of beliefs in D were the same as the number of probabilistically inconsistent pairs of beliefs in E. If in E were more at odds with each other than were in D and if other things were equal, then D would be more probabilistically consistent than E.

14 BonJour gives a nice comparison of Brand Blanshard, A. C. Ewing, and C. I. Lewis on this issue. Blanshard (according to BonJour) says that D is fully coherent just in case each of the believed propositions entails and is entailed by the other believed propositions. Ewing drops the first part, saying that D is fully coherent just in case each of the believed propositions is entailed by the other believed propositions. Lewis switches to congruence, and replaces “is entailed by” with “is probabilified to some degree”—so that the view, then, is that D is congruent just in case each of the believed propositions is probabilified to some degree by the other believed propositions (where p is probabilified by q just in case P(p/q) is greater than P(p)). See Blanshard 1939, Ewing 1934, and Lewis 1946.

15 See Feldman and Conee 1985, 15-34.

16 The relation of fit, when the relata are a belief and an experience, is notoriously hard to make sense of. Conee and Feldman acknowledge this, and then (rather unsatisfyingly) leave matters at that.

17 See Armstrong 1973.

18 Conee’s and Feldman’s construal is (strictly speaking) ambiguous on this issue. For “the person’s occurrent and dispositional mental states, events, and conditions” can be read as either “the person’s occurrent and dispositional mental states, events, and conditions at t” or “the person’s occurrent and dispositional mental states, events, and conditions both at t and prior to t”.

19 The difference between a mind-internalist according to whom the justification facts at t are fixed exclusively by certain mental facts at t and a mind-internalist according to whom the justification facts at t are fixed by certain mental facts at t together with certain mental facts prior to t could then be marked by calling the former a “strong mind-internalist” and the latter a “weak mind-internalist”.

20 The test, so as to allow for mind-internalists according to whom past mental facts are important, should really be whether the demon’s victim, who is identical to his counterpart in the actual world in both past and present mental facts, is identical in justification to his counterpart in the actual world.

21 I say “a bit paradoxically” because it make it sound as if the access-internalist were saying that S needs to have access to the things that are required for justification (but that are not also required for belief). This, however, would start a vicious regress, since, on access-internalism, this access is itself one of the things required for justification, and since, thus, S would then need to have access to this first access, as well as access to this second access, and so on.

CHAPTER 2

THE ALTERNATIVE-SYSTEMS

AND ISOLATION OBJECTIONS

It is widely thought that coherentism is hopeless. First, it is said that there are lots of incompatible yet fully coherent belief systems such that any self-consistent belief whatsoever is a member of at least one such system, and that, for instance, this makes coherentism too permissive. Second, it is said that coherence neither involves nor requires any connection (e.g., causal) to the outside world, and that coherentism, therefore, fails in securing a connection between justification and truth. And third, it is said that coherence neither involves nor requires any fit between the subject’s beliefs and the subject’s experiences (e.g., visual experiences), so that, to its detriment, coherentism allows for scenarios in which the subject’s beliefs are justified but radically in conflict with the subject’s total body of evidence. The aim in the next three chapters is to rebut, once and for all, each such objection—showing, therefore, that, contrary to orthodox epistemological opinion, no such objection is in any way helpful in settling the foundationalism/coherentism debate.

I shall start with the Alternative-Systems Objection and the Isolation Objection, arguing that each version thereof fails in three very different ways. I shall argue that each version of both objections presupposes that the coherentist is a mind-internalist, and that this is a problem because each version of both objections fails against the mind-externalist coherentist. I shall argue that the mind-internalist foundationalist is faced with similar objections—so that were the mind-internalist coherentist to fall, the mind-internalist foundationalist too would fall. And I

shall argue against the objections directly, arguing that each version thereof, when fully fleshed out, has either a false premise or a fallacious sub-argument (or both).

Then, in the next chapter, I shall take up the Database Objection, which is supposed to be a problem not for coherentism in general, but for explanationist coherentism in particular. I shall argue that each of its three versions is really just a special case of either the Alternative-Systems

Objection or the regress argument, and that since both the Alternative-Systems Objection and the regress argument are utter failures, so too is the Database Objection.

And then, in Chapter 5, I shall argue against the Experience Objection—i.e., the one that says that, since coherence neither involves nor requires any fit between the subject’s beliefs and the subject’s experiences, coherentism allows for scenarios in which the subject’s beliefs are justified but radically in conflict with his total body of evidence. I shall argue, in particular, that

(initial appearances notwithstanding) experiences are unfit to serve as reasons—either for or against beliefs. This will serve both to rebut the Experience Objection, and to refute my theory’s main rivals in the foundationalist camp. For the key debate between the foundationalist and the coherentist, at least in terms of what it would be to have good reason, is on the question of whether experiences can serve as reasons.

2.1 The Objections

The aim in this section is to set out the various versions of the two objections in as clear,

true-to-the-literature, and compelling a manner as is possible. This will prove helpful in securing

the ultimate goal of the chapter (i.e., rebutting the Alternative-Systems Objection and the

Isolation Objection), in that it will ensure that the objections I rebut are in the literature, and that

there are not any better varieties in the offing.

2.1.1 The Alternative-Systems Objection (ASO)

With the first variety of the ASO, the arguer uses the claim that there are lots of

incompatible yet fully coherent belief systems such that any self-consistent belief whatsoever is a

member of at least one such system to get to the sub-conclusion that coherentism is too

permissive, or too liberal—letting people believe, with justification, whatever they want to

believe. Richard Feldman, though not a fan of the objection, is perhaps the clearest in

formulating it:

Consider the proposition that Abraham Lincoln was assassinated. If, as the objectors contend, there are many different, and incompatible, coherent systems of beliefs, there will be some systems that include this belief and others that include its negation. If that belief is part of your actual system, you can imagine a system that replaces everything supporting it or following from it with different propositions. By carefully constructing the new system, you could get one just as coherent as your current system, but including the proposition that Lincoln was not assassinated. Thus, if there are all these different coherent systems, then you can make any belief you want justified simply by picking and choosing the rest of your beliefs appropriately. That cannot be right. (Feldman 2003, 67)

The worry, charitably fleshed out, looks like this:

The ASO 1 (1) If coherentism is true, then justification is solely a matter of coherence.1 (2) There are lots of incompatible yet fully coherent possible belief systems such that any self-consistent possible belief whatsoever is a member of at least one such system.2 ------(3) If coherentism is true, then each such system is justified.3 ------(4) If coherentism is true, then S can believe, with justification, whatever he wants to believe—in the sense that if he wants to believe that p and if there is a possible coherent belief system D having the belief that p as a member, then if he were to have D, his belief that p would be justified. ------(5) Coherentism is too permissive. ------(6) Coherentism is false.

That coherentism is too permissive, then, is supposed to follow from the claim that coherentism

looks with favor on each of the various incompatible yet fully coherent possible belief systems,

and that coherentism, thus, looks with favor on each of the various self-consistent possible beliefs

therein.4

The second variety of the ASO, expressed below by Paul Moser, focuses not on whether coherentism is too permissive, but on whether coherentist justification is connected to truth:

But however coherence relations among propositions are understood, Probability Coherentism faces a serious problem. Mere coherence of a system of propositions, however comprehensive, fails to provide evidential probability concerning how things actually are. There are comprehensive coherent systems of obviously false, evidentially gratuitous propositions, such as propositions in science fiction. And for virtually any coherent system of propositions, we can imagine an alternative system consisting mainly of the denials of the propositions in the first system. But of course two such coherent systems cannot both be probability-providing for a person concerning how things actually are. This is especially clear if we construe “probable” as “more probable than not”. For if a proposition, P, is evidentially more probable than its denial, ~P, then ~P is not evidentially more probable than P. (Moser 1989, 62)

The argument, more fully put, runs as follows:

The ASO 2 (1) If coherentism is true, then justification is solely a matter of coherence. (2) There are lots of incompatible yet fully coherent possible belief systems such that any self-consistent possible belief whatsoever is a member of at least one such system. ------(3) If coherentism is true, then each such system is justified. (4) That a proposition p is likely to be true entails that no proposition q incompatible with p is also likely to be true—such that if there is good reason for thinking that p is true, then for any proposition q incompatible with p, it is not the case that there is good reason for thinking that q is true. ------(5) If coherentism is true, then justification is not connected to truth—in that the fact that a system of beliefs is justified would in no way make it likely, or such that there is good reason for thinking, that the beliefs in it are true. (6) Justification is connected to truth. ------(7) Coherentism is false.

The worry, in short, is that since coherence could in no way discriminate between the various incompatible yet fully coherent possible belief systems (because each such system is fully coherent) and since coherence could in no way make all such systems likely (because such systems are competitors and, thus, because a high likelihood for one would preclude a high likelihood for the others), coherence could in no way make any such system likely—so that coherentism, therefore, severs the tie between justification and truth.5

With the third and final variety of the ASO, the arguer uses the claim that there are lots of incompatible yet fully coherent possible belief systems to get to the sub-conclusion that

coherentism is inapt as a procedure for deciding what to believe (i.e., as a doxastic decision procedure, or “ddp” for short). C. I. Lewis, for instance, can be read as thinking of the objection along such lines:

There undoubtedly is some logical relation of facts—or more than one—to which the name ‘coherence’ might aptly be given. And there is equally little doubt that such logical and systemic relationships are important for assuring credibility—once a sufficient number of antecedent and relevant facts have been otherwise determined. But no logical relationship, by itself, can ever be sufficient to establish the truth, or the credibility even, of any synthetic judgment. That is one point which logical studies of the last half century have made abundantly clear. Unless the beliefs so related, or some of them, have a warrant which no logical principle can assure, no logical relation of them to one another constitutes a scintilla of evidence that they are even probable. Let us assume that the whole of the truth has even that strongest type of coherence illustrated by a system of geometry. The statements of the system (postulates and theorems together) are so related that, if we should be doubtful of any one of them, the other statements of the system would be sufficient to assure it with deductive certainty. But that relationship, as we know, is insufficient to determine any truth about the geometric properties of actual things. If Euclid is thus coherent, then so too are Riemann and Lobachevsky; though given any denotation of the geometric vocabulary, these three geometries are mutually incompatible systems. If the truth about our space is ever to be ascertained, something disclosed in experience must be the final arbiter. Since this is the case for geometric truths, which cohere by the strong relations of deductive logic, a fortiori it must be the case for empirical truth at large, for the determination of which we must so often rely upon induction, which affords a probability only, on the supposition that our premises are certain or that they have some antecedent probability on other grounds. In brief, we have nothing but experience and logic to determine truth or credibility of any synthetic judgment. Rule out datum-facts afforded by experience, and you have nothing left but the logically certifiable. And logic will not do it. (Lewis 1952, 169)

Fully fleshed out, the objection looks like this:

The ASO 3 (1) There are lots of incompatible yet fully coherent possible belief systems such that any possible belief whatsoever is a member of at least one such system. ------(2) Coherence is inapt as a ddp, at least when deciding on a belief system from among the set of possible belief systems.6 (3) If coherentism is true, then justification is solely a matter of coherence. (4) If an account of justification says that justification is solely a matter of X, then the account is apt as a ddp only if X is apt as a ddp. ------(5) Coherentism is inapt as a ddp, at least when deciding on a belief system from among the set of possible belief systems. (6) An account of justification is correct only if it is apt as a ddp. ------(7) Coherentism is false.

The argument, put simply, is that because coherence could in no way discriminate between the incompatible yet fully coherent possible belief systems (given that each such system is fully coherent), and because, thus, coherence could in no way discriminate between the possible beliefs therein, coherence could in no way favor any such belief over its competitors—and therefore could in no way help in deciding what to believe.7,8

2.1.2 The Isolation Objection (IO)

The first variety of the IO is articulated nicely by Susan Haack:

To get this objection to coherentism in as strong a form as possible, it is desirable (though I shall continue to call it the drunken sailors argument) to spell it out literally. The fundamental objection is this: that because coherentism allows no non-belief input— no role to experience or the world—it cannot be satisfactory; that unless it is acknowledged that the justification of an empirical belief requires such input, it could not be supposed that a belief’s being justified could be an indication of its truth, of its correctly representing how the world is. In the end, I believe, this argument really is fatal to coherentism. A theory couched in terms exclusively of relations among a subject’s beliefs faces an insuperable difficulty about the connection between the concepts of justification and truth. How could the fact that a set of beliefs is coherent, to whatever degree and in however sophisticated a sense of “coherent”, be a guarantee, or even an indication, of truth? (Haack 1993, 26-7)

The argument, it seems, goes like this:

The IO 1 (1) In order for it to be highly likely that the beliefs in a belief system are true, the system needs to be connected to the outside world. (2) A belief system’s being coherent neither involves nor entails any sort of connection between the system and the outside world. (3) If coherentism is true, then justification is solely a matter of coherence. ------(4) Coherentist justification is not connected to truth. (5) There is a connection between justification and truth. ------(6) Coherentism is false.

That coherentism is false is supposed to follow from the claim that coherentist justification is not connected to truth. And this, in turn, is supposed to follow from the claim that a system’s being

coherent neither involves nor entails any sort of connection between the system and the world, together with the claim that likelihood of truth requires a system-world connection of some kind.9

The second, and final, version of the IO focuses on the fact that, because coherence neither involves nor entails any sort of system-world connection, coherentism allows for justification in scenarios in which the subject’s belief system is entirely isolated from the world.

Laurence BonJour puts it thus:

Coherence is purely a matter of the internal relations between the components of the belief system; it depends in no way on any sort of relation between the system of beliefs and anything external to that system. Hence if, as a coherence theory claims, coherence is the sole basis for empirical justification, it follows that a system of empirical beliefs might be adequately justified, indeed might constitute empirical knowledge, in spite of being utterly out of contact with the world that it purports to describe. Nothing about any requirement of coherence dictates that a coherent system of beliefs need receive any sort of input from the world or be in any way causally influenced by the world. But this is surely an absurd result. Such a self-enclosed system of beliefs, entirely immune from any external influence, cannot constitute empirical knowledge of an independent world, because the achievement of even minimal descriptive success in such a situation would have to be either an accident or a miracle, not something which anyone could possibly have any reason to expect—which would mean that the beliefs involved would not be epistemically justified, even if they should somehow happen to be true. (BonJour 1985, 108)

The worry is best construed as follows:

The IO 2 (1) If coherentism is true, then justification is solely a matter of coherence. (2) Coherence is solely a matter of internal relations between beliefs, so that coherence neither involves nor entails any sort of connection to the outside world. ------(3) If coherentism is true, then justification is solely a matter of internal relations between beliefs, so that justification neither involves nor entails any sort of connection to the outside world. ------(4) The following scenario is possible: S’s belief system is coherent, though entirely isolated from the world. (5) If a belief system is entirely isolated from the world, then it would be a coincidence were the beliefs in it true. ------(6) It is not likely that S’s beliefs, in the scenario imagined, are true. (7) If X-ist justification (e.g., reliabilist justification) is connected to truth, then truth is likely whenever X (e.g., reliability) is present.10 ------(8) Coherentist justification is not connected to truth. (9) There is a connection between justification and truth. ------

(10) Coherentism is false.

The challenge here, like with the ASO 2, is that coherentist justification is not connected to truth.

This time, however, the worry is that, since coherence neither involves nor entails any sort of system-world connection, coherentism allows for a possibility (i.e., justification in an isolation scenario) not in keeping with the claim that coherentist justification is connected to truth.

2.2 The First Rebuttal

I argued in the previous chapter that a coherentist can be a mind-externalist, saying both that justification is partially a matter of coherence and that justification is partially a matter of things that, unlike coherence, are neither mental nor supervenient on the mental. EMV, for instance, says that justification is partially a matter of coherence, and partially a matter of (among other things) being reliably connected to the outside world.

This, however, spells trouble for the ASO and the IO, given that each version of each such argument assumes that coherentist justification is solely a matter of coherence. The ASO 1, the ASO 2, and the ASO 3 assume it in moving from the claim that there are lots of incompatible yet fully coherent possible belief systems such that any self-consistent possible belief whatsoever is a member of at least one such system to the claim that coherentism looks with favor on each such system. The IO 1 uses it in the move from the claim that high likelihood of truth, but not coherence, requires a system-world connection of some kind to the claim that coherentism fails in securing a connection between justification and truth. And the IO 2 uses it in a sub-argument in support of the claim that if coherentism is true, then justification is solely a matter of internal relations between beliefs, such that justification neither involves nor entails any sort of connection to the world.

The point, then, is that the foundationalist—to win the debate with the coherentist on the

structure of justification—needs more than just the ASO and the IO. Even if they worked against

the mind-internalist coherentist, there would still be the mind-externalist coherentist to deal with.

Reply 1. But it would seem, you might think, that to construe justification, even in part, in terms of things that are neither mental nor supervenient on the mental would be to make room for non-inferential justification—which, of course, would be to abandon coherentism altogether.

Not so. To make room for non-inferential justification would be to make room for the possibility of a justified belief the justification of which is not dependent on logical support from other beliefs. But no coherentist—not even a mind-externalist coherentist—allows for such a possibility, since no coherentist—not even a mind-externalist coherentist—allows for justification in the absence of logical support from other beliefs. The mind-externalist coherentist disagrees with the mind-internalist coherentist not on whether logical support from other beliefs (from within the confines of a coherent belief system) is necessary for justification, but on whether such support is sufficient for justification.

Reply 2. I argued in the previous chapter that the coherentist, when properly construed, grounds inferential justification at least partially—but perhaps not wholly—in coherence. I pointed to a theory that requires both coherence and reliability, and argued that there is a straightforward sense in which it is coherentist; it is in conflict with each of the three core tenets of foundationalism, and it makes heavy use of coherence. I then said that you could instead insist on the standard line (which says that the coherentist grounds inferential justification wholly in coherence) and say that theories such as the one I pointed to are non-coherentist brands of non- foundationalism; after all, this issue is merely taxonomic. You might think, however, that this would be fatal for my rebuttal (in this section) to the ASO and the IO—because then it would not be the case that the sort of theory I was calling “mind-externalist coherentism” is a kind of coherentism, and because then, thus, it would not be the case that the sort of theory I was calling

“mind-externalist coherentism” is a kind of coherentism that is immune to the ASO and the IO.

The overall point would remain, though. It would still be true that the sort of theory I call

“mind-externalist coherentism” is diametrically opposed to foundationalism on the structure of justification, opposing the claim that there is non-inferential justification, opposing the claim that

inferential justification is part-to-part and uni-directional, and opposing the claim that inferential justification is grounded in non-inferential justification. It would still be true that the sort of theory I call “mind-externalist coherentism” says that inferential justification is whole-to-part, and that an essential factor in the justification of a belief system is coherence. It would still be true that the sort of theory I call “mind-externalist coherentism” is immune to the ASO and the

IO. And it would still be true, therefore, that the foundationalist needs more than just the ASO and the IO to show that justification is skyscraper-like in structure, or even to show that it is false that all justification is inferential, or that it is false that inferential justification is whole-to-part, or that it is false that inferential justification is at least partially grounded in coherence.

Reply 3. You might object that I am being a bit unfair, in that I am invoking a kind of coherentism that neither the proponents of the ASO and the IO nor the coherentists that they are attacking have ever even considered. The standard coherentist, you might press, is a mind- internalist according to whom justification is solely a matter of coherence, and so, since the objections are aimed at the standard coherentist and since they succeed against him, they achieve their aim.

I agree, to a certain extent. I agree with the claim that the standard coherentist is a mind- internalist according to whom justification is solely a matter of coherence. I also agree with the claim that the objections are made with the standard coherentist in mind. However, I do not agree with, and shall later argue against, the claim that the objections succeed against the standard coherentist. The important point for now, however, is just that if the foundationalist takes this line, he cannot argue for foundationalism along the following lines:

(1) Either foundationalism is true, or coherentism is true. (2) Coherentism is false, as is shown by the ASO and the IO. ------(3) Foundationalism is true.

The second premise is false, given that there are varieties of coherentism on which justification is more than just a matter of coherence, and given that, thus, each version of the ASO and each version of the IO is unsound.

This is perfectly parallel, notice, to the fact that the coherentist could not establish the truth of coherentism by establishing the falsity of mind-internalist foundationalism. He would still need to address mind-externalist varieties of foundationalism such as Alvin Goldman’s process reliabilism and Robert Nozick’s tracking account (construed as a theory of justification).

The foundationalist could modify the second premise with the hope of making it true, so

that the argument would run like this:

(1) Either foundationalism is true, or coherentism is true. (2) Mind-internalist coherentism is false, as is shown by the ASO and the IO. ------(3) Foundationalism is true.

But then the argument would be invalid, since for all the premises say, it might be the case that mind-externalist coherentism is true. What would be needed is a further argument against either mind-externalism in general, or mind-externalist coherentism in particular.

Reply 4. “But what would be the point in being a coherentist, were mind-externalism true?”, you might ask. Why not instead be a mind-externalist foundationalist, such as a ground reliabilist in the manner of William Alston or Marshall Swain?

Imagine that, like a mind-internalist coherentist, you thought that experiences are unfit to serve as reasons, and that, thus, coherentism is needed to make sense of what it would be to have good reason. Then, of course, mind-externalist coherentism—as opposed to mind-externalist foundationalism—would make perfect sense for you. The foundationalism/coherentism question, whether within the mind-internalist camp or within the mind-externalist camp, is about the structure of justification, and so the answer to the question “Why be a mind-externalist coherentist, not a mind-externalist foundationalist?” should have the same answer as does the question “Why be a mind-internalist coherentist, not a mind-internalist foundationalist?”.

Reply 5. “But the mere possibility of mind-externalist coherentism is nowhere near

sufficient to undermine the dialectical force of the ASO and the IO.”, you might argue. After all,

it might be that mind-externalist coherentism, though possible, is highly implausible, and that the

ASO and the IO, therefore, are decisive against every variety of coherentism with even a shred of

plausibility.

This reply is problematic in at least two respects. The first is that, as will be argued in

Chapter 7, mind-externalist coherentism is far superior to mind-internalist coherentism. The second, and more obvious, respect in which it is problematic is that, dialectically speaking, the foundationalist is in no position to treat the mind-externalist coherentist any less seriously than he treats the mind-internalist coherentist. The only distinguishing feature of the mind-externalist coherentist, in comparison to the mind-internalist coherentist, is his espousal of mind-externalism.

But the foundationalist, even the mind-internalist foundationalist, treats mind-externalism quite seriously—the mind-externalist foundationalist in endorsing it, and the mind-internalist foundationalist in arguing against it. Of course, the foundationalist could do away with mind- externalist coherentism by doing away with mind-externalism in general. But the point would remain: the ASO and the IO, by themselves, are not enough to secure foundationalism; supplementation is needed.

2.3 The Second Rebuttal

The task now is to show that, to undermine the ASO and the IO, the coherentist need not appeal to mind-externalism, or even to the possibility of mind-externalist coherentism. The coherentist can simply point out that mind-internalist foundationalism is faced with similar objections—so that, at best, alternative-systems-type objections and isolation-type objections are decisive against mind-internalism.

I shall use evidentialism, as put forward by Earl Conee and Richard Feldman, as representative of mind-internalist foundationalism; it is relatively easy to work with. The view, to

repeat (from Chapter 1), is that a belief is justified just in case it fits the subject’s evidence, where the subject’s evidence consists of his beliefs and experiences.

On evidentialism, a cognitive system is perfect—vis-à-vis justification—just in case the

non-inferential beliefs fit the experiences (and are not contravened by other beliefs), the first-level

inferential beliefs fit the non-inferential beliefs (and are not contravened by other beliefs), the

second-level inferential beliefs fit the first-level inferential beliefs (and are not contravened by

other beliefs), and so on. Call this property “E”.

The first thing to see is that there are lots of incompatible yet E possible cognitive systems such that any self-consistent possible belief whatsoever is a member of at least one such system. Think of the various incompatible yet fully coherent possible belief systems referred to in the ASO, and then add to the encompassing cognitive systems the appropriate experiences.11

For example, for the belief systems with the non-inferential belief “There is a black cat before

me.”, add (to the encompassing cognitive systems) a visual experience of a black cat. But for the

belief systems with the non-inferential belief “There is a white cat before me.”, add a visual

experience of a white cat.

With this, it is quite straightforward to generate objections analogous to the three versions of the ASO. The analogue to the ASO 1, for starters, would run as follows:

The Alternative E Systems Objection 1 (AESO 1) (1) If evidentialism is true, then justification is solely a matter of E. (2) There are lots of incompatible yet E possible cognitive systems such that any self- consistent possible belief whatsoever is a member of at least one such system. ------(3) If evidentialism is true, then each such system is justified. ------(4) If evidentialism is true, then S can believe, with justification, whatever he wants to believe—in the sense that if he wants to believe that p and if there is a possible E cognitive system D having the belief that p as a member, then if he were to have D, his belief that p would be justified. ------(5) Evidentialism is too permissive. ------(6) Evidentialism is false.

Three comments. First, read the claim “justification is solely a matter of E” as the claim that to be justified is to be in a system with E. Second, the claim “If evidentialism is true, then justification is solely a matter of E.”, though strictly speaking false, is harmless. Yes, the evidentialist allows for the possibility of a justified belief in a system with unjustified beliefs.

And so yes, the evidentialist allows for the possibility of a justified belief in a system not having

E. But this is of no consequence, as the objections could be reconstrued in terms of E’, where a system is E’ with respect to p just in case (a) it has as a member the belief that p and (b) the belief that p fits the subject’s evidence (which, all told, includes the subject’s experiences). And third, read the claim “If evidentialism is true, then each such system is justified.” as the claim that if evidentialism is true, then each belief in each such system is justified.

You might think that the evidentialist, though like the coherentist in being faced with

alternative-systems-type objections, is in a much better position to meet them. After all, S does

not pick or choose his experiences, and thus, because his system is E only if his beliefs fit his

experiences, he is significantly constrained in shaping his doxastic system. But notice, the

coherentist can say similar things, given that S can no more pick his non-inferential beliefs (i.e.,

his beliefs not resulting from reasoning) than pick his experiences.12

Regardless, the point about the spontaneity of experiences is irrelevant. The claim “If evidentialism is true, then S can believe, with justification, whatever he wants to believe.” is to be read as the claim that if evidentialism is true, then if S wants to believe that p and if there is a possible E cognitive system D having the belief that p as a member, then if he were to have D, his belief that p would be justified. The truth of this conditional is in no way threatened by the fact that the man-in-the-street, because of his psychological make-up, cannot pick his experiences.

Now for the analogue to the ASO 2:

The Alternative E Systems Objection 2 (AESO 2) (1) If evidentialism is true, then justification is solely a matter of E. (2) There are lots of incompatible yet E possible cognitive systems such that any self- consistent possible belief whatsoever is a member of at least one such system. ------

(3) If evidentialism is true, then each such system is justified. (4) That a proposition p is likely to be true entails that no proposition q incompatible with p is also likely to be true. ------(5) If evidentialism is true, then justification is not connected to truth—so that the fact that a cognitive system is justified in no way makes it likely, or such that there is good reason for thinking, that the beliefs in it are true. (6) Justification is connected to truth. ------(7) Evidentialism is false.

The evidentialist, it seems to me, should attack the move from (3) and (4) to (5), just as the mind- internalist coherentist should attack the corresponding move from (3) and (4) to (5) in the ASO 2.

The issue, at any rate, is the same.

The exception is the analogue to the ASO 3, in that the evidentialist can respond in a way that the mind-internalist coherentist cannot. Consider:

The Alternative E Systems Objection 3 (AESO 3) (1) There are lots of incompatible yet E possible cognitive systems such that any possible belief whatsoever is a member of at least one such system. ------(2) E is inapt as a ddp, at least when deciding on a belief system from among the set of possible belief systems. (3) If evidentialism is true, then justification is solely a matter of E. (4) If an account of justification says that justification is solely a matter of X, then the account is apt as a ddp only if X is apt as a ddp. ------(5) Evidentialism is inapt as a ddp, at least when deciding on a belief system from among the set of possible belief systems. (6) An account of justification is correct only if it is apt as a ddp. ------(7) Evidentialism is false.

Yes, evidentialism is neutral with respect to the alternative E cognitive systems. And yes, it follows that evidentialism is inapt as a cognitive decision procedure (at least at the global level).

But no, it does not follow that it is inapt as a doxastic decision procedure—even at the global

level. The evidentialist could construe his doxastic decision procedure so that it takes

experiences as inputs. The procedure would then discriminate between the alternative E

cognitive systems, ruling against both (a) the systems with experiences different from the

subject’s experiences and (b) the systems with beliefs not fitting the experiences therein.

Problem solved.

This is good for the evidentialist, in a way; he is not burdened with yet a third alternative- systems objection. But since the ASO 3 is a failure against the mind-internalist coherentist (as I shall argue below), that the foundationalist is thus unburdened in no way tilts the balance—at the end of the day—in his favor.

The second thing to see (with the first being that there are lots of incompatible yet E possible cognitive systems such that any self-consistent possible belief whatsoever is a member of at least one such system) is that E neither involves nor entails any sort of connection (e.g., causal) to the outside world. Think of evil-demon scenarios, where S’s beliefs and experiences, though entirely illusory, are identical in content to his actual beliefs and experiences. S’s cognitive system is E in both scenarios, but in the evil-demon scenario it is entirely isolated from the world (in the relevant sense).13

But because of this, the evidentialist, just like the mind-internalist coherentist, is faced with isolation objections. Here, first, is the analogue to the IO 1:

The E Isolation Objection 1 (EIO 1) (1) In order for it to be highly likely that the beliefs in a cognitive system are true, the system needs to be connected to the world. (2) A cognitive system’s being E neither involves nor entails any sort of connection between the system and the outside world. (3) If evidentialism is true, then justification is solely a matter of E. ------(4) Evidentialist justification is not connected to truth. (5) There is a connection between justification and truth. ------(6) Evidentialism is false.

And here, second, is the analogue to the IO 2:

The E Isolation Objection 2 (EIO 2) (1) If evidentialism is true, then justification is solely a matter of E. (2) E is solely a matter of internal relations between (a) beliefs and experiences and (b) beliefs and beliefs, so that a system’s being E neither involves nor entails any sort of connection between the system and the outside world. ------

(3) If evidentialism is true, then justification is solely a matter of internal relations between (a) beliefs and experiences and (b) beliefs and beliefs, so that a system’s being justified neither involves nor entails any sort of connection between the system and the outside world. ------(4) The following scenario is possible: S’s cognitive system is E though entirely isolated from the outside world. (5) If a cognitive system is entirely isolated from the outside world, then it would be a coincidence were the beliefs in it true. ------(6) It is not likely that S’s beliefs, in the imagined scenario, are true. (7) If X-ist justification is connected to truth, then truth is likely whenever X is present. ------(8) Evidentialist justification is not connected to truth. (9) There is a connection between justification and truth. ------(10) Evidentialism is false.

Neither argument is without flaws. (7) in the second argument, for instance, is false (as I shall argue below). But either way, the evidentialist, with respect to isolation worries, is in the same boat as is the mind-internalist coherentist. Each view allows for the possibility of isolation, and so if the possibility of isolation entails non-truth-conduciveness, then neither mind-internalist coherentism nor evidentialism is truth-conducive.

The lesson is fully general. For any theory of justification—whether foundationalist or coherentist—according to which justification is supervenient, in full, on the mental, there are alternative-systems-type objections and isolation-type objections in the waiting. Why? The first part of the reason is that the key difference between mind-internalist foundationalism and mind- internalist coherentism is that the former, but not the latter, makes appeal to experience. The foundationalist says that the space of reasons (to borrow a phrase from Wilfrid Sellars) includes experiences. The coherentist, in contrast, says that the space of reasons is restricted to the space of beliefs. The second part of the reason is that no mind-internalist relation between belief and experience is fit to either block the existence of an innumerable number of incompatible but possible cognitive systems in which the relation in question is present, or guarantee a connection to the outside world.

2.4 The Third Rebuttal

The task now is to move beyond the dialectical point from the previous section, by

addressing the objections directly. I shall proceed in order, starting with the ASO 1.

2.4.1 Against the ASO 1

It is clear that mind-internalist coherentism is permissive, in saying that if S wants to believe that p and if there is a possible coherent belief system D having as a member the belief that p, then if S were to somehow bring it about that he has D, his belief that p would be justified.

It is also clear, as I argued in the previous section, that mind-internalist foundationalism is equally permissive. The question, now, is whether this permissiveness is vicious, as the ASO 1 alleges.

The chief difference between the mind-internalist and the mind-externalist, whether foundationalist or coherentist, centers on the notion of having good reason. The mind-internalist says that whether S’s belief that p is justified is determined by whether S has good reason for thinking that p, and that this, in turn, is determined by the contents of both his other beliefs and his experiences. The mind-externalist, in contrast, says either that what determines whether S’s belief that p is justified is something other than just whether S has good reason for thinking that p, or that what determines whether S has good reason for thinking that p is something other than just the contents of both his other beliefs and his experiences. The process reliabilist, for example, falls in the first mind-externalist camp, averring that what determines whether S’s belief that p is justified is not whether S has good reason for thinking that p, but whether it was produced by a reliable process. The ground reliabilist, in contrast, falls in the second mind-externalist camp— agreeing with the claim that whether S’s belief that p is justified is determined by whether S has good reason for thinking that p, but disagreeing with the claim that this, in turn, is determined just by the contents of both his other beliefs and his experiences. S’s environment, insists the ground reliabilist, matters too. The reason, then, why the mind-internalist, but not the mind-externalist,

thinks that justification supervenes on the mental is that the mind-internalist, but not the mind- externalist, thinks that justification consists in having good reason, and that this is solely determined by the contents of the subject’s mental states.

Consider, to illustrate, the mind-internalist’s use of evil-demon cases against mind- externalists such as the process reliabilist. The mind-internalist argues that process reliabilism gives the wrong result for your counterpart’s beliefs in the evil-demon world, saying that, since your counterpart’s belief-forming processes are unreliable (because of the evil demon), your counterpart’s beliefs are unjustified. That this is the wrong result, argues the mind-internalist, follows from the fact that, by hypothesis, you and your counterpart have the same beliefs and experiences, and thus the same evidence—so that, because of this and because you have good reason for your beliefs, your counterpart too has good reason, and just as good, for his beliefs.

The implicit claim, notice, is that justification consists in having good reason, and that this is solely determined by the contents of the subject’s mental states.14

Imagine, then, that—as the mind-internalist coherentist allows—S, who lives in the actual world, finds a belief system D such that if he were to have D, then (because D is coherent) his belief that p, which is one of his favorite beliefs, would be justified. Imagine also that, somehow,

S manages to move from his (then current) belief system to D, and thereby (according to mind- internalist coherentism) comes to justifiably believe that p—though, suppose, p is widely thought to be false, and though, suppose further, almost all of the beliefs in D are false. Is this permissiveness vicious? Not if the rationale for mind-internalism is true, for S, notice, is identical mentally to T, who also has D but who lives in a possible world in which it is widely thought that p and in which almost all of the beliefs in D are true. T, in other words, is just a regular cognizer in his world—viz., a cognizer who forms beliefs in the normal ways, and whose true beliefs far outnumber his false beliefs. There would be no problem with mind-internalist coherentism for saying that T’s beliefs are justified, since T in his world is analogous to us in our world. But then there should be no problem with mind-internalist coherentism for saying that S’s beliefs are

justified—given that, by hypothesis, T has good reason for his beliefs, given that S’s beliefs are identical in content to T’s beliefs, and given that, on mind-internalism, having justification

consists in having good reason, and that this, in turn, consists in internal relations between the

contents of the subject’s mental states.

I shall argue in Chapter 7 that there are two very different, but equally legitimate, senses in which you can have good reason, one of which is mind-internalist and one of which is mind- externalist. This will serve to block the possible reply that though (as I have shown) the severe permissiveness of mind-internalist coherentism flows quite straightforwardly from the rationale for mind-internalism, mind-internalist coherentism is still too permissive—given that mind- internalism itself is too permissive.

2.4.2 Against the ASO 2

The charge in the ASO 2, roughly put, is that since coherence could in no way discriminate between the various incompatible yet fully coherent possible belief systems (because each such system is fully coherent) and since coherence could in no way make all such systems likely (because they are competitors and, thus, because a high likelihood for one would preclude a high likelihood for the others), coherence could in no way make any such system likely. This is supposed to show that there is no connection between coherentist justification and truth.

The problem with this line of reasoning is that, were it sound, it would follow that there is no connection between reliabilist justification and truth. The reliabilist allows for false but justified beliefs—viz., for false but reliably-produced beliefs. This means, though, that the reliabilist also allows for incompatible but justified beliefs—viz., for scenarios in which S, via reliable process R, believes that p, in which T, via reliable process R’ (which might be identical to

R), believes that q, and in which p and q are incompatible. It would follow, then, that reliabilist justification is in no way connected to truth—given that reliability could in no way discriminate

between the various incompatible yet reliably-produced possible beliefs (because each such belief is reliably produced), and given that reliability could in no way make all such beliefs likely

(because such beliefs are competitors and, thus, because a high likelihood for one would preclude a high likelihood for the others).

2.4.3 Against the ASO 3

The problem with coherentism, according to the ASO 3, is not that it is useless, across the board, as a doxastic decision procedure (ddp). It is fully acknowledged that coherentism can help with decisions at the local level, such as when the decision is between adding the belief that p and

adding the belief that q (where p and q are incompatible) and when the former, in comparison to

the latter, would yield a system with a higher degree of coherence. It is even fully acknowledged

that coherentism can help with decisions at the global level—for example, if the decision is

between belief system D and belief system E, and if D is coherent whereas E is not coherent. The

supposed problem, rather, is that coherentism is useless as a ddp at the global level when the

choice is between all possible belief systems.

Fine. But so what? The project of giving a theory of justification differs significantly from the project of giving a recipe for building a belief system from scratch. In giving a theory of justification, the epistemologist is not giving a procedure, or algorithm, for constructing or picking a belief system. Rather, he is giving an answer to the question of what makes a justified belief justified. His aim is explanatory—viz., to shed light on why it is that a justified belief is justified, or, in other words, to locate the things on which justification depends.15

R. Eugene Bales makes a similar point in “Act-Utilitarianism: Account of Right-Making

Characteristics or Decision-Making Procedure?”. He starts by distinguishing between the notion of a theory of moral rightness and the notion of a decision-making procedure for moral rightness, saying that whereas a theory of moral rightness is an answer to the question of what makes a

morally right action morally right, a decision-making procedure for moral rightness is an answer to the question of how it is that you should go about deciding what to do. Then he sets out some popular attacks on act-utilitarianism. And last, he uses the distinction between a theory and a decision-making procedure to rebut the attacks. The overall point is that though the attacks might be decisive in showing that act-utilitarianism is inadequate as a decision-making procedure for moral rightness, they nonetheless leave act-utilitarianism untouched—as act-utilitarianism is a

theory, not a decision-making procedure.16

But there is a second, and deeper, point to be made against the ASO 3. The point, put simply, is that its underlying idea—the idea of using a ddp to build or choose a belief system from scratch—is seemingly unintelligible. No clear sense, it seems, can be made of it.

To bring out this seeming unintelligibility, it will help to look at Alvin Goldman’s paper

“The Internalist Conception of Justification”. Though he never speaks to the question of whether

sense can be made of the idea of using a ddp to build, or choose, a belief system from scratch,

what he says in answer to other questions is perfectly parallel to what, it seems to me, should be

said in answer to the inelligibility question.

Goldman (well into the paper) considers, and then attacks, an internalist view according to which a ddp is right just in case it would be chosen, and properly so, from the internal standpoint, where what is distinctive of the internal standpoint is that it is doxastically neutral— not allowing for any epistemically problematic beliefs (such as beliefs about the external world).

The problem, he argues, is not that choosing a ddp from the internal standpoint would itself amount to adopting a doxastic attitude towards a proposition and thus would itself require already having a ddp; a ddp is a policy (not a proposition), and so choosing a ddp, regardless of the standpoint from which it is made, in no way amounts to adopting a doxastic attitude towards a proposition. Rather, the problem is that nothing could be chosen from the internal standpoint:

Although the choice of a unique DDP does not require a prior (meta-) DDP, it does require antecedent doxastic attitudes on which to base a choice. To assess the probable consequences of this or that policy of forming physical-world beliefs, for example, one

needs some doxastic attitudes toward propositions that describe typical events in the physical world and relationships between the physical world and one’s mental states. But such doxastic attitudes are absent in the internal standpoint. (Goldman 2001, 51)

Goldman is supposing that the goal of cognition is getting truth and avoiding falsity, and that a ddp should thus be evaluated in terms of its conduciveness to getting truth and avoiding falsity.

The point, then, is that since choosing a ddp would require beliefs about the degrees to which the proposed ddps are truth-conducive and since the internal standpoint allows for no such beliefs, it thereby also allows for no such choice.

Goldman then goes on to explain, in externalist fashion, how you could come to choose a ddp on the basis of beliefs that are not themselves the result of a ddp-based decision. The key, argues Goldman, is doxastic habit. You start the process with doxastic habit, forming beliefs via pre-programmed/automatic belief-forming processes (e.g., vision); most such beliefs are about things in your immediate environment. Then, becoming a bit self-reflective, you move inward, forming beliefs about such pre-programmed/automatic belief-forming processes themselves, including beliefs about their reliability. And last, you use such beliefs about reliability to finish the process—viz., formulating and subsequently endorsing ddps.

Let me stress that, though I am inclined to, I am in no way endorsing any of these claims.

That is not my purpose. My purpose, rather, is to highlight the structure or form of the argument:

You cannot choose an X (a ddp) from standpoint Y (the internalist standpoint), since (a) choosing an X requires having a Z (a belief) and (b) being in Y precludes having a Z. But that is okay, because you could choose an X from another standpoint (the externalist standpoint), a standpoint in which you can have a Z.

The idea of using a ddp to build, or choose, a belief system from scratch falls to a similarly

structured argument.

Think of how a logician uses the truth-table method to determine whether an argument is

valid. He starts by translating the argument, call it “G”, into his special language “Sentential”.

Then he makes a truth-table for the premises and the conclusion, looking for a bad row (where a

bad row is a row in which the premises are true and the conclusion is false); suppose that he finds

no such row. And last, he runs this finding through his validity function (which, in effect, says that an argument is valid just in case there are no bad rows in its truth-table). What is important here is that this process involves lots of beliefs, most importantly (for my purposes) the belief that

G’s truth-table has no bad rows. The logician’s evaluation of G as valid is dependent, in part, on his belief that its truth-table has no bad rows.

The same is true of how a teaching assistant uses an answer key to determine whether a student’s answer on an exam is correct. He starts by seeing what the student has as the answer; suppose that the student’s answer is “(b) Aristotle”. Then he compares this with the appropriate part of the answer key. And last, he makes his evaluation. The point, again, is that the evaluation, whatever it is, is based, in part, on his belief that the student’s answer is “(b)

Aristotle”.

This generalizes. To use a criterion c, which says that a thing t has evaluative property e just in case it has d, to evaluate something for e requires, or so it would seem, beliefs about the thing in question; at the very least, a belief about whether it has d. Without a belief about whether it has d, in what sense would the person in question be using c to evaluate for e?

But now consider the idea of using a ddp to build, or choose, a belief system from scratch. Let D be your favorite ddp, so that the question at issue is whether D could be used to build, or choose, a belief system from scratch. In terms of building a belief system from scratch,

D would have the form “If x holds relative to proposition p, then add the belief that p.”. In terms of choosing a belief system from scratch, D would have the form “If x holds relative to possible belief system B, then adopt B.”. Either way, it would seem, using D would require already having a belief—viz., either the belief that x holds relative to p (in the first case), or the belief that x holds relative to B (in the second case). This, however, would mean that D could not be used from scratch, and that, thus, it could not be used to build, or choose, a belief system from scratch; to use it at all, you would need to already have a belief system (albeit not a very big one).

With this result in hand, we can make sense of some rather hard to understand passages in Wilfrid Sellars’s “Empiricism and the Philosophy of Mind”. In the last paragraph of section

35, he argues against the possibility of non-inferential knowledge, arguing that S knows that x is

green only if S knows that he is a reliable indicator of green objects. Then in the first paragraph

of section 36, he argues that this has the consequence that observational knowledge of a singular

fact (e.g., that x is green) requires knowledge of a general fact (e.g., that he, S, is a reliable

indicator of green objects), and that this spells doom for the empiricist view according to which

observational knowledge of singular facts underwrites, and therefore is prior to, knowledge of

general facts. And then in the following paragraph, he considers a possible worry:

And it might be thought that there is an obvious regress in the view we are examining. Does it not tell us that observational knowledge at time t presupposes knowledge of the form X is a reliable symptom of Y, which presupposes prior observational knowledge, which presupposes other knowledge of the form X is a reliable symptom of Y, which presupposes still other, and prior, observational knowledge, and so on? (Sellars 1963, 168-9)

The rather hard to understand passages I have in mind (and alluded to above) come, in section 37, in answer to this worry:

37. Thus, all that the view I am defending requires is that no tokening by S now of ‘This is green’ is to count as ‘expressing observational knowledge’ unless it is also correct to say of S that he now knows the appropriate fact of the form X is a reliable symptom of Y, namely that (and again I oversimplify) utterances of ‘This is green’ are reliable indicators of the presence of green objects in standard conditions of perception. And while the correctness of this statement about Jones requires that Jones now cite prior particular facts as evidence for the idea that these utterances are reliable indicators, it requires only that it is correct to say that Jones now knows, thus remembers,1 that these particular facts did obtain. It does not require that it be correct to say that at the time these facts did obtain he then knew them to obtain. And the regress disappears. Thus, while Jones’s ability to give inductive reasons today is built on a long history of acquiring and manifesting verbal habits in perceptual situations, and, in particular, the occurrence of verbal episodes, e.g. ‘This is green’, which is superficially like those which are later properly said to express observational knowledge, it does not require that any episode in this prior time be characterizable as expressing knowledge. (Ibid., 169)

Yes, if Sellars were trying to give a ddp for rationally building, or choosing, a belief system from scratch, then his view that knowledge of singular facts requires knowledge of general facts would be an utter failure. For how would you ever get started? You could not start with knowledge of a

singular fact, since to have such knowledge you would need to already have knowledge of a general fact. Nor could you start with knowledge of a general fact, since to have such knowledge you would need to already have knowledge of a singular fact (i.e., knowledge of a positive instance). But that is not what Sellars is trying to give. Sellars is trying to give a theory of knowledge; he is trying to explain, or shed light on, what it would be for S, at t, to know that x is

F. Sellars is saying that, on his theory of knowledge, what matters for knowledge, at t, is what

S’s belief system is like at t, not what S’s belief system was like prior to t, or even whether S had a belief system prior to t. All that matters is whether S, at t, has the requisite general and particular beliefs to do well at the game of giving reasons—so that S, at t, has the requisite general beliefs to adequately defend his particular beliefs, and so that S, at t, has the requisite particular beliefs to adequately defend his general beliefs.

2.4.4 Against the IO 1

The first sub-argument in the IO 1, remember, is supposed to show that there is no connection between coherentist justification and truth:

(1) In order for it to be highly likely that the beliefs in a belief system are true, the system needs to be connected to the outside world. (2) A belief system’s being coherent neither involves nor entails any sort of connection between the system and the outside world. (3) If coherentism is true, then justification is solely a matter of coherence. ------(4) Coherentist justification is not connected to truth.

The problem is supposed to be that likelihood of truth, but not coherence, requires that there be a connection of some sort between the subject’s belief system and the outside world.

The problem, however, is with the objection, not coherentist justification. (1) gives a necessary condition for the likelihood of truth. If (2) were the claim that coherence in no way makes it likely that the necessary condition in question is satisfied, then (4) would follow just fine. (2), though, is a much weaker claim—viz., the claim that coherence in no way guarantees

that the necessary condition in question is satisfied. And this, of course, is perfectly compatible with the claim that coherence makes it highly likely, though not guaranteed, that the subject’s beliefs are connected in some way to the outside world (so that the necessary condition in question is satisfied), which, in turn, is itself perfectly compatible with the claim that coherence

makes it highly likely that the subject’s beliefs are true (so that (4) is false).

2.4.5 Against the IO 2

The challenge, again, is that since coherence neither involves nor entails any sort of system-world connection, coherentist justification is not connected to truth. This time, however, the worry is that, since coherence neither involves nor entails any such connection, coherentism allows for justification in isolation scenarios, which itself is not in keeping with the claim that coherentist justification is connected to truth.

The reliabilist, it would seem, is again the key (just as he was with the ASO 2). Imagine that you are watching a basketball game, and that Stretch, your favorite player, is about to shoot a free throw. Imagine further that Stretch has made 93% of his free throws over the past five seasons, and that you are well aware of this. And imagine, finally, that you infer by enumerative induction, which, suppose, is highly reliable, that Stretch will make the free throw (i.e, the one he is about to shoot). The reliabilist, it would seem, would say that your belief is justified, given that enumerative induction is reliable (in the world in question), and given that the input beliefs, suppose, are themselves justified. But imagine that Stretch is part of a fix, and that, as a result,

Stretch (unbeknownst to you and to almost everyone else) will try to miss the free throw. Then it would not be likely that your belief is true (given that Stretch, on this occasion, will try to miss)— though, it would seem, the reliabilist would still say that your belief is justified, since, it would seem, your belief would still be the result of a reliable process. Hence, reliabilism, it would seem, makes for a nice counterexample to (7) in the IO 2—i.e., the claim that if X-ist justification is

connected to truth, then truth is likely whenever X is present—in that reliabilism, though seemingly allowing for scenarios in which there is justification but not a high likelihood of truth, nonetheless secures a tight connection between justification and truth.

2.5 Conclusion

Here is the dialectic. First, the mind-internalist foundationalist is not in a position to

press alternative-systems-type objections or isolation-type objections against the coherentist,

since the mind-internalist foundationalist too is faced with such objections. Second, the mind-

externalist foundationalist (e.g., Alvin Goldman) is not in a position to press alternative-systems-

type objections or isolation-type objections against coherentism in general, because the mind-

externalist coherentist is immune to such objections. Third, the mind-externalist foundationalist

could press alternative-systems-type objections and isolation-type objections against the mind-

internalist coherentist, but, in doing so, he could not show that mind-internalist coherentism—qua

a kind of coherentism—is false. After all, the mind-internalist foundationalist, such as the

evidentialist, would fall to similar objections. And fourth, each version of each such objection

fails in and of itself, in that each version of each such objection has a false premise or a fallacious

sub-argument (or both).

The upshot, to repeat, is that, contrary to orthodox epistemology, there is nothing to be learned—vis-à-vis the question of whether justification is skyscraper-like in structure—from either alternative-systems-type worries or isolation-type worries.

1 This premise is needed in order to secure the move from (2) to (3), because if coherentist justification were in part a matter of something besides coherence and if at least some of the various incompatible yet fully coherent possible belief systems were to fair poorly in terms of the something else in question, then— contra (3)—it would not be the case that coherentism looks with favor on each of the various incompatible yet fully coherent possible belief systems.

2 The qualification “possible” is needed in light of the fact that it is not true, strictly speaking, that there are lots of incompatible yet fully coherent belief systems such that any self-consistent belief whatsoever is a member of at least one such system. For example, no one holds the belief that President Bush has thirty- four legs, and thus no fully coherent belief system has as a member the belief that President Bush has thirty-four legs.

3 You might worry that I am being a bit uncharitable, sticking the objector with the claim that the coherentist thinks that justification attaches not just to beliefs but also to belief systems. But remember (from Chapter 1): talk of a belief system’s being justified, as I understand it, is equivalent in meaning to talk of a belief system’s being justification-conferring (so that the beliefs therein are justified). The point, either way, is that the objector is implicitly assuming that the coherentist thinks that the beliefs in a belief system are justified just in case the belief system is coherent.

4 Jack Crumley also understands the objection in this manner. See Crumley 1999, p. 146.

5 Susan Haack and Laurence BonJour also construe the objection along these lines. See Haack 1993, p. 26, and BonJour 1985, p. 107.

6 That (1) is true in no way by itself entails that coherence is inapt—across the board—as a ddp. It does not preclude coherence from helping with decisions at the local level. If, for example, S is wondering whether to add the belief that p or to add the belief that not p, and if S’s belief system is such that adding the belief that p would yield a system with a higher degree of coherence than would adding the belief that not p, then coherence dictates in favor of adding the belief that p. It does not even preclude coherence from helping with decisions at the global level. If, for instance, S is wondering whether to have belief system D or to have belief system E, and if D is coherent whereas E is not coherent, then coherence dictates in favor of D.

7 Louis Pojman can also be read as thinking of the Isolation Objection along these lines. See Pojman 2001, 118-9.

8 Lewis and Pojman can be read in a radically different manner—a manner, moreover, in line with the ASO 2. The worry, perhaps, is that since coherence could in no way discriminate between the incompatible yet fully coherent possible belief systems and thus could in no way make any such system more likely than any other such system, coherence could in no way serve to account for, or make sense of, how it is that you, say, have good reason for your beliefs. Put another way, the worry is perhaps just the ASO 2 again—viz., the worry, in effect, that since coherence could in no way favor any such system over any other such system (because each such system is fully coherent) and since coherence could in no way make all such systems likely (because such systems are competitors and thus because a high likelihood for one would preclude a high likelihood for the others), coherence could in no way make any such system likely. But, because the question of whether this is the better of the two readings is merely an interpretive question, I shall here leave matters at that—saying, in the end, just that Lewis and Pojman can be read in such two very different ways.

9 Jack Crumley can also be interpreted as understanding the IO in this fashion. See Crumley 1999, 149.

10 This objection starts with the claim that coherentist justification is solely a matter of coherence. It then works to the claim that the following scenario is possible: S’s belief system is coherent (and therefore justified), though entirely isolated from the world. And, last, it moves from there together with the claim that it would not be likely that S’s beliefs are true to the claim that coherentist justification is not connected to truth. But in order for the latter move to go through, there needs to be a premise to the effect that X-ist justification is connected to truth only if truth is likely whenever X is present.

11 Strictly speaking, even further modifications would be needed—since the coherentist, but not the evidentialist, allows for circular chains of inferential relations. But this should not be a problem, given that non-circular chains of inferential relations are just as easy to manufacture (in logical space) as are circular chains of inferential relations.

12 There is a sense in which you can pick your non-inferential beliefs; namely, by performing certain actions. You could bring about the belief that the lights are off, for example, by turning off the lights. But, of course, in that sense you can also pick your experiences. You could bring about an experience of darkness, for instance, by turning off the lights.

13 It is true that certain externalist theories of content would not allow for such scenarios. Such theories, though, would be of no help to the mind-internalist foundationalist, dialectically speaking. First, it is quite common for mind-internalist foundationalists to use evil-demon scenarios against mind-externalist foundationalists, such as the Goldman-style process reliabilist. And second, the mind-internalist coherentist could just as well appeal to such theories of content.

14 See Cohen 1984 for more on this objection to mind-externalist views such as process reliabilism.

15 Alvin Goldman makes this point in “The Internalist Conception of Justification” (1980).

16 Holly Goldman draws the same distinction in “Doing the Best One Can”.

CHAPTER 3

THE DATABASE OBJECTION

The Database Objection is aimed not at coherentism in general, but at explanatory, or explanationist, coherentism in particular. In this respect, it differs significantly from the

Alternative-Systems Objection, the Isolation Objection, and the Experience Objection (to be discussed in the next chapter).

The main parties to this dispute are just four: James Cornman in “Foundational versus

Nonfoundational Theories of Justification” (1977), George Pappas in “Explanatory Coherence and Data Sentences” (1992), Keith Lehrer in Knowledge (1974) and Theory of Knowledge (2000), and William Lycan in Judgement and Justification (1988) and “Plantinga and Coherentisms”

(1996). Cornman, Pappas, and Lehrer argue that the Database Objection is decisive against explanationist coherentism. Cornman and Pappas, in addition, argue that this blow to explanationist coherentism is itself a major blow to coherentism in general. Lehrer, of course, disagrees on the latter point, arguing instead that the fall of explanationist coherentism is a significant victory for non-explanationist versions of coherentism such as his. Lycan, on the other hand, uses the Database Objection not against explanationist coherentism, but for explanationist coherentism. Lycan develops a version of explanationist coherentism, and touts it, in part, on the grounds that it can answer the Database Objection.

The objection, in structure, has five parts. First, there is a premise saying that the explanationist coherentist needs to meet the database challenge (my terminology). Second, there is a premise saying that he cannot do so. Third, there is the main conclusion, the claim that

explanationist coherentism is false and should thus be rejected. Fourth, there is an argument in support of the first premise. And fifth, there is an argument in support of the second premise. I shall give three very different readings of the database challenge, and thus three very different readings of the Database Objection itself. I shall then argue that on each such reading the argument fails, since on each such reading the challenge is misplaced; even if, as the second premise says, the explanationist coherentist could not meet the database challenge, it would not follow that explanationist coherentism is false.

3.1 The Objection

The task in this section is to set out the objection, in each of its forms. I shall start with

Lycan’s discussion, highlighting the first version of the database challenge and thus the first version of the Database Objection. Then I shall move to Cornman’s and Pappas’s discussions, drawing out the second and third versions of both the database challenge and the Database

Objection. And then, last, I shall move to Lehrer’s discussion, in which can be found the first and third versions of each.

Let me stress, before starting, that I am not suggesting that Cornman and Pappas mean to be giving multiple versions, or that Lehrer means to be giving multiple versions, or even that these theorists (including Lycan) think that there are multiple versions to be given. Rather, I am suggesting that there are three very different but easily confused worries, that each such worry can (at the very least) be read into the literature (in some cases into the same discussion), and that such theorists might be (but might not be) running the worries together. The aim is to tease out as many versions of the Database Objection as can be teased out, as this will only serve to better establish the overall conclusion of this chapter—viz., the claim that the Database Objection, in all of its possible forms, is a complete failure.

3.1.1 Lycan

Lycan starts by arguing that though at least some justified beliefs are justified in virtue of explaining, it is not the case that all justified beliefs are so justified; some justified beliefs explain nothing. He cites, as an example, his belief that there are little moving spots in his visual field. It is justified, but does not explain anything else he believes. He also gives a regress argument, saying that if the explanationist coherentist were to insist that all justified beliefs are explainers, then the explanationist coherentist would be stuck with either an infinite regress of explainers, or a circle of explainers. The implicit premise, of course, is that neither an infinite regress nor a circle would make for justification.1

The challenge for the explanationist coherentist, says Lycan, is to explain how the regress-stoppers are themselves justified. That is, the challenge is to account for the justification of the beliefs in the database—the beliefs that, in the end, serve as the ultimate explanatory data.

This, in a nutshell, is the database challenge, as Lycan sees things.

Lycan considers and rejects two proposals, the first of which is that a database belief is justified in virtue of being deduced from, or entailed by, an explainer. Yes (says Lycan), some justified non-explainers are justified in virtue of being deduced from, or entailed by, explainers.

But no, it cannot be the case that a database belief is justified in such a fashion. Why? Because a belief is justified in virtue of being deduced from, or entailed by, an explainer only if the explainer is antecedently justified. And because nothing is antecedent in justification to a database belief, given that a database belief is a regress-stopper, and given that nothing, by definition, is antecedent in justification to a regress-stopper.

The second proposal that Lycan considers and rejects is that a database belief is justified in virtue of being explained:

This suggestion is investigated by Lehrer and by James Cornman. I think it holds more than a grain of truth, as I shall argue later on, and it has some prima facie appeal. . . . But this account incurs a difficulty not unlike the one I raised against the preceding suggestion. Unlike the deductive consequences mentioned there, the explainees in

question are presumably going to be the ultimate explainees themselves; that is, they are going to be the initial data that get the whole explanatory enterprise started in the first place. And that, essentially, is our question: How does the whole explanatory enterprise get started in the first place? The present suggestion is that we begin with a set of initial data, explain these by making certain hypotheses, defend the hypotheses on the basis of their ability to explain the data, and then defend the initial data on the ground that they are explained by a set of justified explanatory hypotheses. Right: Any elementary logic student would spot the circularity. (Lycan 1988, 164-5)

Yes (acknowledges Lycan), some justified non-explainers are justified in virtue of being explained. But a database belief cannot be justified in such a fashion. For a belief is justified in virtue of being explained only if the explainer is antecedently justified, and nothing—again, by definition—is antecedent in justification to a regress-stopper.

Thus the problem, argues Lycan, is clear. The explanationist coherentist needs to answer the database challenge, and from within the confines of coherentism (that is, without slipping into foundationalism):

As I see it, the problem for the explanationist is to break into the circle by finding independent justification for the initial data (‘independent’ in the sense of being epistemically prior to their being explained by justified explainers). (Ibid., 165)

This, however, is seemingly impossible:

It seems the explanationist must appeal to something besides coherence entirely, and hence does no longer qualify as even an impure coherentist. (Lycan 1996, 5)

To be a regress-stopper, the justification (it would seem) needs to be non-inferential; this is why the two proposals above fail. Yet, to appeal to non-inferential justification is ipso facto to go foundationalist (or, again, so it would seem).

3.1.2 Cornman and Pappas

There is a lot of overlap between Cornman’s and Pappas’s discussions of the Database

Objection, both in terms of how the database challenge is construed and in terms of how it is argued that the challenge is unanswerable. It will be best, then, to treat them together.

Consider, for starters, what Cornman and Pappas say explicitly on what the explanationist coherentist needs to do:

As stated, (c) does not require such an amendment, but it does require that there be some way to determine which statements are to be explained. A person should not be allowed to decide this arbitrarily, because, as Lehrer notes, he can limit the statements that require explanation to exactly those his favorite system explains very well.13 Thus for (c), (b), and also, I find, for any plausible construal of D5, the crucial problem for an EC-theorist is to find a nonarbitrary way to restrict the statements that are to be explained. (Cornman 1977, 293)

An EC-theorist must find a procedure for selecting a set of statements that are to be explained. He has but two alternatives: allow a person to restrict the set arbitrarily, or provide nonarbitrary but accessible particular restrictions. But the first alternative is objectionable, because it allows a person to pick just those statements that his favorite theory explains. (Ibid., 295)

Let us call statements to be explained “data statements.” Such statements are not data in any sense that recalls the notion of the given; to be a datum is merely to be in need of explanation, relative to some system of beliefs or statements. Then we need to ask how these data statements are selected. That is, we want to know how, given the explanatory coherence theory, any given statement, Si, is justifiably selected as a statement to be explained. Only if there is some way to make such a justifiable selection from within the confines of the explanatory coherence theory is such a theory apt to prove plausible. (Pappas 1992, 170-1)

For Cornman, the explanationist coherentist needs to give a non-arbitrary but accessible means of determining what is to be explained, or, put slightly differently yet (supposedly) equivalently, to give a set of non-arbitrary but accessible restrictions on what is to be explained. For Pappas, the explanationist coherentist needs to give a means of justifiably selecting what is to be explained.

Thus as Cornman and Pappas put it, the database challenge is to give a means, of the right kind, of determining or selecting what is to be explained—where the qualifier “of the right kind” is spelled out by Cornman in terms of non-arbitrariness and accessibility, and by Pappas in terms of justification.

But what would it be to give a means of determining or selecting what is to be explained?

Would it be to give an account of how a database belief is antecedently justified, which is what

Lycan would count as answering the database challenge? Or should the challenge be read quite literally, so that to give a means of determining or selecting what is to be explained is just that—

to give a means of determining or selecting what is to be explained? Even then, however, there would be questions. Is the means supposed to be for the subject, so that the challenge is to give

the subject a means of building, or choosing, a database? Or is it instead supposed to be for the explanationist coherentist himself, to be used by him in adjudicating between alternative possible databases?

Consider, first, how Cornman and Pappas argue against what I will call the “Experience

Proposal”, a proposal as to how the explanationist coherentist might try to answer the database challenge:

The natural way to eliminate further is not available to an EC-theorist. That is, he cannot limit the set to those statements that are confirmed by observation, individually and independently of any explaining theory. Such statements would be self-justifying, contrary to what N2 requires. (Cornman 1977, 293-4)

A natural answer to the question of the justified selection of data statements is that we select those statements that are individually confirmed by experience. Thus, statements to be explained would be observation statements such as “I see small spots before my eyes.” Other statements believed by a person would be justified for that person provided they serve to explain the statement about the small spots. However, it is plain that this option is closed to the defender of the explanatory coherence theory. For this approach would allow some statements to be justified on extra-systematic grounds, while the guiding principle behind all coherence theories is that all justification derives from systematic considerations. (Pappas 1992, 171)

To appeal to experience, the argument goes, is to appeal to non-inferential justification, which itself, of course, is to abandon coherentism. Hence, no theorist could consistently both endorse explanationist coherentism and endorse the Experience Proposal.

This suggests, quite straightforwardly, that what it would be to give a means of

determining or selecting what is to be explained, according to Cornman and Pappas, is to give an

account of how a database belief is (antecedently) justified. For if Cornman and Pappas

understood “use X to determine or select what is to be explained” as “a database belief is justified

by X”, then their move from “use experience to determine or select what is to be explained” to “a

database belief is justified by experience” would make perfect sense. That is, it would make

perfect sense for them to read the Experience Proposal as suggesting that a database belief is

justified by experience, since “use experience to determine or select what is to be explained” would simply mean “a database belief is justified by experience”.

But now consider how they argue against what I will call the “Social-Agreement

Proposal”. Pappas writes:

Another approach is “social”; statements to be explained are those that would be assented to by nearly everyone under specific circumstances. The statement concerning the spots would be affirmed by nearly anyone in the right circumstances—when spots are present. Thus, what would serve to justify such an observation statement would be community agreement rather than experience. To see a problem here, let us label the spot statement “X.” Now consider the further statement “The statement ‘X’ would be assented to by nearly everyone in appropriate circumstances.” The foregoing method of picking data statements is adequate only if we are justified in believing the latter statement; but what would its justification be? Either it would be an explaining member of a maximally coherent system of statements or it, too, would be a statement to be explained. Presumably it is not the latter, so it would have to be an explaining member of the system. But what does it explain? We can give it an explanatory role. First we label it, Y: The statement “X” would be assented to by nearly everyone in approporiate circumstances. and then we note that Y helps to explain another statement, namely Z: Person A assents to statement “X.” Of course, now our problem breaks out again regarding Z; why should we assume that it is a statement to be explained, a data statement? We could solve this problem by stipulating that another statement, Y2, is true, Y2: The statement “Z” would be assented to by nearly everyone in appropriate circumstances. The truth of Y2 makes Z a data statement; but what justifies Y2? Presumably it, too, is an explaining member of the maximally coherent system, so perhaps it would explain W: Person B assents to statement “Z.” It should be clear from just this fragment that the method of community agreement for selecting data statements leads to an infinite regress. That is, this method requires an infinite number of explaining statements and an infinite number of data statements, each of which is justified in order for “X” to be justified. Thus, this regress seems to be vicious, an excellent reason for rejecting the method of community agreement for selecting data statements. (Ibid., 171-2)

The only difference with Cornman’s evaluation is his diagnosis of the problem. Yes, using social agreement to determine or select what is to be explained would generate an infinite regress. But no, that is not what makes the proposal inadequate:

The preceding procedure for justifying G1 requires infinitely many explaining sentences and sentences to be explained. But that is not its problem. Like the problem for N1, its problem is that it is unjustifiably arbitrary. We need only substitute “~e1” for “e1” in G1 and e2, and, keeping all else the same, ~e1 is justified for s at t, instead of e1. But, as was

noted previously, any procedure should be rejected, if it allows someone to justify whichever one of a statement and its denial he desires to justify. (Cornman 1977, 294)

The real problem, argues Cornman, is that the proposal would make explanationist coherentism overly permissive.

This, in contrast to how they argue against the Experience Proposal, suggests (again, quite straightforwardly) that what it would be to give a means of determining or selecting what is to be explained is to give a means of rationally building, or choosing, a database—viz., that the challenge is to give a doxastic decision procedure (ddp) for rationally building, or choosing, a database. This would certainly explain the worries about an infinite regress. To use social agreement to rationally add the belief that p to your database, you would need to be justified in believing that p is socially agreed on. If, as Cornman and Pappas suggest, the latter belief would have to be an explainer, then it, to be justified, would have to explain a justified database belief— for example, the belief that such and such person assents to p. Let this proposition be q. To have used social agreement to rationally add the belief that q to your database, though, you would have had to have been justified in believing that q is socially agreed on. And off we go.

What we have so far, then, are two very different readings of the database challenge.

First, there is the reading suggested by their critiques of the Experience Proposal, on which the challenge is to give an account of how a database belief is justified. This is how Lycan reads the worry. Second, there is the reading suggested by their critiques of the Social-Agreement

Proposal, on which the challenge is to give a means of rationally building, or choosing, a database.2

It would seem, though, that neither reading can make sense of both their critiques of the

Experience Proposal, and their critiques of the Social-Agreement Proposal. On one hand, it would seem that were the first reading correct, then the problem with the Social-Agreement

Proposal would be not that it would generate an infinite regress or that it would make explanationist coherentism too permissive, but that it would make a justified database belief non-

inferentially justified. It would be saying that social agreement is sufficient for justification—so that it could be the case that S’s belief that p is a database belief (because p is socially agreed on), even though S has no reason whatsoever for thinking that p. On the other hand, it would seem that were the second reading correct, then the problem with the Experience Proposal would be not that it would make a database belief non-inferentially justified, but either that it would generate an infinite regress, or that it would make explanationist coherentism too permissive. To use

experience to rationally add the belief that p to your database, you would need to be justified in

believing that p fits with your experiences. If, as Cornman and Pappas might suggest, the latter

belief would have to be an explainer, then it, to be justified, would have to explain a database

belief—for example, the belief that q. To have used experience to rationally add the belief that q

to your database, though, you would have had to have been justified in believing that q fits with

your experiences. Etc.3

The only way (I can think of) of reconciling the reading suggested by their critiques of the Social-Agreement Proposal with the reading suggested by their critiques of the Experience

Proposal is by invoking strong access-internalism, the view (roughly put) that X makes S’s belief that p justified only if S is aware of X. Given the reading suggested by their critiques of the

Experience Proposal, the Social-Agreement Proposal would say that a database belief is justified by social agreement. This together with strong access-internalism would then imply that your belief that p (which, suppose, is a database belief) is justified only if you are aware of the fact that p is socially agreed on—where to be aware of this would be to have a justified belief to the effect that p is socially agreed on. And with this, the regress would be off and running.

Pappas, though, is not an access-internalist, let alone a strong access-internalist. Nor is a clear-headed explanationist coherentist. So though this alternative reading of their critiques of the

Social-Agreement Proposal would better fit with the reading suggested by their critiques of the

Experience Proposal (since each such critique would involve the same version of the database challenge), it would make their argument in support of the claim that the explanationist

coherentist cannot answer the challenge an easy target—as strong access-internalism is itself an easy target.

But perhaps there is a way of reconciling their critiques of the Experience Proposal with their critiques of the Social-Agreement Proposal. Perhaps, that is, there is a way of reading their critiques of the Experience Proposal as involving the version of the database challenge according to which the challenge is to give a means of rationally building, or choosing, a database. You might think not, reasoning as follows: if so (that is, if the challenge, in their minds, were to give a means of rationally building, or choosing, a database), then the charge against the Experience

Proposal would be similar in form to the charge against the Social-Agreement Proposal; however, it is not similar in form; the charge, instead, is that to invoke experience is to invoke non- inferential justification, and thus to abandon explanationist coherentism altogether. But that would be too quick. Perhaps they think that if X is being proposed as a means of rationally building, or choosing, a database, then, in addition, X is being proposed as what it is that makes a database belief justified; this would explain their charge that the Experience Proposal invokes non-inferential justification. And perhaps the reason why they give just the argument about invoking non-inferential justification—not also giving an argument similar in form to the argument they give against the Social-Agreement Proposal—is that, since the argument about invoking non-inferential justification is sufficient for disposing of the Experience Proposal, giving a further argument would be overkill. This reading, as far as I can tell, is consistent with everything they say.

I should stress, again, that I am not saying that Cornman and Pappas mean to be giving two versions of the database challenge, or that charity dictates that they be read as giving two versions of the database challenge, or that there are no plausible readings on which they are giving just one version of the database challenge. The claim, rather, is that there are (at least) two very different but easily confused database worries, and that each such worry is (at least) suggested by things they say.

Yet a third database worry is suggested by Cornman’s argument in support of the claim that the explanationist coherentist needs to meet the database challenge:

The worry, it seems, is that without an answer to the database challenge, the explanationist

coherentist would be open to a variant of the over-permissiveness version of the Alternative-

Systems Objection (i.e., the ASO 1). The challenge, then, is to give a means—to be used, by the

explanationist coherentist, in eliminating possible belief systems for justification—of eliminating

possible databases for justification. The resulting view would say that though explanationist

coherence is necessary for justification, it is not sufficient; it is further required that the database

in question be of the right kind.

But why non-arbitrary? And why accessible? The restrictions would need to be non-

arbitrary, I take it, because otherwise the resulting view would be ad hoc. It would, without any

reason, legislate against certain databases and doxastic systems in favor of certain other databases

and doxastic systems. The restrictions would need to be accessible, in turn, because otherwise the

resulting view would allow both for unjustified beliefs in accordance with the available evidence,

and justified beliefs not in accordance with the available evidence. Consider, for instance, what

Cornman says against what I will call the “Truth Proposal”:

He might propose, instead, that the desired set include all and only true observation statements. But then the following objection arises. Let us assume that either Newton’s theory or Einstein’s theory (but not both) explains only true observation statements. If this present version of EC-theory is correct, then for all we can determine, the Newtonian theory and what it explains might be justified for us, because it explains only what is true. Most of us, following what scientists tell us, would believe—mistakenly and with no

means for discovering our mistake—that Einstein’s theory is justified. And anyone who doggedly has maintained the Newtonian theory would be justified in his belief even though he has no way to support his view except to say that what his theory explains is true. This is surely an objectionable feature of a theory of empirical justification, and it should be avoided. It seems, then, that an EC-theorist needs something more than the mere (lucky) truth of observation statements in order to limit the set of statements to be explained. He needs some nonarbitrary means that is more reliable than a mere guess or dogmatic assertion for choosing which of such conflicting theories is more reasonable. (Ibid., 294)

The point, it seems, is that, since truth is not accessible in the right way, the Truth Proposal would allow for possible worlds in which, though the evidence at hand differs in no way from the evidence at hand in the actual world, the crank’s belief (which is not in accordance with the available evidence) in Newton’s theory is justified while our beliefs (which are in accordance with the available evidence) in Einstein’s theory are unjustified.

I should note that Pappas gives no argument for the claim that the explanationist coherentist needs to meet the database challenge. He simply lays it down that explanationist coherentism is plausible only if the database challenge is met. Presumably, though, he has in mind something like the over-permissiveness worry.

Cornman’s and Pappas’s discussions, in over-view form, run as follows. First, they spell out explanationist coherentism in terms of the notion of having a maximum of explanationist coherence. Second, they spell out the notion of having a maximum of explanationist coherence in terms of the notion of a database. Third, they give the database challenge, which is either (a) a challenge to account for the (antecedent) justification of the beliefs in the database, or (b) a challenge to give a means of rationally building, or choosing, a database, or (c) a challenge to give a means—to be used not by the subject but by the explanationist coherentist—of adjudicating between alternative possible databases. Fourth, they argue (explicitly in Cornman’s case and implicitly in Pappas’s case) that the explanationist coherentist needs to meet the challenge, since otherwise the view would be too permissive. And fifth, they argue that the challenge is unaswerable, and that, therefore, explanationist coherentism is false. Cornman argues for this by arguing against the Experience Proposal, the Truth Proposal, and the Social-

Agreement Proposal. Pappas argues for it by arguing against the Experience Proposal, the Social-

Agreement Proposal, and Lycan’s proposal (to be discussed later).

3.1.3 Lehrer

Lehrer starts by construing explanationist coherentism in terms of the notion of having a maximum of explanationist coherence:

S’s belief that p is justified if and only if (a) it plays an explanatory role in his belief system and (b) his belief system has a maximum of explanationist coherence (with respect to the possible belief systems of which he can conceive).

Then he construes the notion of having a maximum of explanationist coherence in terms of the notion of having greater explanationist coherence:

With respect to the possible sets of beliefs C1, C2, C3, . . . of which S can conceive, C1 has a maximum of explanationist coherence if and only if it is not the case that there is a set

of beliefs Cn such that (a) C1 ≠ Cn and (b) Cn has a greater degree of explanationist coherence than C1.

Then he construes the notion of having greater explanationist coherence in terms of the notions of

being consistent, explaining more, and explaining better:

C has a greater degree of explanationist coherence than C’ if and only if either (a) C is consistent, C’ is inconsistent, and other things are equal, or (b) C explains more than C’ and other things are equal, or (c) C explains better than C’ and other things are equal.

And then, after all of this, he gives the database challenge:

The first problem raised by our explanatory theory of justification concerns comparing systems with respect to explanatory coherence. Our theory tells us that one system has greater explanatory coherence than a second if the first leaves less unexplained or explains better what it does explain than does the second. Even so, one system may leave less unexplained and explain better what it does explain by containing less to be explained. One system may admit statements of unexplained facts that the other excludes. To reduce what is unexplained, one may refuse to concede the truth of those statements that need explanation. Explanation involves those statements that do the explaining, on the one hand, and those that describe what is to be explained, on the other. One can increase the explanatory coherence of a system either by adding statements that explain or by subtracting statements to be explained. The method of increasing the explanatory coherence of a system by decreasing what is to be explained must be limited. Otherwise, we may obtain a maximum of coherence only by securing a minimum of content.

The foregoing remarks may be illustrated with a very simple formal example. Compare any system of beliefs within science to the following. Take a language with one observation predicate ‘O’ and one theoretical predicate ‘T.’ Then adopt a system affirming that everything is T and that everything T is O, hence, that everything is O. Let the system contain only these sentences. We can now get a maximum of coherence by adding just those observation sentences to our system that fit with our one empirical law. More concretely, if we wish to have the law affirming that all dragons breathe fire, we may then add the ‘observation’ sentences that object 1 is a fire-breathing dragon, object 2 is a fire-breathing dragon, and so forth. The coherence between the law and observation statement will be perfect, and the absurdity of the system will be manifest. To avoid this sort of implausibility, philosophers have imposed further limitations on what kinds of statements may belong to a justificatory system. (Lehrer 2000, 108-9)

The challenge, according to Lehrer, is to impose permissibility restrictions for database beliefs, so that there is a means of adjudicating between alternative possible databases and thus between alternative possible belief systems. Otherwise, a subject could make his system have a maximum of explanationist coherence by simply restricting his database beliefs so as to fit his favorite explainers.4

So far, then, it looks as though Lehrer’s version of the database challenge is identical to the third version suggested in Cornman’s discussion. The challenge is to give a means—to be used not by the subject but by the explanationist coherentist—of adjudicating between alternative possible databases.

However, some of the things that Lehrer says in support of the claim that the challenge is

unanswerable suggest a reading on which the challenge is to account for how a database belief is

justified. For instance, he argues against the Conditioned-Response Proposal and the Spontaneity

Proposal (my terms) in the way that Cornman and Pappas argue against the Experience Proposal.

Namely, he argues that invoking either conditionedness or spontaneity would be tantamount to

abandoning explanationist coherentism (not saving it), since invoking either conditionedness or

spontaneity would be tantamount to invoking non-inferential justification.

3.1.4 Summary

What we have, all told, is three very different versions of the Database Objection. First, we have a version on which (a) the database challenge is to account for the antcedent justification of the database (i.e., the ultimate data of explanation), (b) the rationale for the challenge is that without such an account the explanationist coherentist would be stuck with either an infinite regress of explainers or a circle of explainers (neither of which would make for justification), and

(c) the worry is that meeting the challenge seemingly requires invoking non-inferential justification (which, of course, is an anathema to the explanationist coherentist). Second, we have a version on which (a) the database challenge is to give a ddp for rationally building, or choosing, a database, (b) the rationale for the challenge is that without such a ddp people could build their databases so as to fit their favorite explainers (so that the resulting view would be much too permissive), and (c) the worry is, first, that the resulting view, if the challenge were met, would make justification require an infinite regress (Pappas’s worry) and, second, that the resulting view, if the challenge were met, would still be much too permissive (Cornman’s worry). And third, we have a version on which (a) the database challenge is to find a means, to be used by the explanationist coherentist himself, of adjudicating between alternative possible databases, (b) the

rationale for the challenge is that without such a means the explanationist coherentist would have

to allow for any (probabilistically) consistent possible database whatsoever and so would be open

to a variant of the over-permissiveness version of the Alternative-Systems Objection, and (c) the

worry is that—given the failure of proposals such as the Experience Proposal, the Truth Proposal,

the Conditioned-Response Proposal, and the Spontaneity Proposal—no such means can be found.

I should add that an alternative rationale for the second version of the database challenge is hinted at in Lycan’s discussion. Suppose that you wanted to rebuild your belief system, and to do so in accordance with the correct account of justification. Suppose further that you wondered

how the rebuilding process would work were explanationist coherentism (in the manner of

Lycan’s account) the correct account of justification. It would be fairly clear how you should go about making inferences, given a set of beliefs from which to infer; namely, you should make valid deductive inferences, and cogent instances of inference to the best explainer. It might not be clear, though, how you should go about getting the whole thing started. How is it, you might wonder, that you should go about getting a database from which such inferences could be made?

The challenge to the explanationist coherentist—a challenge from you to him—would be to give you a procedure for building a database, a procedure for getting the data on the basis of which you could then make valid deductive inferences and cogent instances of inference to the best explainer. The rationale for the second version of the database challenge, then, would be not that without such a procedure explanationist coherentism would be too permissive, but that without such a procedure you would have no way of getting the rebuilding process underway.

3.2 The First Version Rebutted

The foundationalist and the coherentist, remember (from Chapter 1), disagree both on how inferential justification works and on what grounds inferential justification. The foundationalist, on one hand, says that inferential justification is part-to-part (moving from belief to belief) and uni-directional (moving in just one direction), and that every inferentially justified belief gets its justification, in the end, from one or more non-inferentially justified beliefs. The coherentist, on the other hand, says that inferential justification is whole-to-part (moving from belief system to belief), and that every inferentially justified belief gets its justification, at least partially, from the coherence of the belief system of which it is a member.

It should be clear right away, then, that the version of the database challenge in the first version of the Database Objection is misplaced. The explanationist coherentist is not obliged to give an account of the antecedent justification of the beliefs in the database, or to give an account of the justification of the beliefs in the database that differs in kind from the account he gives for the beliefs not in the database. The explanationist coherentist is a holist, and so thinks that every

justified belief, whether in the database or not, gets its justification from the belief system of which it is a member, which, in turn, gets its justification (or status as justification-conferring), at least partially, from the fact that it is coherent.

Consider, again, how the objection (in Lycan’s hands) works, noticing the numerous points at which it assumes the foundationalist’s part-to-part-and-uni-directional conception of inferential justification. Why is it, according to the objection, that there needs to be a database, a set of beliefs that are justified but not in virtue of explaining? Because some beliefs are justified in virtue of explaining, because a belief is justified in virtue of explaining only if the belief explained is antecedently justified, and thus because without a database the explanationist coherentist would be stuck with either an infinite regress of explainers or a circle of explainers.

Why is it, according to the objection, that it cannot be the case that the beliefs in the database are justified in virtue of being entailed by beliefs not in the database? Because a belief is justified in virtue of being entailed only if the entailer is antecedently justified, and because nothing, by definition, is antecedent in justification to a database belief. Why is it, according to the objection, that it cannot be the case that the beliefs in the database are justified in virtue of being explained by beliefs not in the database? Because a belief is justified in virtue of being explained only if the explainer is antecedently justified, and because nothing—again, by definition—is antecedent in justification to a database belief. And why is it, according to the objection, that the challenge is seemingly unanswerable. Because to be a regress-stopper the justification (attaching to the beliefs in the database) would need to be non-inferential, and because the explanationist coherentist, qua coherentist, cannot appeal to non-inferential justification. The explanationist coherentist, at each such point, should reject the worry by simply invoking his holistic, or whole- to-part, conception of inferential justification—pointing out that on such a conception there are no beliefs antecedent in justification to other beliefs, and that, because of this, on such a conception there are no regresses of justification.

This is similar to the dialectic pertaining to the foundationalist’s regress argument in support of non-inferential justification. The foundationalist gives the regress argument, and in

doing so assumes (quite commonsensically) that inferential justification is part-to-part and uni-

directional. The coherentist agrees with the second part of the argument, that if the regress (once

started) were to stop with an unjustified belief, or circle back to the target belief, or continue on

without end, then the target belief would not be justified. The coherentist further agrees with the

third part, that there is justification. The fourth part—viz., the conclusion that there is non-

inferential justification—the coherentist finds entirely mysterious. The coherentist thus rejects

the first part (i.e., the part-to-part-and-uni-directional conception of inferential justification), and

in so doing moves to the whole-to-part conception—thus keeping the regress from ever starting in

the first place.

The thing to see, to put the point in a slightly different way, is that the non-explanationist coherentist is faced with an objection that is perfectly parallel to the first version of the Database

Objection, and that, just like the first version of the Database Objection, the objection facing the non-explanationist coherentist is really just an instance of the foundationalist’s regress argument.

Compare:

(1) Some beliefs are justified in virtue of explaining. (2) A belief is justified in virtue of explaining only if the belief explained is antecedently justified. ------(3) Without a database (i.e., a set of beliefs that are justified but not in virtue of explaining), the explanationist coherentist would be stuck with either an infinite regress of explainers or a circle of explainers. ------(4) The explanationist coherentist needs to supply a database. (5) The explanationist coherentist cannot do so by saying that the beliefs in the database are justified in virtue of being entailed by beliefs not in the database—because a belief is justified in virtue of being entailed only if the entailer is antecedently justified, and because nothing, by definition, is antecedent in justification to a database belief. (6) The explanationist coherentist cannot do so by saying that the beliefs in the database are justified in virtue of being explained by beliefs not in the database—because a belief is justified in virtue of being explained only if the explainer is antecedently justified, and because nothing, by definition, is antecedent in justification to a database belief.

(7) The explanationist coherentist cannot do so by saying that the beliefs in the database are justified non-inferentially—because the explanationist coherentist rejects the view that there is non-inferential justification. ------(8) The explanationist coherentist cannot supply a database.

(1) Some beliefs are justified in virtue of being deductively or inductively supported by other beliefs. (2) A belief is justified in virtue of being deductively or inductively supported by other beliefs only if the supporting beliefs are antecedently justified. ------(3) Without a base (i.e., a set of beliefs that are justified but not in virtue of being deductively or inductively supported by other beliefs), the non-explanationist coherentist would be stuck with either an infinite regress of supportees or a circle of supportees. ------(4) The non-explanationist coherentist needs to supply a base. (5) The non-explanationist coherentist cannot do so by saying that the beliefs in the base are justified in virtue of being deductively supported by beliefs not in the base— because a belief is justified in virtue of being deductively supported by other beliefs only if the supporting beliefs are antecedently justified, and because nothing, by definition, is antecedent in justification to a base belief. (6) The non-explanationist coherentist cannot do so by saying that the beliefs in the base are justified in virtue of being inductively supported by beliefs not in the base— because a belief is justified in virtue of being inductively supported by other beliefs only if the supporting beliefs are antecedently justified, and because nothing, by definition, is antecedent in justification to a base belief. (7) The non-explanationist coherentist cannot do so by saying that the beliefs in the base are justified non-inferentially—because the non-explanationist coherentist rejects the view that there is non-inferential justification. ------(8) The non-explanationist coherentist cannot supply a base.

Each objection involves a regress worry. Each regress worry is premised on the foundationalist’s part-to-part-and-uni-directional conception of inferential justification. And thus each objection is unsound, given that neither the explanationist coherentist nor the non-explanationist coherentist accepts the part-to-part-and-uni-directional conception of inferential justification.

Whether the explanationist coherentist is right in thinking that inferential justification is holistic, or whole-to-part, is another issue. For all I have said, it might be that he is wrong in so thinking. The Alternative-Systems Objection and the Isolation Objection are but two of a slew of objections to the coherentist’s brand of whole-to-part-ism. The point now is just that the

challenge in the first version of the Database Objection is premised on something the explanationist coherentist rejects, and that, in that respect, the challenge is misplaced.

Lycan’s response is much different. First, he argues for what he calls the “Principle of

Conservatism”, the view that you should prefer hypothesis h over hypothesis h’ if h better coheres with your beliefs—i.e., not just your beliefs for which you have good reasons, but also your beliefs for which you do not have good reasons—than does h’ (and if other things are equal).

Second, he argues from the Principle of Conservatism to the view that a belief is at least a little bit justified solely in virtue of being held.5,6 Third, out of the view that a belief is at least a little

bit justified solely in virtue of being held he formulates what he calls the “Principle of Credulity”,

according to which you should accept at the outset whatever seems true to you. And fourth, he

argues from the Principle of Credulity to the rebuttal of the Database Objection:

At any given time we involuntarily find ourselves holding any number of beliefs—at least those produced by perception and memory, though I shall not yet make any appeal to those faculties as justifying. Call such unconsidered beliefs “spontaneous” beliefs.10 They are mostly about our immediate environment, about past events, perhaps about our own mental states, and probably more. Now, since all their contents are things that seem true to us, the Principle of Credulity tacks those propositions in place long enough for them to serve as data for explanation. And since they are justly available for explaining, they soon acquire the kinds of coherence that are constituted by a proposition’s being explained (on which more shortly). (Lycan 1996, 5-6)

The problem, remember, is that the beliefs in the database need justification that is antecedent to being entailed or explained—in order to give justification to the beliefs explaining them. The

Principle of Credulity, argues Lycan, saves the day, since it entails that the beliefs in the database

have at least a modicum of (prima facie) justification solely in virtue of being held; the

explanatory enterprise is thus off and running.

But what about the second horn of the supposed dilemma? Has not Lycan, in pumping for the view that a belief is at least a little bit justified solely in virtue of being held (regardless of whether the subject has good reason, or any reason for that matter, for thinking it is true), abandoned coherentism in favor of foundationalism?

“No. Well, maybe. But if so, not as strong a foundationalism as Chisholm’s.” is Lycan’s response:

My own proposal bears some similarity to Chisholm’s principle B, in his Theory of Knowledge (1977, p. 76): “For any subject S, if S believes, without ground for doubt, that he is perceiving something to be F, then it is beyond reasonable doubt for S that he perceives something to be F.”7 But Chisholm’s principle is quite differently motivated. He believes in it because, according to him, perceptual beliefs in particular are “self- presenting” and therefore can serve as (part of) the foundation of knowledge. I am not a foundationalist—at least not as “strong” a foundationalist as Chisholm is. (Lycan 1988, 170-1)

My suspicion is that he thinks that whether his account is foundationalist is a mere terminological

issue, and that because of this it is not worth exploring.

Lycan, at that, sets this issue aside, turning to the nature and role (in his account of

justification) of the Principle of Credulity. He is quick to stress that the justification conferred on

a belief by the Principle of Credulity is highly defeasible. He is also quick to stress that it is

insufficient for the justification part of knowledge:

Does it not follow that, since our explanatory chain is only as strong as its first link (the ultimate explainees), all of our subsequent explainers will be only very weakly justified as well, and that although our beliefs may be made very minimally reasonable in this way, they are still not justified in the strong sense required for knowledge, or anything approaching it? It would follow, if that were all there was to it. In fact, I think, the minimal justification conferred on spontaneous beliefs by the Principle of Credulity can be and is reinforced in several ways by the superadded explanatory structure. This is where explanatory “coherence” comes in. (Ibid., 167, ephasis Lycan’s)

So the Principle of Credulity, as Lycan sees it, accounts for the initial, though minimal and highly

defeasible, justification of the beliefs in the database. This justification then filters to the

explainers, and as the system gains in explanatory coherence, both the beliefs in the database and

the explainers gain in justification.

But then why is the initial justification needed in the first place? Suppose, for the sake of

simplicity, that knowledge requires 5 units of justification and that the Principle of Credulity

confers 1 unit of justification on the beliefs in the database. The question, then, is why 1 unit of

justification together with coherence can yield 5 units of justification, though coherence by itself

cannot? That is, if you are willing to say that coherence can generate justification, then why not go all the way with it, saying that coherence can generate the full five units?

What is more, Lycan in no way uses the Principle of Credulity in answering the

Alternative-Systems Objection. He simply acknowledges that his view has the consequence that each of the alternative systems is justified, and says that that is how it should be.

It seems to me, therefore, that the Principle of Credulity does Lycan more harm than good. Yes, it helps with the Database Objection. But (a) it is not needed to rebut the argument

(as I argued above), (b) it threatens to put his account in with the foundationalist accounts, (c) he does not use it in rebutting the Alternative-Systems Objection, and (d) he already has to say that coherence is a justification-generator. He would be better off without it.

3.3 The Second Version Rebutted

The second version of the Database Objection is easy enough to dispense with. It makes

use of the version of the database challenge on which the challenge is to give a means of

rationally building, or choosing, a database. The problem, again, is that the challenge is

misplaced. The explanationist coherentist, like the non-explanationist coherentist and like the

foundationalist, is not in the game of giving a recipe for building a belief system. Rather, he is in

the game of giving an account—an explanatory account, to be specific—of justification. His aim

is to shed light on the structure and grounding of justification. So even if the objector were right

in saying that the explanationist coherentist cannot give a means of rationally building, or

choosing, a database, it would not follow that explanationist coherentism is false—since it would

not follow that it is false that what makes a justified belief justified is that it plays an inferential

role (of the right kind) in an explanatorily coherent belief system.

What is more, this kind of challenge is unintelligible (as I argued in the previous chapter);

no clear sense can be made of it. A logician, in using the truth-table method to determine whether

an argument is valid, needs to form a belief about the argument’s truth-table (e.g., that there are

no bad rows). A teaching assistant, in using an answer key to determine whether a student’s answer on an exam is correct, needs to form a belief about the student’s answer (e.g., that it is

“(b) Aristotle”). And (in the same way) a cognizer, in using a ddp to determine whether to add

the belief that p, needs to form a belief about whether the input conditions are met. This,

however, means that no ddp could be used from scratch, and that, thus, no ddp could be used to

build, or choose, a belief system from scratch; to use a ddp at all, a cognizer would need to

already have a belief system (albeit not necessarily a very big one).

This, remember, is not to say that the idea of using a ddp is unintelligible. With a belief system already in place, the idea of using a ddp to add to or subtract from the existing system makes perfect sense. The problem is not with the idea of using a ddp, but with the idea of using a ddp from scratch.

3.4 The Third Version Rebutted

The third version of the Database Objection is the one on which the database challenge is to find a means of adjudicating between alternative possible databases, so that there is a means of adjudicating between alternative possible beliefs systems. The worry, remember, is that without such a means the explanationist coherentist would have to allow for any (probabilistically) consistent possible database whatsoever, and so would be open to a variant of the over- permissiveness version of the Alternative-Systems Objection.

To the charge that he cannot meet the challenge, the explanationist coherentist should simply fall back on his response to the first version of the Alternative-Systems Objection. He should stress, first, that the objection is premised on the mistaken assumption that the explanationist coherentist is a mind-internalist, according to whom justification is solely a matter of explanationist coherence. He should stress, second, that both the mind-internalist non- explanationist coherentist and the mind-internalist foundationalist are faced with similar

objections. And he should stress, third, that the extent to which mind-internalist explanationist coherentism is permissive is just as it should be.

3.5 Conclusion

The upshot is twofold. First, the Database Objection fails (in each of its versions), since the database challenge is misplaced (in each of its versions). The explanationist coherentist is in no way obliged to give an account of the antecedent justification of the beliefs in the database.

Nor is he under any obligation to give a doxastic decision procedure for building, or choosing, a database, or to find a means of adjudicating between alternative possible databases. Second, the

Database Objection is not a problem in addition to the problems with which the explanationist coherentist, as a coherentist, is already faced. The first version of the objection, the one on which the challenge is to give an account of the antecedent justification of a database, is really just a special case of the foundationalist’s regress argument. The second version, where the challenge is to give a means of rationally building, or choosing, a database, is really just a version of the third version of the Alternative-Systems Objection. And the third version, on which the challenge is to give a means of adjudicating between alternative possible databases, is really just a special case of the first version of the Alternative-Systems Objection.

This should not be terribly surprising, notice. The debate between the explanationist coherentist and the non-explanationist coherentist is on the question of what makes two beliefs

(or sets of beliefs) inductively hang together. The explanationist says that what does it is that

there is a virtuous explanatory relation between them. The non-explanationist says that at least

sometimes what does it is something else. Thus other than seemingly non-explanatory yet cogent

inferential relations, any problem for the explanationist coherentist could be rephrased as a

problem for the non-explanationist coherentist.

1 Lycan has yet a third argument: “Explanation is asymmetric, in that it proceeds from stronger propositions to weaker propositions. (In a paradigm case, when P explains Q, P entails Q but is not entailed by Q.) It is

hard to imagine how a lot of little, weak propositions could band together to explain a big, strong general proposition; so it is hard to see how there could be even a nonvicious circle of explanations. But any finite or nonfinite regress of explanations would produce weaker and weaker explanations as explananda, and it seems clear that we will come to a point at which one of our explainees will be so weak and obvious and uninteresting that one could not plausibly think it justified by its ability to explain some even weaker proposition.” (Lycan 1988, 163).

2 The regress worry can be read in a slightly different manner, as the worry that if explanationist coherentism were true, then, because of the regress, the explanationist coherentist himself could not have justified beliefs, by way of the Social-Agreement Proposal, to the effect that such and such of the subject’s beliefs are database beliefs. But since what I shall say in criticism of the first reading holds equally well of this alternative reading, I shall hereafter ignore it.

3 If, instead, your belief that p fits with your experiences would have to be a database belief, then to have used experience in rationally adding it to your database you would have had to have been justified in believing that p’s fitting with your experiences itself fits with your experiences. But then what about this new belief? Would it have to be an explainer? Or would it have to be a database belief? . . .

4 This reading is supported by what Lehrer says in his summary section: “First, the explanatory coherence of a system could be increased by decreasing what needs explanation. We thus reduce the problem of explanation by systematically denying the truth of those statements describing whatever is unexplained until we obtain a very simple system in which everything is perfectly explained because there is almost nothing to explain. No explanatory function or role of statements suffices to prevent this artificial manipulation of explanatory systems.” (Ibid., 120).

5 The argument is this: “This [Principle Of Conservatism] is vindicated by our earlier considerations of the inefficiency and instability of gratuitous changes of mind. But it has a slightly startling consequence. Consider my present belief set B, and three theories T1, T2, and T3. T1 is logically equivalent to B; T2 is logically stronger than B; and T3 is incompatible with B, though T3 and B may have a large intersection. Now, T3 may outweigh T1 and T2 in explanatory advantages, in which case we should reject those theories in favor of T3. But suppose this is not so. Then we should prefer either T2 or T1, according to our strengthened canon of conservatism. T2, by virtue of its greater strength, may have an explanatory advantage over T1, or it may not, but in either case our rule of conservatism justifies us in accepting one of the two, absent any competitor other than T3. But either entails B, my present belief set. It follows that I am justified in accepting B, merely in virtue of my already holding B. Our rule of conservatism, then, is tantamount to the claim that the bare fact of one’s holding a belief renders that belief justified, to some degree; any belief at all is at least minimally warranted.” (Lycan 1988, 162).

6 Harman uses lost-evidence cases in arguing for general foundationalism, which is similar to this conservativism-inspired justification principle: No-foundations and special foundations theories require that all or most beliefs be associated with non-trivial justificatory arguments, if they are to be justified. Consider the implication of such a requirement for things one came to believe for reasons one has now forgotten. One no longer associates reasons or justificatory arguments with such beliefs. The requirement that such beliefs be associated with justificatory arguments if they are to be justified implies that such beliefs are not justified. Given the normative aspect of justification, one ought to give up any such belief and one is at fault for continuing to believe it. People tend to have many beliefs of this sort, beliefs about historical dates (the Battle of Hastings) and various other factual beliefs (about state or national capitals). The skeptical implications of the justificatory argument requirement can be quite substantial. On the other hand, there are no such skeptical implications to general foundationalism, since it does not impose any general requirement that beliefs be associated with justificatory arguments. (Harman 2003, 10) For more on lost evidence cases and/or what is standardly called “conservativism” (i.e., the view that a belief is justified just in case it is not in conflict with the subject’s other beliefs), see Goldstick 1971, Sklar

1975, Goldman 1979a, Pappas 1980, Foley 1982, Harman 1986, Adler 1990, Vogel 1992, and Christensen 1994.

CHAPTER 4

THE EXPERIENCE OBJECTION

4.1 The Objection

The Experience Objection is supposed to be a problem for coherentism in general, unlike the Database Objection (discussed in the previous chapter) which is supposed to be a problem just for explanatory, or explanationist, coherentism. The alleged problem, in short, is that coherence neither involves nor requires any fit between the subject’s beliefs and the subject’s experiences

(e.g., visual experiences), so that, to its detriment, coherentism allows for scenarios in which the subject’s beliefs are justified but radically in conflict with the subject’s total body of evidence.

Paul Moser, for one, sets out the problem quite lucidly:

The basic claim of my isolation objection is simply this: so long as the holding of coherence relations between believed or accepted propositions is sufficient for empirical epistemic justification, such justification can be divorced from one’s total empirical evidence. That is, some propositions that are justified on the coherentist account will actually be unlikely to be true given one’s total empirical evidence, including the subjective contents of one’s perceptual experiences. So my objection is: Epistemic coherentism entails that one can be epistemically justified in accepting a contingent empirical proposition that is incompatible with, or at least improbable given, one’s total empirical evidence. This isolation objection is universally applicable to coherence theories of justification so long as we acknowledge that one’s empirical evidence extends beyond the propositions one believes or accepts. The problem of course is that there is no necessary connection between (a) the holding of coherence relations, however comprehensive, between propositions one believes or accepts and (b) conformity to the subjective contents of one’s perceptual experiences. But empirical epistemic justification, by definition, requires that if one has such justification for a proposition, P, then P is likely to be true for one, relative to all one’s empirical evidence, including the contents of one’s perceptual experiences. (Moser 1989, 177)

The argument, then, has two parts. In the first part, it is claimed that since coherentist justification is solely a matter of coherence, coherentism allows for scenarios in which a belief is fully justified yet radically at odds with the subject’s experiences. In the second part, it is claimed that since the class of evidence includes not just the class of beliefs but also the class of experiences, coherentism—in allowing for scenarios in which a belief is fully justified yet radically at odds with the subject’s experiences—allows for scenarios in which a belief is fully justified yet radically at odds with the subject’s total body of evidence. The conclusion, of course, is that coherentism is false.1

Matthias Steup too presses this objection against coherentism. He gives a pair of cases, where in both you are visiting a zoo and looking at an enclosure with aardvarks, where in both you believe that there are two aardvarks in the enclosure, where in both your belief system is coherent, and where in the first you are visually experiencing two aardvarks in the enclosure while in the second you are visually experiencing three aardvarks in the enclosure. The intuition is supposed to be that though your belief is justified in the first case (where you are visually experiencing two aarvarks), your belief is unjustified in the second case (where you are visually experiencing three aardvarks)—so that, contra coherentism, it cannot be the case that justification is solely a matter of relations between beliefs. The underlying assumption, again, is that the class of evidence includes not just the class of beliefs, but also the class of experiences.2

The coherentist has but two options: (1) attack the premise according to which coherentism allows no role for experience in justification; (2) attack the premise according to which the class of evidence includes the class of experiences. I shall opt for (2)—arguing that, though quite intuitive, the claim that the class of evidence includes the class of experiences is

false.

4.2 The Trilemma Challenge

I shall start by articulating a general argument against, or challenge to, the view that

experiences can serve as reasons for beliefs—an argument that is rooted in, but goes well beyond,

the writings of Wilfrid Sellars, Laurence BonJour, and Donald Davidson. I call it “the trilemma

challenge”. Then I shall argue that none of the many extant foundationalist accounts successfully

answers the trilemma challenge, and that this suggests, quite straightforwardly and quite strongly,

that the problem is intractable.

But let me first alert you to two alternative possible responses, one of which is

articulated, though neither explicitly accepted nor explicitly rejected, by Jonathan Kvanvig and

Wayne Riggs in “Can a Coherence Theory Appeal to Appearance States?”. It starts with the

claim that experiences are propositional in content. It then uses this to support the claim that the

coherentist can define coherence in terms of both belief and experience, so that a belief is

justified only if it fits with the subject’s experiences. And, last, it uses this to support the claim

that it is not the case that coherentism—across the board—allows no role for experience in

justification.

The envisioned account, though allowing a role for experience in justification, would still

be a kind of coherentism, argue Kvanvig and Riggs. It would still not allow for non-inferential

justification, as it would still not allow for scenarios in which a belief is justified but not in any

way connected (logically) to other beliefs in the subject’s belief system; fitting with the subject’s

experience would be necessary, but not sufficient, for justification. And it would still require

coherence, albeit coherence partially defined in terms of experience.

I agree on both points. I agree that there is nothing, qua coherentist, precluding the

coherentist from defining coherence not just in terms of relations between beliefs but also in

terms of relations between (a) beliefs and experiences and (b) experiences and experiences.3 I also agree that defining coherence in such a manner would safeguard the coherentist against the

Experience Objection. It would be a grave mistake, though, for the coherentist to so-define

coherence, as he himself would then be faced with the trilemma challenge (or a variant thereof).

The trilemma challenge—were it successful at all—would be telling not just against the claim that experiences, by themselves, can serve as reasons for, and hence justify, beliefs, but also against the claim that experiences together with beliefs can serve as reasons for, and hence justify, beliefs. The result, that is, would be not just that there is no foundationalist role for experience in justification, but also that there is no coherentist role for experience in justification.

The other alternative possible response to the Experience Objection is suggested, but not endorsed, by William Lycan, in reponse to an instance of the Experience Objection put forth by

Alvin Plantinga:

Case (4) is puzzling in a different way. We are asked to imagine a radical split between Ric’s beliefs and his perceptual experience. This is hard to do. Some philosophers argue for a conceptual connection between perceiving and the formation of perceptual beliefs; e.g., D. M. Armstrong maintains (as a piece of conceptual analysis) that perception is nothing but a certain means of acquiring a belief.28 Chez such a philosopher, Plantinga’s example is simply incoherent. I am not among those, but I still find the case hard to imagine in detail. (Lycan 1996, 16)

The strategy, then, would be to argue for a theory in the philosophy of mind that precludes the possibility of scenarios in which there is a radical conflict between belief and experience, thereby rebutting the claim that coherentism allows for scenarios in which there is both justification and a radical conflict between belief and experience.4

I am unsure of the plausibility of this line of response, since I am unsure both of which, if any, of the extant accounts in the philosophy of mind is correct, and of which, if any, such accounts preclude the possibility of scenarios in which there is a radical conflict between belief and experience. I shall thus simply acknowledge that this line of response might work (leaving it

to the experts in the philosophy of mind to determine whether it does), and stress that, either way,

it is not needed to rebut the Experience Objection.

4.2.1 Accounts on which the Target Beliefs are about the External World

The trilemma challenge has two levels. The first level is for accounts on which the target beliefs—viz., the beliefs for which experiences can serve as reasons—are about the external world. The second level is for accounts on which the target beliefs are about the internal world of experiences. S’s visual experience of a blue book before him, for example, would, on the latter such accounts, give him good reason not for thinking that there is a blue book before him, but for thinking that he is having a visual experience of a blue book before him. It will prove best to start with the first level.

Suppose, first, that experiences are non-propositional in content. The worry, then, would be that, since logical relations stand only between propositions (or things that are propositional in content), experiences are unable to stand in logical relations to beliefs and thus are unable to serve as reasons for beliefs. How else, the worry might be pressed, could experiences serve as reasons for beliefs? Let this be the Content Problem.

This worry is familiar from Davidson’s writings, specifically “A Coherence Theory of

Truth and Knowledge”:

The relation between a sensation and a belief cannot be logical, since sensations are not beliefs or other propositional attitudes. What then is the relation? The answer is, I think, obvious: the relation is causal. Sensations cause some beliefs and in this sense are the basis or ground of those beliefs. But a causal explanation of a belief does not show how or why the belief is justified. (Davidson 2000, 157)

The only difference with this expression of the worry is that it is being claimed, instead of being merely supposed for the sake of argument, that experiences are non-propositional in content.

It is important that—contrary to what Davidson’s claim that “its [i.e., coherentism’s]

partisan rejects as unintelligible the request for a ground or source of justification of another ilk”

might suggest—the coherentist, in pressing the trilemma challenge against the foundationalist,

not be read as claiming that nothing can count as a source of justification except a belief. Since

then the coherentist would be in no position (dialectically) to give the trilemma challenge—given

that (at least some) coherentists posit coherence, which is not a belief but a property of a belief system, as a source of justification.5,6 The coherentist should instead be read as claiming that nothing can serve as a reason, or as evidence, for a belief except a belief (or set of beliefs).

Suppose, next, that experiences are propositional in content, so that experiences can stand in logical relations to beliefs. Would that help the foundationalist? Would it follow that experiences, in virtue of being propositional in content, can serve as reasons for beliefs? It would seem not. For entertainings and hopings, to give just two examples, are propositional in content but cannot serve as reasons for beliefs. Imagine, for instance, that S is entertaining the thought that there are lots of dark clouds to the near west. Would it follow, just from this and the fact that there is a cogent inductive inferential relation between the claim that there are lots of dark clouds to the near west and the claim that it will rain shortly, that S has good reason for thinking that it will rain shortly? Surely not. But then the mere fact that experiences were propositional in content would not mean that experiences could serve as reasons.

Richard Heck makes a similar point against John McDowell, who argues at length for the claim that perceptual experiences are conceptual in content. Heck argues that since states such as mere entertainings are conceptual in content and yet are unable to serve as reasons for beliefs, showing that perceptual experiences are conceptual in content is insufficient, by itself, for showing that perceptual experiences can serve as reasons for beliefs:

All sorts of mental states have content: Desires and intentions, as well as beliefs; even mere considerings, entertainings of Thoughts. But none of these can be a reason for a belief: So, if a sense-datum theorist were to be converted to the view that perceptual experience has conceptual content, but held that it was like the entertaining of a Thought, she would not be much closer to a resolution of the epistemological problems about perception than she was before. (Heck 2000, 507)

The key for Heck, as I shall discuss later (much later), is not conceptual content but representational content—or “presentational” content, as he calls it.

The problem with states such as entertainings and hopings (vis-à-vis serving as reasons for beliefs), it seems, is that endorsement matters. S, in entertaining the thought that q, is not yet

taking a stand on how the world is, and so S, in entertaining the thought that q, is not yet in a position to take a stand on the claim that p. S, in hoping that q, is not yet taking a stand on how the world is, and so S, in hoping that q, is not yet in a position to take a stand on the claim that p.

And so on. The lesson, it seems, is that the difference between propositional states that can serve as reasons and propositional states that cannot serve as reasons is that the former, but not the latter, are endorsement states.

This, though, would seem to spell trouble for experiences, since experiences, it would

seem, are non-endorsement states. That, for example, S could visually experience a blue book

before him without endorsing the claim that there is a blue book before him (and vice versa), it

seems, is perfectly intelligible. The worry then would be that experiences are too much like

entertainings and hopings—and not enough like believings. How is it, to put the worry in terms

of a question, that non-endorsement states such as experiences, but not non-endorsement states

such as entertainings and hopings, can serve as reasons for beliefs? Let this be the Endorsement

Problem.

But perhaps experiences are endorsement states, so that, for instance, S’s visually

experiencing a blue book before him involves S’s endorsing the claim that there is a blue book

before him. Would this be of use to the foundationalist, in making sense of how experiences can

serve as reasons for beliefs? You might think so, since there is no problem in understanding how

endorsement states such as beliefs can serve as reasons for beliefs. Were experiences

endorsement states, then, presumably, experiences would serve as reasons for beliefs in the same

way that beliefs serve as reasons for beliefs. The problem, however, is that experiences, as

endorsement states, would be ill-suited to serve as regress-stoppers—since to justify, they

themselves would need to be justified, just as beliefs, to justify, themselves need to be justified.

Let this be the Terminus Problem.

The argument, to this point, is aimed just at accounts on which experiences serve as reasons for external-world beliefs. On the first horn, the challenge is to make sense of how

experiences—construed as non-propositional in content—could serve as reasons. The worry is that the only way to make sense of the relation x-is-a-reason-for-y is (at least partially) in terms of x’s standing in a logical relation to y. On the second horn, the challenge is to make sense of how

experiences—construed as non-endorsement states—could serve as reasons. The worry is that

the only way to account for the difference—qua ability to serve as reasons—between believings,

on one hand, and states such as entertainings and hopings, on the other hand, is in terms of

endorsement. And on the third horn, the challenge is to make sense of how experiences— construed as endorsement states—could serve as regress-stoppers. The worry is that endorsement

states, to justify, would themselves need to be justified.

4.2.2 Accounts on which the Target Beliefs are about Experiences

Things are a bit different with accounts on which the target beliefs are about experiences, in that, with such accounts, neither the Content Problem nor the Endorsement Problem nor the

Terminus Problem is immediately pressing. The reason why, in brief, is that, with such accounts, experiences serve as reasons not because of how their contents are related to the contents of the target beliefs, but because of how their contents are related to the subject—so that it is irrelevant whether the contents are propositional, whether, if propositional, the contents are endorsed, and whether, if endorsed, the contents are endorsed with justification.

The same is true of (at least some) accounts on which second-order beliefs are non- inferentially justified. Imagine, for instance, that Jim believes that he believes that there is a blue

Great Dane before him. Imagine also that though this second-order belief is true, the first-order belief (viz., Jim’s belief that there is a blue Great Dane before him) is unjustified. The accounts I have in mind are such that the unjustifiedness of the first-order belief would in no way preclude the second-order belief from being justified—given that, on the accounts I have in mind, whether the second-order belief is justified hinges not on whether there is an inferential relation of the

right kind between the first-order belief and the second-order belief, but on whether Jim is related in the right way to the first-order belief.

It is not the case, though, that such accounts are free of content worries, endorsement worries, and terminus worries. Quite the contrary. Each such account paints a justificatory picture in which there is a third component—viz., a component over and above the experience component and the belief component. With some, it is acquaintance. With others, it is awareness. With still others, it is attention. But what about this third component? Is it propositional in content? If not, how is it that it could figure in giving the subject reason for holding the target belief? If, on the other hand, it is propositional in content, is it an endorsement state? If not, what makes it different from non-endorsement states such as entertainings and hopings? If, in contrast, it is an endorsement state, is it justified? If not, what differentiates it from unjustified endorsement states such as unjustified beliefs? So, the Content Problem, the

Endorsement Problem, and the Terminus Problem—though now directed not at the experience component but at the acquaintance/awareness/attention component—are still very much in play.7

A similar but slightly different set of worries can be found in Sosa’s “Priveleged Access”.

He distinguishes between experiencing awareness (e-awareness) and noticing awareness (n- awareness): for you to be e-aware of an experience’s F-ness is for you to have an experience with

F; for you to be n-aware of an experience’s F-ness is for you to have a justified true belief to the effect that you have an experience with F. He then uses this distinction in attacking the view that if (a) S has an experience with F and (b) S is aware of the experience’s F-ness, then S’s belief that he has an experience with F is justified. If, on one hand, the awareness were construed as an e- awareness, the view would be open to counterexample. For though, for instance, you could have an experience with forty-eight speckles (or, better, an experience of something with forty-eight speckles) and thus be e-aware of the experience’s forty-eight-speckle-ness, your belief that you have an experience with forty-eight speckles—were you to have it as the result of guessing—

would not be justified. And if, on the other hand, the awareness were construed as an n- awareness, the view would fail explanatorily:

The latter [i.e., n-awareness] will not enable the desired explanation, since the concept of ‘noticing’ is itself epistemic in a way that unsuits it for the explanatory work that it is being asked to do. What we want is an explanation in non-epistemic terms of how a non- inferential, foundational belief can acquire epistemic status in the first place, so that holding it is not just arbitrary, so that conclusions drawn from it can inherit epistemic status. Our explanation hence cannot properly rest with ‘noticings’ that are supposed to have epistemic status already. The question will remain as to how these beliefs constitutive of the ‘noticings’ have acquired their status. (Sosa 2003, 277)

The alleged problem on the first horn (the one on which the awareness is construed as an e- awareness) is not the Endorsement Problem, but a much different problem: the problem of the speckled hen. But the alleged problem on the second horn (the one on which the awareness is construed as an n-awareness) is, in effect, just the Terminus Problem. The worry is that the

awareness would be incapable of providing for a terminus of justification, and that, thus, the

foundationalist’s proposed explicans would be incapable of providing for a terminus of

(epistemic) explanation.

I should note, before continuing on, that though Sellars is often credited with first articulating worries about whether experiences can serve as reasons for beliefs, Sellars’s attack on the given is very much different than the trilemma challenge. Yes, Sellars presses a dilemma against the proponent of the given. And yes, on one horn sensings are propositional while on the other horn sensings are non-propositional:

Now if we bear in mind that the point of the epistemological category of the given is, presumably, to explicate the idea that empirical knowledge rests on a ‘foundation’ of non-inferential knowledge of matter of fact, we may well experience a feeling of surprise on noting that according to sense-datum theorists, it is particulars that are sensed. For what is known, even in non-inferential knowledge, is facts rather than particulars, items of the form something’s being thus-and-so or something’s standing in a certain relation to something else. It would seem, then, that the sensing of sense contents cannot constitute knowledge, inferential or non-inferential; and if so, we may well ask, what light does the concept of a sense datum throw on the ‘foundations of empirical knowledge’? The sense- datum theorist, it would seem, must choose between saying: (a) It is particulars which are sensed. Sensing is not knowing. The existence of sense data does not logically imply the existence of knowledge, or (b) Sensing is a form of knowing. It is facts rather than particulars which are sensed.

On alternative (a) the fact that a sense content was sensed would be a non-epistemic fact about the sense content. . . . On the second alternative, (b), the sensing of sense contents would logically imply the existence of non-inferential knowledge for the simple reason that it would be this knowledge. But, once again, it would be facts rather than particulars which are sensed. (Sellars 1963, 128-9)

But otherwise, the worry is very much different:

For they have taken givenness to be a fact which presupposes no learning, no forming of associations, no setting up of stimulus-response connections. In short, they have tended to equate sensing sense contents with being conscious, as a person who has been hit on the head is not conscious, whereas a new-born babe, alive and kicking, is conscious. . . . But if a sense-datum philosopher takes the ability to sense sense contents to be unacquired, he is clearly precluded from offering an analysis of x senses red sense content s as x non-inferentially knows that s is red only if he is prepared to admit that the ability to have such non-inferential knowledge as that, for example, a red sense content is red, is itself unacquired. And this brings us face to face with the fact that most empirically minded philosophers are strongly inclined to think that all classificatory consciousness, all knowing that something is thus-and-so, or, in logicians’ jargon, all subsumption of particulars under universals, involves learning, concept formation, even the use of symbols. It is clear from the above analysis, therefore, that classical sense- datum theories—I emphasize the adjective, for there are other, ‘heterodox’, sense-datum theories to be taken into account—are confronted by an inconsistent triad made up of the following three propositions: A. X senses red sense content s entails x non-inferentially knows that s is red. B. The ability to sense sense contents is unacquired. C. The ability to know facts of the form x is φ is acquired. A and B togather entail not-C; B and C entail not-A; A and C entail not- B. (Ibid., 131-2)

The worry on the horn where sensings are propositional (because sensings are knowings) is not

that sensings would be ill-equipped to serve as regress-stoppers, but that the ability to sense

would be acquired—contrary to the position allegedly held by the proponent of the given. The

worry, in other words, is not that sensings would themselves need evidential support and thus

could not themselves stop any regress of knowledge or justification, but that sensings would

require prior learning and concept-acquisition, and so could not do the non-epistemic work that

the proponent of the given allegedly invokes them to do.

4.3 Accounts that Ignore the How-Question

The dialectic, to this point, runs as follows. The foundationalist argues that coherentism is false, on the grounds that it divorces justification from experience. The coherentist, at the same time, argues by way of the trilemma challenge that foundationalism (or at least the kinds of foundationalism on which experiences can serve as reasons for beliefs) is false, on the opposite grounds: namely, that it weds justification to experience. Each side has a point, and thus each side has work to do. The foundationalist needs to make sense of how experiences could serve as reasons, addressing the trilemma challenge head-on; the foundationalist needs to give not just an account on which experiences can serve as reasons, but also a rationale as to how experiences could serve as reasons. The coherentist, in turn, needs to make sense of how it is that, though false, it is so intuitive that experiences can serve as reasons.

You might expect Roderick Chisholm to be of help on this matter, as he is perhaps the twentieth-century’s most pre-eminent proponent of foundationalism. Just the opposite, unfortunately. Chisholm (at least in his writings) worries not about the question of how an experience could serve as a reason, but about the question of when an experience could serve as a reason.

Consider, for instance, Chisholm’s paper “A Version of Foundationalism”. He says explicitly and clearly that experiences can justify beliefs about experiences, so that, for instance,

S’s being appeared to redly can justify S’s belief that he is being appeared to redly. That experiences can do this is supposed to follow quite straightforwardly from the claim that experiences are self-presenting (which means that, for example, you would believe that you are being appeared to redly if you were, in fact, being appeared to redly and if you were, in fact, to consider whether you are being appeared to redly) together with the claim that beliefs about the self-presenting are justified. He never says, though, how experiences can justify beliefs about experiences (or, more generally, how the self-presenting can justify beliefs about the self-

100

presenting)—simply laying it down that, in fact, experiences can justify beliefs about experiences.

The same is true of Chisholm’s discussion, later in the paper, of whether experiences can justify beliefs about the external world. He says explicitly and clearly that experiences can justify beliefs of the form “There is something that is appearing W-ly to me.”:

S’s belief that there is something that is appearing W-ly to him is justified if (a) S is being appeared to W-ly and (b) S has no reason for doubting the claim that there is something that is appearing W-ly to him.

He also says explicitly and clearly that experiences can justify beliefs of the form “I am perceiving something that is F.”:

S’s belief that he is perceiving something that is F is justified if (a) S perceptually takes there to be something that is F (i.e., S is appeared to in such a way that he believes he is being appeared to in the way in question by something that is F) and (b) S has no reason for doubting the claim that he is perceiving an F.

But he nowhere explains how experiences can justify claims of either form, focusing entirely on the task of giving, in precise form, the conditions under which they can.

Susan Haack’s discussion in Evidence and Inquiry is no different in this respect. Haack, like Chisholm, develops and defends an account on which experiences can serve as reasons for beliefs.8 But Haack, like Chisholm, entirely ignores the how-question—giving no rationale as to how experiences, but not states such as entertainings and hopings, can serve as reasons for beliefs.9

You might argue, in defense of such foundationalists, that I am being a bit unfair, in asking for an answer to the trilemma challenge. Yes, each such foundationalist needs to give an argument in support of his version of foundationalism. And yes, each such foundationalist, therefore, needs to give an argument against the trilemma challenge—viz., an argument showing that, contra the trilemma challenge, experiences can serve as reasons for beliefs. But no, it is not the case that each such theorist needs to explain, in terms of the trilemma challenge, how

101

experiences can do this; showing that they can is enough. The question, after all, is not why experientialist foundationalism is true, but whether experientialist foundationalism is true.

You might also argue in the same way in defense of the foundationalists to be discussed

in sections 4, 5, and 6. That is, you might argue that even if I am right that they fail in answering

the trilemma challenge, they still have their arguments in support of their versions of

foundationalism, and thus still have their arguments in support of the claim that, contra the

trilemma challenge, experiences can serve as reasons for beliefs.

But that would be to miss the point of the trilemma challenge, or, better, to mislocate its

place in the overall dialectic. The trilemma challenge is supposed to cast doubt not just on

whether it can be explained how experiences can serve as reasons for beliefs, but also on whether,

in fact, experiences can serve as reasons for beliefs. The trilemma challenge is supposed to cast

doubt not just on whether it can be explained how experientialist foundationalism is true, but also

on whether, in fact, experientialist foundationalism is true. The foundationalist is giving an

explanatory theory—with the data consisting of both real and hypothetical cases. He is putting

forward his theory as true on the grounds that it best explains the data. The trilemma challenge is

supposed to cast doubt on this argument, by casting doubt on the conclusion, and perhaps even on

the premises—if the premises include contentious data cases. Hence, the foundationalist cannot

simply rest content with his explanatory argument in answer to the whether-question (i.e., the

question of whether experientialist foundationalism is true)—since without an answer to the how-

question, the explanatory argument in answer to the whether-question would fail.

Consider, by way of analogy, the situation with the substance dualist, who thinks that the

mind, though causally related to physical substances such as the human body, is a non-physical

substance. The charge that mental-physical causal relations would be entirely mysterious were

substance dualism true is supposed to cast doubt not just on whether it can be explained how

there can be causal relations between non-physical substances and physical substances, but also

on whether, in fact, there can be causal relations between non-physical substances and physical

102

substances. The charge is supposed to cast doubt not just on whether it can be explained how substance dualism is true, but also on whether, in fact, substance dualism is true. It would not do, then, for the substance dualist to answer the charge by arguing that substance dualism is true and that substance dualism entails that, somehow, there can be, and are, causal relations between non- physical substances and physical substances—given that without an answer to the how-question

(i.e., the question of how there can be causal relations between non-physical substances and physical substances), the argument in answer to the whether-question (i.e., the question of whether there can be causal relations between non-physical substances and physical substances) would fail.

“But what about the strong intuition that experiences can serve as reasons for beliefs?”, you might ask. Should not this intuition tip the balance in favor of the foundationalist? Not so, at least if what I shall argue for in chapters 6 and 7 is right. I shall argue that, though experiences cannot serve as reasons for beliefs, there is a contingent respect in which experiences are nonetheless required for justification—in that, though they cannot themselves serve as reasons, they can (and contingently do) help in enabling beliefs to serve as reasons.

4.4 Acquaintance Accounts

There are a number of foundationalist accounts that try to make sense of non-inferential justification in terms of acquaintance, or awareness, or some other such notion. The common thread running through these accounts is that what makes it such that S has good reason for thinking that p obtains—and thus what makes it such that S’s belief that p obtains is non- inferentially justified—is that S is acquainted with p, or aware, in some special way, of p.

The aim in this section is to canvass the leading theories in this camp, show that none of them adequately answers the trilemma challenge, and, in showing this, show that there is little, if any, promise of answering the trilemma challenge by way of acquaintance, or awareness, or some other such alleged special mode of access to the world. The basic problem, as you will see, is that

103

precious little can be said, in terms of positive elucidation, about notions such as acquaintance, so that, in the end, no appeal to notions such as acquaintance could help in making sense of non- inferential justification.

4.4.1 Chalmers

In “The Content and Epistemology of Phenomenal Belief”, David Chalmers argues for a rather peculiar variety of foundationalism—peculiar in that it is limited just to the domain of phenomenal beliefs, or beliefs about experiences. Some justified phenomenal beliefs, the view goes, are non-inferentially justified. The other justified phenomenal beliefs are inferentially justified, depending for their justification on the non-inferentially justified phenomenal beliefs.

The two key notions for Chalmers are the notion of a direct phenomenal belief and the notion of acquaintance. The view, overall, is that (other things being equal) a direct phenomenal belief is non-inferentially justified, and that what secures the justification is the subject’s acquaintance with the phenomenal quality that the phenomenal belief is about.

Chalmers distinguishes between three very different ways of picking out phenomenal

qualities, such as phenomenal redness. The first is relational, picking it out as the phenomenal

quality typically caused by paradigmatically red things.10 The second is indexical, picking it out via demonstrations such as “This quality.” or “This sort of experience.”—where, in fact, phenomenal redness happens to be the phenomenal quality being ostended. The third is direct (or non-relational), picking it out in terms of what it is like, subjectively, to have an experience of it.

It is this third sort of concept, which Chalmers calls “a pure phenomenal concept”, that figures in direct phenomenal beliefs.

But not just any old pure phenomenal concept can figure in a direct phenomenal belief,

according to Chalmers. Only some can. The key distinction is between direct pure phenomenal

concepts and standing pure phenomenal concepts. The distinctive mark of a direct pure

104

phenomenal concept is that it is partly constituted by the phenomenal quality of which it is a concept, so that having the concept requires having an experience with the quality. This is not the case with a standing pure phenomenal concept, in that you can have a standing pure phenomenal concept of a phenomenal quality long after the experience with the quality has left. Only a direct pure phenomenal concept, says Chalmers, can be deployed in a direct phenomenal belief.

The first point at which acquaintance enters the picture, for Chalmers, is with the explanation of why we can have direct concepts of our phenomenal qualities, but not of our non- phenomenal qualities, such as our height qualities, our age qualities, and our chemical qualities.

The explanation, he suggests, is that we are related to our phenomenal qualities, but not to our non-phenomenal qualities, in an especially intimate way. He calls this intimate relation

“acquaintance”.

The second, and (for my purposes) most interesting, point at which acquaintance enters the picture is with the justification of direct phenomenal beliefs. Chalmers argues that what secures the justification of a direct phenomenal belief is the subject’s acquaintance with the phenomenal quality that the belief is about, and that this works in a way analogous to the way in which a belief secures the justification of another belief. S’s belief b justifies S’s belief b’ only if

(1) the content of b’ is appropriately related to the content of b, (2) b’ is appropriately caused by b, and (3) b is justified. In the same way, S’s experience e justifies S’s direct phenomenal belief b only if (1) the content of b “mirrors” the quality of e, (2) b is appropriately constituted by e, and

(3) S is acquainted with the quality of e.

But what is it to be acquainted with a phenomenal quality? Is it a relation, as he suggests in some passages? Or is it not a relation but a mental state of some kind? Or is it something else entirely?

Unfortunately, Chalmers has little to say on this—by way of positive elucidation. First

and foremost, he says that acquaintance is a relation and that it explains both why we can have

direct concepts of our phenomenal qualities but not of our non-phenomenal qualities, and why

105

direct phenomenal beliefs are justified. Second, he says that acquaintance can be conceived of as either a relation between a subject and a phenomenal property, or as a relation between a subject and an experience with a phenomenal property:

Most fundamentally, it might be seen as a relation between a subject and an instance of a property: I am most directly acquainted with this instance of phenomenal greenness. This acquaintance with an instance can then be seen to confer a derivative relation to the property itself. Or in the experience-based framework, one might regard acquaintance as most fundmantally a relation between a subject and an experience, which confers a derivative relation between the subject and the phenomenal properties of the experience. (Chalmers 2003, 250)

Third, he says that acquaintance can be seen as an epistemic relation, and a basic one at that:

Acquaintance can be regarded as a basic sort of epistemic relation between a subject and a property. (Ibid., 250)

Fourth, he says that acquaintance is neither assertive nor conceptual nor justificatory:

Theoretically: acquaintance with a property makes the property available to a subject in a manner that makes concepts and assertions involving the property possible, and that enables these assertions to be justified. There is no reason why this requires acquaintance to itself involve an assertion. (Ibid., 252)

On my view, our acquaintance with qualities requires neither concepts nor language. (Ibid., 263)

Acquaintance is not itself a conceptual relation: rather, it makes certain sorts of concepts possible. And it is not itself a justificatory relation: rather, it makes certain sorts of justification possible. (Ibid., 251)

And fifth, he says that acquaintance can be thought of as just the instantiation of a phenomenal property, so that it is nothing over and above a subject’s having an experience with a phenomenal property:

It is also worth noting that one need not regard the acquaintance relation that a subject bears to a phenomenal property as something ontologically over and above the subject’s instantiation of the property, requiring a subject-relation-quality ontology at the fundamental level. It is arguable that it is a conceptual truth that to have a phenomenal quality is to be acquainted with it (at least in so far as we have a concept of acquaintance that is not wholly theoretical). Certainly it is hard to conceive of a scenario in which a phenomenal quality is instantiated but no one is acquainted with it. If so, then the picture I have sketched is combined with a simple subject-quality ontology, combined with this conceptual truth. The ontological ground of all this might lie in the nature of phenomenal qualities, rather than in some ontologically further relation. (Ibid., 254)

106

What we are told about acquaintance, then, is hardly anything—as all we are told is (1) that it might or might not be a relation, (2) that, if it is a relation, it stands between subjects and either phenomenal properties or experiences with phenomenal properties, (3) that (though perhaps epistemic and basic) it is non-assertive, non-conceptual, and non-justificatory, and (4) that it explains both why we can have direct concepts of phenomenal qualities but not of non- phenomenal qualities, and why direct phenomenal beliefs are justified.

Neither this obscurity as to what acquaintance comes to nor the resulting obscurity as to how acquaintance can secure justification is troubling to Chalmers, however. Chalmers thinks that acquaintance earns its keep by its explanatory role in an epistemic theory that accounts for the special epistemic character of the phenomenal domain—viz., the (alleged) fact that experiences can make direct phenomenal beliefs justified, as well as the (alleged) fact that direct phenomenal beliefs are immune to doubt.

I disagree. The trilemma challenge casts doubt, though not conclusive doubt, on the view that experiences can serve as reasons for, and thus justify, beliefs. The foundationalist, in light of this, needs to do more than just give an account of when experiences can serve as reasons for, and

thus justify, beliefs. He needs to give a rationale, explaining how experiences could do such

things. Chalmers fails miserably on this score, however. If, on one hand, acquaintance is nothing

over and above the instantiation of a phenomenal quality, then the view is just an instance of the

more general view according to which experiences can justify beliefs; there is the experience,

there is the belief about the experience, and, somehow, the experience makes the belief justified.

If, on the other hand, acquaintance is something over and above the instantiation of a phenomenal

property, then, since nothing is said as to what it comes to or as to how it makes phenomenal

beliefs justified, the view is just a fancy way of saying that experiences can justify beliefs about

experiences. The view is just that there is an X, call it “acquaintance”, such that X enables

experiences to justify beliefs about experiences.

107

4.4.2 Fumerton

The same is true of Richard Fumerton’s discussion in Metaepistemology and Skepticism.

He builds an account of non-inferential justification in which acquaintance plays a significant role. But he gives hardly anything by way of positive elucidation on the nature of acquaintance.

There are two parts to the account. There is a claim about when a belief would be non- inferentially justified. The claim is that S’s belief that p is non-inferentially justified if (a) S is acquainted with p, (b) S is acquainted with his thought that p, and (c) S is acquainted with the correspondence between p and his thought that p. And second, there is a claim about the kinds of things with which we are acquainted, and thus about the kinds of beliefs to which non- inferential justification can attach. The claim, in particular, is that we are acquainted with mental states (or the features thereof). The view, then, is that at least some beliefs about mental states are non-inferentially justified, and that what makes such beliefs justified is (a) the fact that the subject is acquainted with the mental state (or feature thereof) that the belief is about, (b) the fact that the subject is acquainted with his thought about the mental state (or feature thereof), and (c) the fact that the subject is acquainted with the correspondence between the state and the thought.

Here Fumerton seems to say something quite helpful, equating acquaintance with direct awareness:

Classical acquaintance theorists like Russell appropriately emphasized the role of acquaintance with particulars, properties, and even facts in grounding justification. But a fact is not a truth, and what one needs to end a regress of justification is a direct confrontation with truth. To secure that confrontation, one needs to be directly aware of not just a truth-maker (a fact to which a truth corresponds) but also a truth-bearer (a thought) and the correspondence that holds between them. (Fumerton 1995, 75)

This would be helpful, were he to mean it, since we already have somewhat of a purchase on what it is to be aware of something. Oftentimes, we use the word “aware” to mean “has a true

belief”, as when, because my wife has a true belief to the effect that Bush is the president of the

United States, I say that my wife is aware that Bush is the president of the United States.

108

Things are not so happy, though. For in other places he stresses that acquaintance is nothing like a true belief. To be acquainted with a fact, he stresses, is not to be in an intentional state, such as a true belief. It is not to know it, or to justifiably believe it, or to linguistically express it, or even to represent anything about it. Rather, it is to be related to it in a certain, hard to describe, way.

Fumerton, like Chalmers, is untroubled by this largely negative characterization. There is

just no getting around it, he argues. By way of description, the only (positive) things that can be

said are that acquaintance is a relation and that it stands between a subject and a fact. No other

relation is like it in any other way, and thus nothing further can be said.

“The solution to Sellars’s dilemma is to be found in the notion of acquaintance.”, avers

Fumerton. Hardly. Fumerton gives not a solution, but a dismissal. Fumerton simply dismisses

the worries involved in the trilemma challenge, saying (not arguing) that we are related to our

mental states in such a way that our beliefs about them (or at least some such) are non-

inferentially justified.

4.4.3 Fales

The foundationalist to turn to now is Evan Fales. For you might think that Fales’s A

Defense of the Given offers a way out for Chalmers and Fumerton.

Fales argues, by way of analogy to basic deductive inferences, that what enables an experience to make a belief justified is an inferential relation. The move from an experience to a belief about it is inferential, the idea goes, in that, like a basic deductive inference, there is a right way to do it and a wrong way to do it, in that, like a basic deductive inference, no believed proposition could be appealed to in defense of its (i.e., the move’s) legitimacy without already supposing the legitimacy of moves from experience to belief, and in that, like a basic deductive

109

inference, the ability to do it properly involves at least the potential grasp of the propositions that

(setting aside worries about begging the question) would suffice to defend its legitimacy.

Fales touts his “insight” on the inferential nature of experience-to-belief-about-it moves by noting that it is quite helpful in answering BonJour’s objection to views on which experiences can justify beliefs, an objection that is just a simpler version of the second level of the trilemma challenge. Fales’s theory, says Fales, is not a third-state theory, and thus is immune to BonJour’s objection, which is aimed at third-state theories. And it is more than just the claim that experiences can justify beliefs about experiences. Problem solved—or so he thinks.

But notice, there are some important differences between a deductive inference and an experience-to-belief-about-it move. First, there is the fact that with a deductive inference the input state (i.e., the premise belief) is a belief, whereas with an experience-to-belief-about-it move the input state (i.e., the experience) is not a belief; Fales stresses that an experience is not a belief. Second, there is the fact that with a deductive inference there is a logical relation between the content of the input state (i.e., the premise belief) and the content of the output state (i.e., the conclusion belief), whereas with an experience-to-belief-about-it move there is just a correspondence relation between the content of the input state (i.e., the experience) and the content of the output state (i.e., the belief about the experience); Fales says explicitly that an experience is non-propositional in content. What makes these differences important is that, it would seem, what makes it such that a deductive inference can yield justification is that the input state is a belief, and that the content of the input state entails the content of the output state—so that the subject has good reason for the output state if he has good reason for the input state. That is, the putative problem is that what makes it such that a deductive inference can yield justification is not what it has in common with an experience-to-belief-about-it move, but what it does not have in common with an experience-to-belief-about-it move. The onus is still on Fales, then, to make sense of how the ways in which an experience-to-belief-about-it move is inferential make it such that an experience-to-belief-about-it move can yield justification.

110

There are passages in which Fales is more forthcoming by way of positive elucidation as to how an experience-to-belief-about-it move is supposed to work, and in ways that seem to help in making it intelligible how an experience-to-belief-about-it move could yield justification. Here he says that it involves singling out the content, or at least part of the content, of the experience:

The way in which a basic belief is “inferred” from an experience E can be thought of, therefore, as involing two steps. First, a protopropositional constituent of the complex protopropositional content of E is singled out. Second, a thought is constructed in such a way that its propositional content is the “same” as (that is, maps onto) the protopropositional content of that constituent of E. (Fales 1996, 169)

And here he says that it involves recognizing that the proposition to be believed corresponds to the content, or at least part of the content, of the experience:

Having said all this, I nevertheless do not mean to rest too much weight on the term “inference,” as I use it to connect basic beliefs to experience. Indeed, it might be more precise to speak here of simple recognition of a kind of isomorphism between some feature or features of an experience, and a propositional content. To form a properly basic belief is to recognize the existence of a relation, which is just the truth-making relation of correspondence, between a state of affairs given in experience, and a primary judgment. We can think of this relation as involving a structural mapping plus a content mapping. (Ibid., 168)

It is quite intuitive, I fully grant, that with the recognition of correspondence the belief about the experience is justified, at least provisionally. What better reason could you have for a belief?

The problem, of course, is that with the recognition of correspondence the view now seems to be a third-state theory, so that it is imperative that Fales address the trilemma challenge head-on. Is the recognition propositional in content? If not, how is it that it could figure in giving the subject reason for holding the target belief? If, on the other hand, it is propositional in content, is it an endorsement state? If not, what makes it different from non-endorsement states such as entertainings and hopings? If, in contrast, it is an endorsement state, is it justified? If not, what differentiates it from unjustified endorsement states such as unjustified beliefs?

The answer, for Fales, is that the recognition neither is nor involves a third mental state of any sort, so that such questions are not to the point. Here, in summing up his theory, he speaks

111

not in terms of recognizing but in terms of apprehending, stressing that apprehending neither is nor involves a propositional state of any kind:

Here, the relation between a suitable fact and a basic belief is an act of apprehending, which is not itself, and does not contain as a constituent, any mental state analyzable as a propositional attitude. Apprehending is to be understood, rather, as a kind of (or as analogous to an) act of inferring. Such an act, together with the way in which it binds the belief that P to its object, is transparent to one, in the sense that when this total state of affairs occurs, it takes no further act on the part of the subject to apprehend its occurrence. (Ibid., 172)

And here he answers trilemma-type worries by saying, at least implicitly, that there is no third mental state involved in an experience-to-belief-about-it move, and that, thus, there is no third mental state with respect to which questions such as “Is it cognitive or non-cognitive?” and “Is it judgmental or non-judgmental?” can even be framed:

But what about the act of inferring a belief from an experience? Is this act (to use BonJour’s terminology) cognitive or noncognitive, judgmental or nonjudgmental? Does it have a propositional content? In response, we might say that it does—in a sense. We might say that the propositional content of a judgment corresponding to the act of accepting an “ordinary” inference is given by the conditional whose antecedent is its premise and whose consequent is its conclusion, and we might say that to judge a deductive inference valid is somehow equivalent to judging this conditional to be a tautology. But even in this case, it would be a grave mistake to identify the act of inferring itself (which we might nevertheless take to invole a species of judging) with the acceptance of that tautology: this we have learned from Achilles. In that sense, the act does not have propositional content.37 Nor, for the inferences from experience at issue here, can the conditional in question be formed—since the “antecedent” is not itself a proposition. Thus BonJour’s dilemma, which he takes to be decisive, has no force against the present way of understanding the warrant for basic beliefs. (Ibid., 167-8)

So though he speaks at times in terms of singling out, recognizing, and apprehending, and though

these terms suggest, quite straightforwardly, that there is a third mental state in the overall

justificatory picture, in reality there is just the experience, the belief about the experience, and the

transition from the former to the latter.

But now the problem is that, at the end of the day, nothing has been said to make it

intelligible how the belief about the experience is justified. The question is not of whether it is

intuitive that the belief about the experience is non-inferentially justified, and that the experience

itself plays a key role in the justification. Even the coherentist grants that it is intuitive—at least

112

pre-philosophically. The question, rather, is of whether there is a rationale for thinking that the belief about the experience is non-inferentially justified, and that the experience itself plays a key role in the justification. Fales, in the end, is unhelpful on this score.

Here, however, is where you might think that Fales has a leg up on Chalmers and

Fumerton. For Fales at least gives independent reasons for the existence of the non-judgmental recognitions and apprehensions he has in mind. Fales draws on Lewis Carroll’s “What the

Tortoise Said to Achilles”, arguing that justification-yielding deductive inferences involve the very same non-judgmental recognitions and apprehensions that, if he is right, are involved in experience-to-belief-about-it “inferences”.

The problem, though, is that Fales’s argument for this is premised on internalism—viz., access-internalism. Fales is explicit in saying that he is an internalist, and that internalism puts special constraints on inferential justification:

If foundationalism is to carry out its program of articulating the structure of our knowledge of the world, it must hold that the domain of the given covers more than the ground of primary perceptual beliefs. In particular, it must be the case that the validity of certain rules of inference is given, and possibly also certain other a priori principles. At least, this must be the case if foundationalism is to respect an internalist conception of knowledge, a conception that requires that if x knows that p then the basis for x’s knowledge that p is available to x as part of his or her subjective conception of the grounds for p. If p is a derivative empirical belief, then it can count as part of what x knows only if x knows that p can be properly derived from other knowledge that x has— hence, only if x understandingly employs the relevant rules of inference, that is, in a way that involves grasping their legitimacy. Since rules of inference can only be demonstrated by being built up out of elementary rules of inference, the legitimacy of some rules of inference must be given. (Ibid., 14)

It is these constraints that, when coupled with the lessons from Carroll, lead to Fales’s claim that justification-yielding deductive inferences involve the very same non-judgmental recognitions and apprehensions that, if he is right, are involved in experience-to-belief-about-it inferences.

This is problematic because access-internalism, which differs greatly from mind-internalism, is itself problematic, and because, moreover, access-internalism is not needed to differentiate

between inferences that yield justification and inferences that do not yield justification.11

113

4.4.4 Peacocke

Christopher Peacocke, in “Conscious Attitudes, Attention, and Self-Knowledge”, argues that S’s belief that he has mental state m is (non-inferentially) justified if (1) m is conscious and

(2) S’s belief that he has m is based on m. The sense in which m would justify, says Peacocke, is that m would serve as a reason—making it such S has good reason for thinking that he has m.

Peacocke’s theory is perfect at this point in the progression, in that it builds on, or differs from,

Fales-type theories in ways that might seem promising for the foundationalist.

Peacocke stresses that he is not imagining a scenario in which you infer from or perceive a conscious mental state. So though, for instance, S’s experience of pain might give him good reason for thinking that he is experiencing pain, it would not do so, as Peacocke is imagining things, in virtue of any inference from or perception of the experience itself.

Perhaps, then, what matters is the content, the content both of the conscious mental state

and of the appropriately-related self-ascription. Not so, according to Peacocke. There is a failure

of transitivity in cases of consciously-based self-ascription, in that what serves as a reason for the

conscious mental state (when the conscious mental state is a belief) need not also serve as a

reason for the self-ascription. For example, though S’s apparent memory that Dubcek was the

prime minister of Czechoslovakia when the Soviet Union invaded gives him reason for believing

that Dubcek was the prime minister then, and though this, in turn, gives him reason for believing

that he holds the belief that Dubcek was the prime minister then, his apparent memory that

Dubcek was the prime minister then does not give him reason for believing that he holds the

belief that Dubcek was the prime minister then. And the reason for this (failure of transitivity) is

that what enables a conscious mental state to serve as a reason for an appropriately related self-

ascription is in part the conscious state itself (or, better, the nature thereof).

Peacocke argues at length that there is a tight, though non-constitutive, connection between S’s believing that p and S’s believing that he believes that p. It is this sort of connection,

114

you might think, that, as Peacocke sees things, enables a concscious mental state to serve as a reason for an appropriately-related self-ascription.12,13 I doubt it. First, he never says so.

Second, he starts the discussion (of the tight connection) with a long paragraph about the intrinsic interest of the question “Why is it that there is a tight, though non-constitutive, connection between S’s believing that p and S’s believing that he believes that p?”. And third, he says that his conscious-states-as-reasons view is helpful in answering this question about the tight connection between second-order beliefs and first-order beliefs—thus saying, by implication, that his view on the tight connection should not be understood as rounding out, or as helping to further explicate, his view on conscious-states-as-reasons.

The natural suggestion, at this point, is that the key for Peacocke is consciousness, so that the view, all told, is that it is in virtue of the consciousness of a conscious mental state that a conscious mental state can serve as a reason for an appropriately-related self-ascription. What else could it be, given that it is something to do with the state itself (over and above the state’s content), and given that it is not the fact that the process to which the state is the input (and of which the self-ascription is the output) is highly reliable?

Peacocke never explicitly says that the key for him is consciousness, and thus never explicitly says how the consciousness of a conscious mental state is supposed to enable a conscious mental state to serve as a reason for an appropriately-related self-ascription. It is as though he thinks it follows straightforwardly from the nature of a conscious mental state, and so from the nature of a consciously-based self-ascription, that the consciousness of a conscious mental state enables a conscious mental state to serve as a reason for an appropriately-related self-ascription, so that no clarification is called for.

There are two particularly pertinent passages, one coming before the justification thesis

(i.e., the thesis that S’s belief that he has mental state m is justified if (1) m is conscious and (2)

S’s belief that he has m is based on m) and one coming after. In the passage coming before, he says that though conscious thoughts (e.g., conscious beliefs) differ from experiences and

115

sensations (which are conscious states) in not having objects of attention, conscious thoughts nonetheless involve the occupation of attention:

If we are to describe correctly the relation between conscious thought and attention, we must respect the distinction between the object of attention and what is occupying attention. In a normal case of perceptual attention to some physical object, feature, or event, there is something to which the subject is attending. The object of attention is perceived; it causally affects the subject. No doubt we may want to say that there is, or can be, some sort of object of attention in a pure case of perceptual hallucination. But such cases are plausibly understood as parasitic on the central case of genuine perception. In the case of pure hallucination, it is for the subject as if there were a genuine object of attention. In conscious thought, by contrast, there is no object of attention (nor is it as if there is). The notion of an object of attention which is inapplicable in conscious thought is that of an experienced object, event, or state of affairs. In mental states other than those of conscious thought, a genuine object of one’s attention might be a material object, or a continuing event, or the continuing or changing features of an object or event, or an object’s changing relations to other objects or events. Having a sensation is also an experience. A pain, for instance, can equally be an object of attention. But thinking is not experiencing. There are objects of thought, but an object of thought is not thereby an experienced object, and is not an object of attention in the sense in question. All the same, in conscious thought, your attention is still occupied—as it is also occupied in the perceptual cases and in cases of imagination. It would be a crude non sequitur to move from the true point that there is no object of attention in conscious thought to the false conclusion that conscious thought does not involve attention. (Peacocke 1998, 65)

In the passage coming after (the justification thesis), he says that consciously-based self- ascriptions, like inferentially-based beliefs and experientially-based beliefs, involve a sensitivity both to the content of the basing state and to the basing state itself:

The idea of consciously based self-ascriptions is sometimes regarded with great suspicion. In fact, in respect of the rational sensitivities required for consciously based self-ascription to proceed properly, these ascriptions are importantly similar to other, very different cases. Consider for a moment beliefs which are reached by inference. When a belief is reached by inference from certain premisses, the contents of some of the thinker’s states are taken by the thinker to support the inferred conclusion, and they do so in the case of valid inference. Now the thinker who successfully reaches new beliefs by inference has to be sensitive not only to the contents of his initial beliefs. He has also to be sensitive to the fact that his initial states are beliefs. He will not be forming beliefs by inference from the contents of his desires, hopes, or daydreams. Another pertinent case is that of beliefs reached by endorsing the content of one of the thinker’s perceptual experiences. Here too the thinker makes a transition—and this time not an inferential transition—from one state with a certain content to a belief with an overlapping, or an appropriately related, content (depending on your views about the nature of perceptual content). Again, the sensitivity does not involve merely some grasp of relations of content between the two states involved in the transition. The thinker is

116

also sensitive to which kind of intitial state it is that has the content. He will not be prepared to take the content of imaginings, for instance, at face value in the same way. In cases of consciously based self-ascription of attitudes and experiences, a thinker similarly makes a transition not only from the content of some initial state, but also because the initial state is of a certain kind. There is, though, a difference from the cases of inference and perception mentioned in the last paragraph. In the cases of consciously based self-ascription, the distinction between those events which are occurrent attitudes of the right kind to sustain the resulting judgement and those which are not is a distinction which is (partially at least) conceptualized by the thinker. The self-ascriber thinks of his state as belief, or as experience of a certain kind, or whatever it may be. He also thinks of himself as the state’s subject. Possession of these important conceptual capacities does, of course, go far beyond the ability to make judgements rationally in response to one’s own conscious states. (Ibid., 73)

Perhaps the idea is that what secures the sensitivity involved in a consciously-based self- ascription is the attention involved in the consciousness of the conscious mental state. That is, perhaps the view is that S’s sensitivity both to the content of m and to m itself is due to, or just is, the attention involved in the consciousness of m. It would thus make sense, you might think, that

S’s conscious mental state m can serve as a reason for his consciously-based self-ascription that he has m, for the consciousness of m would involve S’s attending both to the content of m and to the nature of m itself.

Be clear, though, that were this the right way to read Peacocke, the attending in question would not be the kind of awareness involved in veridical commitment, or endorsement, states

(such as the awareness involved in my wife’s true belief that Bush is the president). Peacocke distinguishes between conscious states that are reason-led (such as beliefs) and conscious states that are not reason-led (such as experiences and sensations), saying that the former are individuated partly in terms of what states would be reasons for being in them and what states they would be reasons for being in, and that being in a reason-led state involves being committed to something. As neither experiences nor sensations are reason-led states and thus as neither experiences nor sensations are commitment, or endorsement, states, the attending involved in the consciousness of such states could not itself be the kind of awareness involved in veridical commitment, or endorsement, states.14 It would still be obscure, therefore, how a conscious

mental state could serve as a reason for an appropriately-related self-ascription.

117

It should also be noted, against the reading under consideration, that what is occupying

S’s attention when he is having a conscious mental state, according to Peacocke, is the content of the state, not the state itself. This is clearest with experiences and sensations, where the things attended to are things such as material objects, continuing events, the continuing features of objects, and so on. It also holds with conscious thoughts, though:

If what I have said about the occupation of attention by conscious thought is along the right lines, then the occupation of attention at least in directed thought also performs a function of selection. One can expect that the parallel will not be precise, just because of the difference noted between the presence of objects of attention in the perceptual and sensational cases and their absence in the case of conscious thought. None the less, when a thinker is engaged in directed thinking, he is in effect selecting a certain kind of path through the space of possible thoughts—thought contents—available to him. (Ibid., 70)

The “things” attended to, in directed thinking, are thought contents, not thoughts themselves.

Hence Peacocke should not be read as saying that the sensitivity involved in consciously-based

self-ascriptions is to be understood in terms of the attention involved in the consciousness of

conscious mental states—since the sensitivity is of, while the attention is not occupied by, the

conscious mental states themselves.15

So perhaps Peacocke should instead be read as saying that the sensitivity is to be understood in terms of sub-personal causal mechanisms (as Peacocke himself might put it). The view would be that S’s underlying psychological make-up is such that S’s having the conscious belief that p starts a causal process ending in S’s holding the belief that he believes that p (as opposed to ending with S’s holding the belief that he believes that q, or S’s holding the belief that he is entertaining p), and S’s having an experience of x starts a causal process ending in S’s holding the belief that he is experiencing x (as opposed to ending with S’s holding the belief that he is experiencing y, or S’s holding the belief that he hopes that p), and so on. It could still be held that it is the consciousness of a conscious mental state that makes it such that a conscious mental state can serve as a reason.

Then, though, we would be without what we were after in the first place. What we were

after was a rationale, a story as to how a conscious mental state could serve as a reason for an

118

appropriately-related self-ascription. What we would be left with, however, is just the claim that a conscious mental state can serve as a reason in virtue of the fact that the consciousness of a conscious mental state involves the subject’s attending to the content of the state—with no hint as to how in attending to x the subject has reason for thinking that he is having (say) an experience of x.

It would not help to respond with the claim that it is the sensitivity, not the consciousness, that explains how a conscious mental state can serve as a reason. The non-foundationalist can readily allow for the possibility of, and even the actuality of, a cognizer the consciously-based self-ascriptions of which are causally sensitive to, via sub-personal causal mechanisms, his conscious mental states. The question would then be, “But how is it that this sensitivity enables the conscious mental states to give him reason for assenting to the contents of the consciously- based self-ascriptions?”.

4.4.5 BonJour

BonJour, interestingly, has come to reject the brand of coherentism that he defends in The

Structure of Empirical Knowledge.16 But even more interestingly, BonJour has moved not to a

different kind of coherentism (or to skepticism), but to a rather old-fashioned kind of

foundationalism—a kind on which the foundation of justification is comprised of beliefs about

beliefs and beliefs about experiences, and on which, like with Peacocke’s account, the key to non-

inferential justification is the consciousness of conscious mental states. I say “even more

interestingly” since, as I noted in section 2, BonJour has spent much time in criticism of such

foundationalist views.

BonJour starts with second-order beliefs. He says, or suggests, that S’s conscious belief that p involves a built-in awareness both of p and of his (i.e., S’s) attitude of acceptance toward p,

119

stressing that the awareness, though, is neither a second-order awareness to the effect that he holds the belief that p, nor even a mental state separate from the belief itself:

My suggestion is instead that an essential and intrinsic aspect of having an occurrent belief just is being consciously aware of the two correlative aspects of its content: first, its propositional content, in this case the proposition that foundationalism is much more defensible than most philosophers think; and, second, the assertory rather than, e.g., questioning or doubting character of one’s entertaining of that content. These two awarenesses (or rather, more plausibly, two aspects of one awareness) are, I am suggesting, not in any way apperceptive or reflective in character: they do not require or involve a distinct second-order mental act with the propositional content that I have the belief in question. Instead, they are (at least partly) constitutive of the first-level state of occurrent belief or thought itself in that they are what make it the very occurrent belief that it is, rather than some other occurrent belief or a different sort of state altogether. (BonJour and Sosa 2003, 62)

Then he argues that the built-in awareness, unlike the belief itself, is not in need of justification, and that, moreover, the built-in awareness can serve as a reason for, and hence justify, the second- order belief that he, S, holds the belief that p.

BonJour then moves to experiential beliefs, modeling the story there on the story about second-order beliefs. He says that S’s sensory experience with sensory content c is a conscious state and thus involves an awareness of c—an awareness, again, that is neither a second-order awareness to the effect that he has an experience with c, nor even a mental state separate from the experience itself. He next says that the awareness is not in need of justification, for the same reason that the awareness involved in a conscious first-order belief is not in need of justification.

And, last, he says that the awareness can serve as a reason for, and thereby justify (non- inferentially), S’s experiential belief that he, S, has an experience with c.

But how is it, you might wonder, that the built-in awareness that BonJour has in mind can serve as a reason for a belief? It cannot do so in the way that most readily comes to mind— namely, by standing in a logical relation to the target belief. BonJour is quite explicit in saying that with a conscious experience the built-in awareness is not just not second-order, but also not conceptual (and hence not propositional). How is it, then, that the built-in awareness, whether in the belief case or in the experience case, is supposed to do it?17

120

The relation, argues BonJour, is not logical, but descriptive. Consider, to illustrate, the case where S has an experience with sensory content c, and where, in addition, he believes that he has an experience with sensory content c. The belief, on one hand, is supposed to accurately describe the experience’s sensory content, so that there is supposed to be a descriptive relation between the experience and the belief. The experience’s built-in awareness, on the other hand, is of c—the very sensory content that the belief is supposed to describe. And hence S, in virtue of having the awareness, has quite the reason for thinking that he has an experience with c:

But the important point for our purposes is that where such a relation of description exists, the character of the non-conceptual object is what determines whether the conceptual description is correct or true. And thus an awareness of that non-conceptual character can seemingly constitute a kind of reason for thinking that the description is true or correct (or equally, of course, untrue or incorrect)—thus apparently providing a basis for the justification of the conceptual claim. (Ibid., 72)

The claim is that the built-in awareness (which is not only not second-order but also not conceptual and thus not propositional) can serve as a reason for the belief since the awareness is an awareness of what it is that makes the belief true (when it is true).

To sum up, the view (to focus just on experiences and experiential beliefs) is that (a) a sensory experience involves a built-in awareness of its sensory content, (b) the built-in awareness can serve as a reason for, and thus justify, an appropriately-related experiential belief, and (c) what enables the awareness to serve as a reason for the belief is that it is an awareness of what it is that makes the belief true. Let this view be BJ1.

The problem with BJ1—or, better, a problem with BJ1—is that the rationale is at odds with the metaphysics. The rationale is that the built-in awareness is an awareness of what it is that makes the belief true (when it is true). The metaphysics, though, is such that the built-in awareness is an awareness of just part of what it is that makes the belief true. Here BonJour seems to say, by implication, that the built-in awareness is not of the subject:

The crucial point is simply that an occurrent belief is, after all, a conscious state, and that what one is primarily conscious of in having such a belief is precisely its propositional content (together with the fact that this content is being accepted rather than, say, doubted or wondered about. (BonJour 2002, 212, emphasis mine)

121

And here BonJour seems to say that the built-in awareness is not even of the state:

My diagnosis is that Rosenthal (along with many others including, perhaps, even Descartes himself) has confused two subtly but crucially different things: first, the consciousness of the content of a conscious mental state, which is, I have suggested, intrinsic to the occurrence of that state itself; and, second, the reflective or apperceptive consciousness of that state itself, i.e., the consciousness that such a state has occurred, which I agree requires a second-level or apperceptive state. This confusion is plainly reflected in Rosenthal’s statement that “conscious states are simply mental states we are conscious of being in” (p.462): to be conscious of being in a state requires consciousness of the state itself as a kind of object of thought and not merely a consciousness of its content. Conflating these two things thus leads quite inevitably to the view that a mental state could be intrinsically conscious only by somehow, paradoxically, having both its ordinary content and the further, self-referential content that it itself occurs. Rosenthal is surely right to reject such a view (pp. 469-70), but wrong that it is the only alternative to his higher-order thought theory. (BonJour 2003, 68, emphasis mine)

The problem is this. What it is that makes the target belief true, in both the belief case and the experience case, has three components:

(1) the subject (2) the state with the content (3) the having relation (the relata of which are the subject and the state)

But what it is that the built-in awareness is an awareness of, in both cases, is just the content (of the state) and, in the belief case, the assenting to the content. The awareness involves neither an awareness of the subject nor an awareness of the state nor an awareness of the having relation.18,19

You might argue, pointing to passages such as the one below, that I am misreading

BonJour:

Instead, I suggest, to have a conscious occurrent belief just is, in part, to have a conscious awareness of the content of that belief (and also of one’s accepting attitude toward that content), an awareness that is not reflective (or “second-order”) in nature, but is instead partly constitutive of the first-level occurrent belief state itself. (BonJour 2002, 211, emphasis mine)

The claim in passages such as this, you might think, is that, in being aware of his attitude of assent, S is aware of himself—thus undermining the premise, in the objection above, according to which the built-in awareness is not of the subject. So too, then, with the built-in awareness involved in a sensory experience. Let this view, whether it be BonJour’s or not, be BJ2.

122

This is certainly a coherent reading, as claims having the form “S is aware of T’s n.”

(where “n” is a noun) and claims having the form “S is aware of T’s v-ing.” (where “v” is a verb) have a coherent reading on which the claim, in part, is that S is aware of T. The claim “Jim is aware of Spot’s running.”, for example, can perfectly well be read as involving the claim that Jim is aware of Spot; suppose that Jim is watching Spot run after a tennis ball.

That the alternative reading is coherent, however, in no way entails that it is true, since claims having the form “S is aware of T’s n.” and claims having the form “S is aware of T’s v- ing.” have a weaker reading. With claims of the first form, the claim is that S is aware of n— where it just so happens that n is T’s. With claims of the second form, the claim is that S is aware of v-ing—where it just so happens that the v-ing is T’s. Suppose, for instance, that Jim is looking at what happens to be Jack’s ten-dollar bill, and that Jim is thus aware of what happens to be

Jack’s ten-dollar bill. Suppose further that Jim is entirely unaware of Jack, having no idea whatsoever who Jack is. The claim “Jim is aware of Jack’s ten-dollar bill.”, then, should be read weakly. Or suppose that Jim hears someone in an adjacent apartment say that the cat is on the mat, and that he has no idea who the someone in question is. Suppose that, in fact, it is George.

The claim “Jim is aware of George’s saying that the cat is on the mat.”, though perfectly intelligible, would thus demand the weaker of the two possible readings.

Notice also that there are times at which BonJour gives claims of the form “S is aware of x’s n.” where it is clear that the weaker of the two possible readings is correct:

The main distinction is between (a) a belief that is about another, distinct belief (and thus reflective or “second-order”) and (b) the conscious awareness of a belief’s own content that is, I am claiming, a constitutive or intrinsic feature of any conscious, occurrent belief, without the need for a second, independent belief. (Ibid., 212, emphasis mine)

That the weaker reading is correct follows from the fact that, in contrasting his account of consciousness with Rosenthal’s account, BonJour stresses that the built-in awareness is not of the state itself. Hence, there is precedent for the reading according to which BonJour, in saying that

“to have a conscious occurrent belief just is, in part, to have a conscious awareness of the content

123

of that belief (and also of one’s accepting attitude toward that content)”, is not saying that the built-in awareness is of the subject himself.

But what if the stronger reading were correct? Or what if, though the weaker reading is correct, BonJour were to amend the account so that the built-in awareness is, in part, of the

subject? Would either possibility be of help to the foundationalist?

I think not. What makes the target belief true, in both the belief case and the experience case, has three components, one of which is the state with the content. However, the built-in

awareness—even if it were, in part, of the subject himself—would still not be of the state with the

content. Think back to BonJour’s discussion of Rosenthal’s higher-order account of

consciousness, in which BonJour stresses that the built-in awareness is of just the content together with the attitude of assent—but not the belief itself. The rationale would still be at odds with the metaphysics.

I read the passage quoted above about Rosenthal so that the claim, in part, is that the built-in awareness is not of the belief, but just of the propositional content together with the attitude of assent. But there is another way of reading the passage, a way on which the claim is not that the awareness is not of the belief, but that the awareness is not a judgmental state about the belief. It is quite intuitive, the proponent of this alternative reading might further suggest, that believing that p is nothing over and above assenting to p, so that it is quite plausible that BonJour, in saying that S is aware of his assenting to p, is saying that S is aware of his believing that p. Let this view (which involves an analogous claim on the experience side) be BJ3.

The nice thing about BJ3, in contrast to BJ1 and BJ2, is that the metaphysics is in line with the rationale. On the second-order belief side, the built-in awareness is of p, p’s being assented to, and p’s being assented to by S. On the experiential-belief side, the built-in awareness is of c, c’s being experienced, and c’s being experienced by S.

The problem now, though, is a problem that has been with us from the very beginning.

The rationale is quite appealing on the surface, since “aware” is often used in the sense of “truly

124

believes” and “perceives”. For BonJour, though, “aware” is not to be read in either such sense.

But then how should it be read? Without an answer, we would be without a rationale. The view

(focusing just on the experience side) would simply be that there is an X, call it “a built-in awareness”, such that X enables experiences to serve as reasons for beliefs about experiences.

BonJour, unfortunately, never says.

But perhaps there is more to the view. Here, for instance, is a passage in which BonJour seems to say that there is a second awareness involved, so that the built-in awareness is not itself doing all of the justificatory work:

Such a built-in awareness of sensory content is thus also apparently available to justify genuinely foundational beliefs in a way parrallel to the case already discussed. Thus if I have the belief that, e.g., my present visual experience includes a red, square patch in the approximate middle of my visual field, and if my constitutive or built-in awareness of the content of my actual conscious visual experience includes an awareness of such an element, then this latter awareness seems to provide a clear and compelling reason for thinking that the belief is true. And if I am aware of this reason, as of course I might not be, then, at least as long as other things are equal, the belief is seemingly justified in a way that does not rely on any further belief, thus making it basic in the sense required by foundationalism. (Ibid., 70, emphasis mine)

There is, first, the built-in awareness, which, according to BonJour, can serve as a reason—and one that is clear and compelling—for the belief. And there is, second, an awareness of the built- in awareness.20 And, further, here is a passage in which BonJour seems to say that there is a comparison involved:

Thus when I have a conscious state of sensory experience, I am, as already argued, aware of the specific content of that state simply by virtue of having that experience. And hence if an apperceptive belief that I entertain purports to describe or conceptually characterize that perceptual content, albeit no doubt incompletely and abstractly, and if I understand the descriptive content of that belief, i.e., understand what an experience would have to be like in order to satisfy the conceptual description, then I seem to be in a good, indeed an ideal, position to judge directly whether the conceptual description is accurate as far as it goes, and if so, to be thereby justified in accepting the belief. (Ibid., 73, emphasis mine)

The subject, it seems, needs to see, via a comparison of some sort, that the conceptual description

is accurate; otherwise, the experiential belief would not be justified (let alone non-inferentially

justified). Let the first view suggested be BJ4, and the second view suggested be BJ5.

125

It is clear that BJ4 would be of no help. It would simply invite the question “But in what does the second-order awareness consist?”, so that the problems at the first level would also arise at the second level. At the first level, there would still be the question of what S’s built-in awareness of c (i.e., the experience’s sensory content) amounts to. And at the second level, there would be a new question—viz., the question of what S’s awareness of the built-in awareness at the first level amounts to.

It is also clear that BJ5 should not be read as the claim that (in order for the built-in

awareness to serve as a reason) the subject needs to compare (in the judging sense) the

experiential belief with the experience and hence needs to be aware (in the judging sense) of the

fit between the experiential belief and the experience—since were it read in such a manner, it

would fall prey to the Terminus Problem. After all, the judgement involved in the comparison, to

justify, would itself need to be justified.

The clue to understanding BJ5—and, it turns out, to seeing how BJ3, BJ4, and BJ5 fit together—is to be found in BonJour’s response to the worry that the subject would need to make a judgmental comparison of fit between the belief and what the belief is about:

But don’t I still have to judge that the content reflected in the “built-in” awareness and that claimed in the second-order belief are the same, and doesn’t this higher-order judgment itself require some independent justification, thus continuing the regress? Internalist epistemologists are sometimes charged with “overintellectualizing” the situation of justification, an accusation that generally seems to me to have little force, being based as it is on little more than the internalist demand that justification involve a genuine and intelligible reason for thinking that the belief to be justified is true. Here, however, I am inclined to regard the suggestion that a direct comparison of two conscious states need involve an independent judgment that must in turn be justified by something other than the conscious contents of the states themselves as a clear case of objectionable overintellectualization. If any intellectual comparison or assessment can ever be direct and unmediated by a further judgment, surely this one can. And to deny that this is ever possible is to guarantee vicious regresses in all directions, rendering the operation of the intellect inherently futile. (Ibid., 65)

The response is not that no comparison is needed. Nor is it that though a comparison is needed, it is not needed that the comparison be justified. Rather, it is that though a comparison is needed, it is not needed that the comparison involve, or consist in, a higher-order state—a state about the fit

126

between the first-order belief and the second-order belief. The subject, the response goes, can make the comparison directly and without mediation.

The sense in which, at least for BonJour, BJ3, BJ4, and BJ5 are not at odds with each other is that, at least for BonJour, they involve the same metaphysics. The extra awareness in BJ4 and the comparison in BJ5 (a comparison which involves the extra awareness in BJ4) are comprised of nothing over and above the built-in awareness in BJ3 (with the exception, perhaps, of the built-in awareness in the target belief).

The problem with BJ3, though, would then plague BJ4 and BJ5. The key question of how the notion of a built-in awareness is supposed to be understood would remain unanswered.

Yes, we would still be told that it is not a belief, or an endorsement state of any kind. Yes, we would still be told that it is non-conceptual, in the experience case. And yes, we would still be told that it is partly constitutive of the experience in the experience case, and of the first-order belief in the first-order belief case. But what else is it? And how is it supposed to help in understanding how experiences can serve as reasons? These questions would still be unanswered.

I can offer only a conjecture, in answer to the question of what it is. It is not a thing or

state at all—BonJour’s comments suggesting directedness notwithstanding.21 Rather, it is a way

of assenting to a proposition—in the belief case. It is a way that is very much different, though in

respects hard to describe, from the way in which a proposition is assented to in the case of a non-

conscious belief. It is a way that contributes to what it is like, subjectively speaking, for the

subject. The same is true in the experience case. It (i.e., the built-in awareness) is a way of

entertaining, or being presented with, a sensory content—viz., a way that contributes to what it is

like for the subject.

This, if correct, would help in understanding what the notion of a built-in awareness comes to. But, of course, it would not help in understanding how experiences, to take just the experience case, can serve as reasons. For it just is the claim that experiences can serve as reasons.

127

4.4.6 Brewer

Bill Brewer is yet another proponent of the thesis that experiences can serve as reasons for beliefs. But unlike the other theorists so far discussed in this section, Brewer thinks that experiences can serve as reasons for external-world beliefs. And unlike Peacocke and BonJour, the key for Brewer is not the consciousness involved in the having of an experience. The key, instead, is the understanding involved in the grasping of the content of an experience. It is this understanding that enables an experience to serve as a reason.

Here is the theory, in a nutshell. First, there is the claim that perceptual experiences are conceptual in content, and that their contents have the form “That is thus.”—where “That” picks out an object in the external world, and where “thus” picks out a property of the object). Second, there is the claim that the contents of the beliefs for which perceptual experiences serve as reasons also take the form “That is thus.”—where “That” and “thus” pick out the object and the property referred to by the corresponding experiences. And third, there is the claim that what enables the experiences to serve as reasons for the beliefs is that the subject, in grasping the contents in question, recognizes or appreciates that the contents are the joint upshot of how the world is:

My answer is that they furnish the subject with certain essentially experiential demonstrative contents—‘That is thus’ (fully conceptual, as they must be)—his grasp of which (defeasibly—see 7.3) provides him with a reason to endorse them in belief. For a person’s grasp of such contents, as referring to the mind-independent objects which they do, and predicating the mind-independent properties which they do, essentially involves his appreciation of them as the joint upshot of the way things are anyway, in the mind- independent world around him, and his current point of view upon them and other relevant circumstances of perception. That is to say, he necessarily understands that his current apprehension that things are thus and so is in part due to the very fact that they are. He therefore recognizes the relevant content as his apprehension of the facts, his epistemic openness to the way things mind-independently are out there. (Brewer 1999, xvi-xvii)

The view is that there is an intimate connection between grasping experiential contents of the form “That is thus.” and seeing that the contents in question are veridical, and that, in virtue of this, experiences can serve as reasons for beliefs.

128

But what is it, it should be asked, for you to recognize or to appreciate or to understand

that the contents of your perceptual experiences are due to what the world is like? Is it to truly

believe it? Or just to believe it, whether truly or not? Or something else?

The first reading—i.e., the one on which to recognize or to appreciate or to understand

that p is to truly believe that p—is quite natural, given everyday parlance. In addition, passages

such as the following seem to require it:

As I have already said (6.1), such reasons are open to rational reflection and rejection in the sense that further background beliefs may make it unreasonable for him to endorse the content in question, if he believes, with reason, that he may well be hallucinating, for example, or that it would be extremely unlikely to come across anything which is thus in his present environment. Furthermore, his possession of such a reason at all is defeasible in the following sense. It is not indubitable for him whether or not he has one. A subjectively indistinguishable correlate, in sense (1c), of a true perceptual demonstrative content appears in this sense to give him equal right to endorse the content in question in belief. In such a case, though, since the content in question is entirely illusory, he is both wrong about the world and about his apparent epistemic openness to how things are out there. Thus his apparent reason depends upon his rightly recognizing his openness to the world. In the illusory case, he can have no such genuine recognition, hence any epistemic right to belief is merely apparent. (Ibid., 232, emphasis mine)

Brewer is herein responding to the objection that because every veridical experience of the kind he has in mind has a subjectively indistinguishable but non-veridical correlate, no such experience is sufficient for justification; it is further required, the objection goes, that the subject have reason for thinking that the experience in question is veridical. Brewer’s response, it seems, is that when the experience is non-veridical, the subject is wrong that the contents in question are the joint upshot of how the world is, and that the subject’s belief “That is thus.” is thereby unjustified.

Were this the right reading, though, the view would fall prey to the Terminus Problem.

The belief “That is thus.” is supposed to be non-inferentially justified. But what about the belief that the experiential content “That is thus.” is the joint upshot of how the world is? Were it not itself justified, then it would be impotent to help justify the belief “That is thus.”. Were it justified, then though it could help justify the belief “That is thus.”, the justification would not be non-inferential.

129

Though natural and though supported by passages such as the one above about

subjectively indistinguishable but non-veridical correlates, the first reading is explicitly

rejected—it seems—in other passages. So too for the second reading, the one on which to

recognize or to appreciate or to understand that p is just to believe that p. The recognition

involved in Brewer’s rationale, it seems, is not a belief at all—let alone a true belief. Here, for

starters, is a passage in which Brewer seems to deny that the subject, to have good reason for

believing “That is thus.”, needs to believe that the experiential content “That is thus.” is due to

what the world is like:

Perhaps this will provoke a further objection, though. For my account may now appear to exhibit a damaging structural similarity to the classical foundationalism which I rightly rejected above (4.2). Does it not effectively offer the following inference in response to the acknowledged need to explicate the role of a person’s perceptual experiences in the provision of reasons for his empirical beliefs? (1) I am entertaining the content that that is thus. (2) This is my epistemic openness to the mind-independent facts. ∴ (3) That is thus. Yet if my suggestion does amount to anything like this, then it is surely susceptible to analogous objections to those which I gave against the classical foundationalist account. First, there are all the difficulties raised by the apparently foundational epistemic status of the first premise. Second, the success of any such inference in delivering knowledge of its conclusion depends upon independent knowledge of the second premise; and this faces all the familiar problems of circularity. For knowledge that certain contents are a reliable guide to the way things actually are with certain external objects surely requires prior knowledge of the relevant mind-independent states of affairs. (Ibid., 238-9)

The problem, he suggests, would be with the justification of the premise beliefs. What would account for the justification of the subject’s belief that his experience has the content “That is thus.”? And what would account for the justification of the subject’s belief that the content is the joint upshot of how the world is? The nice thing about his view, the idea seems to be, is that no such beliefs are required. And here, for yet further evidence, are two passages in which Brewer stresses that the experiences he has in mind are non-endorsement states:

Perceptual demonstrative contents are delivered by experience unendorsed in belief. That is, experience simply provides a person with the resources to grasp certain contents of the form ‘That is thus’. (This is the belief-independence of perceptual experiences again.) As unendorsed contents, though, there is no question of the subject’s reasons for them. They are not the kind of condition of a person for which it is appropriate to seek after her

130

reasons at all. . . . They are not yet in the arena of epistemic appraisal. For they are not yet something to which the subject is in any way committed. (Ibid., 223)

This does indeed introduce an analogy with belief in my view, although I also stress an important disanalogy: perceptual experience does not, whereas belief does, involve assent (P&R, 5.3.3). (Brewer 2001, 16)

If the subject’s experience involves no endorsement, then so too for the subject’s recognition.

For the subject’s recognition is involved in the subject’s grasping (of the experience’s content)— which, in turn, is involved in, or just is, the subject’s having of the experience.

The problem now, though, is that the rationale itself is mysterious and in need of explication. The recognition stuff is supposed to shed light on how an experience, which is a non-endorsement state, could serve as a reason for, and thus justify, a belief. But what is it to have a non-endorsement recognition? And how is it that a recognition, given that it is a non- endorsement state, could serve as a reason for, and hence justify, a belief?

Either way, then, Brewer’s account is inadequate as a response to the trilemma challenge.

If the recognition is an endorsement state, the account falls to the Terminus Problem. If, instead, the recognition is a non-endorsement state, the account falls to the Endorsement Problem—in that, in the end, it simply ignores the Endorsement Problem. And without the recognition, the account is without a rationale.

4.4.7 The Lesson

The foundationalist cannot say that acquaintance (or awareness, or whatever) is an endorsement state, since then, it would seem, S would need justification for the acquaintance and, if he had it, the target belief would thus not be non-inferentially justified. The foundationalist, therefore, needs to say that acquaintance is something else. But this is why acquaintance offers little, if any, promise of helping with the trilemma challenge. For the heart of the trilemma challenge just is the question of how a non-endorsement state, such as an experience, could serve as a reason for a belief, or make it such that a belief is justified. Hence even if the foundationalist

131

succeeded in fleshing out the notion of acquaintance, the trilemma worries pertaining to experience would simply transfer, mutatis mutandis, to acquaintance—so that nothing, philosophically, would be gained by invoking acquaintance.

4.5 Inference Accounts

The accounts to be discussed in this section are significantly different from the accounts discussed in the previous section, and not just in the fact that the accounts to be discussed in this section say that experiences can serve as reasons for external-world beliefs. The distinctive feature of the accounts to be discussed in this section is that (paradoxically?) they try to account for non-inferential justification in terms of inference—in particular, explanatory inference.

I shall start with Peacocke’s account in The Realm of Reason, as it is a nice illustration of the first of the two forms that an inference account can take. I shall then examine Moser’s account in Knowledge and Evidence, which nicely illustrates the second of the two forms that an inference account can take. The dilemma for the foundationalist, I shall argue, is that the problems with accounts such as Peacocke’s push the foundationalist to accounts such as Moser’s, while the problems with accounts such as Moser’s push the foundationalist to accounts such as

Peacocke’s.

4.5.1 Peacocke

Peacocke’s account, in short, is this:

If S has a perceptual experience with the content that-p, if perceptual experiences with the content that-p are instance-individuated (so that, in normal conditions, what causally explains why a perceptual experience has the content that-p is that p), and if S has no reason for thinking that not-p, then S is entitled to believe that p.

Thus since, for instance, perceptual experiences with the content that-there-is-a-blue-book-a-bit- to-the-left are instance-individuated, S is entitled to believe that there is a blue book a bit to the

132

left if S has no reason for thinking otherwise and if S has a perceptual experience with the content that-there-is-a-blue-book-a-bit-to-the-left.

But in virtue of what, according to Peacocke, does an instance-individuated perceptual experience serve as a reason for a belief about the external world? This is where the notion of explanation enters the picture.

Peacocke’s rationale comes in three steps. First, he argues that (other things being equal) a complex phenomenon is likely to have a complexity-reducing explanation. Second, he argues that an instance-individuated perceptual experience is a complex phenomenon. And third, he argues that the best complexity-reducing explanation of an instance-individuated perceptual experience is that it is the result of a reliable representational device selected for by natural selection:

I suggest that the explanation of the occurrence of a perceptual experience with the instance-individuated content that p which most reduces complexity is that the experience is produced by a device which has evolved by natural selection to represent the world to the subject. This involves, other things being equal, the perceptual experiences produced in such a subject being predominantly correct. The occurrence of an experience with the representational content that p would hardly be a coincidence if its representational content held of the world, and the subject had a properly functioning perceptual system whose holistic complexities were adapted to its spatial embedding in the world. For a subject with such a perceptual system its being the case that p would in the predominance of cases then explain the occurrence of an experience with whatever complex relational property is involved in representing it as being the case that p. (Peacocke 2004, 87-8)

It is this, the idea goes, that enables a perceptual experience to serve as a reason for, and hence justify, a belief about the external world.

This, if sound, would make for a nice solution to the Endorsement Problem. For it is certainly true that states such as entertainings and hopings are not instance-individuated. There would thus be a principled reason for putting experiences in the class of non-endorsement states that can serve as reasons while putting states such as entertainings and hopings in the class of non-endorsement states that cannot serve as reasons.

Things are not so happy, however. Were S to believe that he is having a visual experience of a blue book a bit to the left and were S to further believe that—given that this

133

experience is the result of a reliable representational device selected for by natural selection—the best explanation of his having such an experience is that there is, in fact, a blue book a bit to the left, it would make perfect sense to say that S has good reason for thinking that there is a blue book a bit to the left. But without such beliefs, it remains a complete mystery as to how it is that

S, just in virtue of the experience and the explanatory fact, has good reason—or any reason, for that matter—for thinking that there is a blue book a bit to the left. It would seem that, without such beliefs, S would be entirely unaware of both the fact that he is having such an experience and the fact that the best explanation of his experience is that there is, in fact, a blue book a bit to the left. After all, it is not true in general that the mere fact that there is a phenomenon x such that p is the best explanation of x makes it such that S has good reason for thinking that p. The lesson, it would seem, is that the explanatory facts adduced by Peacocke would help in answering the

Endorsement Problem only if it were added that the subject is aware of them. Since only then would it make sense that the subject—in contrast to someone else who is aware of both the subject’s experience and the fact that the best explanation of it is that it is veridical—has good reason.

4.5.2 Moser

Moser’s account, like Peacocke’s, accords a key role to inference and explanation. It differs from Peacocke’s account, though, in according a key role to the awareness involved in the having of an experience. In this respect, it offers what looks to be a promising solution to the problem plaguing Peacocke’s account.

The account, roughly put, is this:

134

Moser, to his credit, gives detailed and precise explications of the notions of bestness (vis-à-vis explanations), contravening, and basing. It will prove best, though, to ignore such explications, as no such notion will play a significant role in the criticism to be given below.

Let me start, then, with the association condition, according to which it needs to be the case that S has associated, or is associating, p and c:

S has associated, or is associating, p and c just in case (a) S has been de re aware, or is de re aware, of c’s supporting p and (b) S, as a non-deviant result of the de re awareness in question, is so disposed that he would focus on c were he to focus on his evidence for p.

It is not enough, the idea goes, that the subjective non-conceptual contents of S’s visual experience of a blue book serve as a reason for, or make likely, the claim that there is a blue book before him. It further needs to be the case that S “sees”, or “appreciates”, that such contents serve

as a reason for, or make likely, such a claim. Otherwise, S’s belief that there is a blue book before him is merely justifiable.

This leads quite naturally to the experience condition. e itself, Moser says, involves an awareness—viz., an awareness of c—which, like the awareness involved in S’s de re awareness of c’s supporting p, is spelled out in terms of direct attention-attraction. S, in virtue of having e, is directly aware of c, in that c is the direct object of S’s attention.

It would seem that Moser’s account has a leg up on Peacocke’s. For whereas Peacocke’s account does not require any awareness of the justifying experience or the justifying explanatory relation, Moser’s account seems to require both such awarenesses.

The key question, though, is of what it would it be for something to be the direct object of

S’s attention? Moser is quick to point out that it would not be for S to have any beliefs about it.

But then what would it be? What would it be, for instance, for the subjective non-conceptual

contents of S’s experience of an apparent blue book (cbb) to be the direct object of S’s attention?

And what would it be for S’s attention to have as its direct object cbb’s supporting the claim that there is a blue book before him?

135

Here, by way of negative elucidation, Moser says what direct attention-attraction is not, saying that it is not predicative (since it is not conceptual), that it is not mere sensory stimulation, and that it is not attention-focusing:

Such attention attraction [i.e., direct attention-attraction], being nonconceptual, does not itself essentially involve one’s predicating something of the presented contents; yet of course it can be accompanied by such predicating. (Moser 1989, 81)

And such attention attraction is different from mere sensory stimulation, since it essentially involves direct awareness, albeit nonconceptual awareness, of what is presented in experience. (Ibid., 81)

Direct attention attraction is also different from one’s focusing attention on something, where one at least implicitly predicates the feature of individuality or of isolatability of what is presented in experience. Attention-focusing, as various psychologists have stressed, essentially involves psychological selection of some sort.22 Such selection essentially involves a form of conceptualization, since it essentially involves a form of objectual categorization. Attention attraction is not thus selective; and so in ordinary English we use the passive voice to characterize it, whereas we use the active voice to characterize attention-focusing. (Ibid., 81-2)

And here, by way of positive elucidation, Moser says what direct attention-attraction is, saying

that it is direct psychological presentation and that the relata (of the presentation relation) include,

in total, just the subject and the experiential contents of which he is aware:

In essence, direct attention attraction is one’s being directly psychologically “affected” by certain contents in such a way that one is psychologically presented with those contents. (Ibid., 82)

For current purposes I take the notion of psychological presentation to be conceptually basic. This notion is familiar from, and receives elucidation in, various foundationalist writings.23 Russell used the notion of presentation to clarify his famous notion of acquaintance: “to say that S has acquaintance with O is essentially the same thing as to say that O is presented to S” (1911, pp. 202-3). But I do not identify my notion of presentation with Russell’s notion of acquaintance, because (a) Russell’s notion allows for one’s being acquainted with objects of which one is not currently aware (p. 203), and (b) Russell’s notion seems to rely on a notion of selective attention (1914a, p.131).24 On my notion of presentation, one is presented with nonconceptual contents only if one is directly aware of those contents; and the directness of such awareness consists in its not essentially involving awareness of any other contents. Presentation, as I understand it, essentially involves direct nonconceptual noticing, but does not essentially involve conceptual noticing as or noticing that. (Ibid., 82)

For most of us, I suspect, the relevant notion of psychological presentation or direct attention attraction is adequately defined ostensively, by an appeal to common experiences. (Ibid., 83)

136

What it would be, then, for something to be the direct object of S’s attention is for S to be psychologically presented with it, where this is to be distinguished from predication, mere sensory stimulation, and attention-focusing.

Thus the overall view, with the claims about direct awareness (or psychological presentation) added in and the claims about contravening and basing subtracted out, runs as follows. First, S—in virtue of having e with subjective non-conceptual contents c—is directly aware of c. Second, p is the best explanation of why e has c. Third, S—in virtue of having associated, or associating, p and c—has been, or is, directly aware of c’s supporting p. And fourth, S—in virtue of the two awarenesses and the explanatory relation—has reason for thinking that p, which, in the end, is what makes S’s belief that p justified.

The question now, though, is of how it is that S, in virtue of (a) the fact that he is directly

aware of c, (b) the fact that p is the best explanation of e’s having c, and (c) the fact that he is

directly aware of c’s supporting p, has reason for thinking that p. How is it, to put the question in

terms of psychological presentation, that S, in virtue of (a) the fact that he is psychologically

presented with c, (b) the fact that p is the best explanation of e’s having c, and (c) the fact that he

is psychologically presented with c’s supporting p, has reason for thinking that p?

Were a direct awareness a belief, it would make perfect sense to say that S, in virtue of the two awarenesses and the explanatory relation, has reason for thinking that p. S would be aware of c in the sense that S would believe that e has c. S would be aware of p’s being the best explanation of e’s having c in the sense that S would believe that p is the best explanation of e’s having c. And (if such beliefs were themselves justified) S would thus have reason for thinking that p. Moser stresses, however, that the non-conceptual experiences he has in mind are non- endorsement states and so are not themselves in need of justification (or supporting reasons). The question remains, then, of how is it that S, in virtue of the two awarenesses and the explanatory relation, has reason for thinking that p. Put another way, Moser’s two awarenesses in no way

137

advance the debate, since they are not beliefs and since Moser says nothing by way of answering the Content Problem for them.

The problem is fully general. If a foundationalist sides with Peacocke, he is faced with the question of why the subject need not be aware of his experience, or of the fact that the best explanation of it is that it is veridical. If, on the other hand, he sides with Moser, he is faced with the question of how the two awareness should be understood. If he says that they are beliefs, then the regress would be up and running—in that the awarenesses, to justify, would themselves need to be justified. And if, following Moser even further, he says that they are non-beliefs of a rather mysterious sort, then his view is dialectically question-begging—amounting to nothing more than the view that there is an X, call it “direct awareness” or “direct attention-attraction”, such that X enables experiences to serve as reasons.

4.5.3 Heck

You might think that Heck provides a middle way between Peacocke and Moser—or, if not a middle way, at least a different way. Heck argues that, though (contra McDowell) perceptual experiences differ from beliefs in being non-conceptual in content, perceptual experiences are like beliefs in being representational (or presentational) in content, and that, because of this, perceptual experiences are also like beliefs in being able to serve as reasons.

This, you might think, is a promising avenue for the foundationalist, given that on this line (in

contrast to Peacocke’s line and Moser’s line) there is no issue as to whether the subject is aware

of the experience, and given that, in being representational in content, this line would provide for a principled epistemic distinction between experiences (on one hand) and states such as entertainings and hopings (on the other hand).

Consider:

The reason is that, as different as perception may be from beliefs—as isolated in certain ways as perceptual experience is from the influence of our beliefs—there is yet

138

something similar: Both purport to represent how the world is; both, we might say (borrowing some terminology from the philosophy of language) have assertoric force.26 Even when the world appears to be a way I know it not to be—when a stick I know to be straight looks to be bent when I partially immerse it in water (to use a tired example)—it still looks as if the stick is bent. That is to say, my experience represents the world as containing a bent stick: In a different way, to be sure, than my beliefs would were I to believe that the stick was bent, but it represents it as being that way nonetheless. (Heck 2000, 508)

What makes perceptual experiences more like beliefs than imaginings and entertainings, according to Heck, is that perceptual experiences, like beliefs but not like imaginings or entertainings, purport to represent how the world is. This is the sense in which, according to

Heck, experiences are representational, or presentational, in content.

It is hard to tell, though, exactly what Heck means in saying that a perceptual experience purports to represent how the world is. Some of the things he says make it seem that he is thinking of a perceptual experience as an endorsement state. But other things he says make it seem that he is thinking of a perceptual experience quite differently, as a non-endorsement state.

Here, for starters, he says that a perceptual experience is more like a voice saying “p.” than like a voice saying “Consider p.”:

That is, if perception were like a little voice telling me, “Consider the Thought (the possibility) that there is a desk in front of you,” then it might be clear enough how I might get from this Thought to the belief that there is a desk in front of me: No translation would be required. But it would remain obscure why I ought so to believe. (Ibid., 507)

Although the comparison with testimony can be misleading if not treated with care—it tends to suggest the representational theory considered in the last section—we might say that perception is more like a little voice saying, “There is a desk in front of you.” (Ibid., 508)

Here he says that a perceptual experience, though not a belief, is like a belief in involving an assertive attitude:

Perceptions are not beliefs: But they may yet be attitudes of some other kind, even if they are not desires, intentions, or entertainings. In fact, I suggest, perceptions are attitudes, attitudes that are like beliefs insofar as to be in a perceptual state is to hold an assertive, or presentational, attitude towards a certain content. (Ibid., 509)

Of course, it would be nice to know more about what it means to say that perceptual states are “assertive.” Unfortunately, I do not know how to explain this. Nor, however,

139

do I know how to explain what is involved in a belief’s being assertive: What exactly does it mean to say that beliefs purport to represent how the world is? I think an answer to this question could be converted into an answer for the case of perception, too: At least, that is the point of my relying upon this analogy here. (Ibid., 509)

And here he says that perceptual experiences justify beliefs in pretty much the way that beliefs justify beliefs:

If that is right, then we can begin to answer the question how perceptions justify beliefs, and how they provide us with reasons for belief, thus: pretty much the same way beliefs do—whatever that may be. (Ibid., 509)

Instead, let me say this: When I look around, the result is, not just that I come to be in a state with a certain content, but that I come to have a presentational attitude towards a particular nonconceptual content; then there is an inference30 or transition—call it whatever you like—from my perceptual state to some belief I recognize it to underwrite.31 My belief will then be justified by the perception on which I base it, in much the same way it might have been justified by another belief upon which I based it. Moreover, my perception is my reason for my belief: At least, we have not yet seen any reason it should not be. (Ibid., 510-11)

Together, these passages suggest the first reading, the reading on which Heck thinks that a perceptual experience is an endorsement state. To have a perceptual experience with content c is,

in part, to endorse c as an accurate representation of how the world is, in just the way that to have

a belief with content p is, in part, to endorse p as an accurate representation of how the world is.

The second reading—viz., the one on which a perceptual experience is a non- endorsement state—is suggested by Heck’s explicit adherence to the view that perceptual experiences are not themselves in need of reasons or justification. Why is it so intuitive that perceptual experiences are not themselves in need of reasons or justification? Because the subject, in having the experience, is not taking a stand on how the world is. The experience might end up causing the subject to take a certain stand on how the world is, a stand for which the subject needs reasons and justification. The experience itself, though, is not a stand on what the world is like—at least not a stand being made by the subject.

Heck at one point says that the reason why perceptual experiences are not themselves in need of reasons or justification is that perceptual experiences are involuntary, both in acquisition and in retention:

140

In a sense, something is “given” on this view: My perceptions, in particular, are something with which I just find myself “saddled.” They are merely caused: They are not something I can justify; but nor are they something that it makes sense to ask me to justify (or expect me to revise). But that does not prevent my perceptions from providing me with reasons for my beliefs about the world, nor from being what justify them. (Ibid., 522)

But that cannot be the reason. For beliefs too are sometimes involuntary, though beliefs are never not in need of reasons or justification. It might be true that S does not need reasons or justification to be blameless in his involuntary beliefs. But that would be an entirely different normative matter.

Consider, again, the claim that a perceptual experience is more like a voice saying “p.” than like a voice saying “Consider p.”. The crucial question is “Whose voice is it?”. If, on one hand, it is S’s, then it sure looks as though a perceptual experience is an endorsement state and that, thus, a perceptual experience, to justify, is itself in need of justification. If, on the other hand, it is not S’s but, say, S’s sub-personal cognitive system’s (in the way way that a thermometer can “say” that the temperature is 73 degrees), then, again, it is hard to see how a perceptual experience could serve as a reason for a belief; S, in having a perceptual experience, would not yet being taking a stand on how the world is, and thus S, it seems, would not yet be in a position to take a stand on the claim that p.

4.6 Reliabilist Accounts

There is an important division in the reliabilist camp between process reliabilists such as

Alvin Goldman and ground reliabilists such as William Alston, Marshall Swain, and Ernest Sosa.

The process reliabilist thinks that whether S’s belief that p obtains is justified hinges on whether the belief-forming process that produced it is reliable. The ground reliabilist thinks that the key is whether the ground—or reason state—on which it is based is a reliable indicator of p’s obtaining.

I said in Chapter 1 that, at least for the purposes of this dissertation, I mean to remain neutral on the truth of process reliabilism. The reason, remember, is that process reliabilism

141

allows (or at least seems to allow) for justification in scenarios in which the subject does not have good reason, and that, insofar as I am interested in what it would be for a belief to be justified, I am interested in what it would be to have good reason.

Things are a bit different with ground reliabilism, for two reasons. First, it requires having good reason, in the form of a good (or adequate) ground, for justification. Second, it allows for scenarios in which what makes it such that the subject has good reason for thinking that p obtains is that he has an experience that, in fact, is a reliable indicator of p’s obtaining. For instance, it allows for scenarios in which what makes it such that S has good reason for thinking that there is a blue book before him is that he is having a visual experience of a blue book before him. It thus needs to be asked whether ground reliabilism can supply an answer to the trilemma challenge.

Take Alston’s brand of ground reliabilism, for example:

S’s belief b is justified if and only if (a) b is based on a ground g, (b) g is adequate, and (c) S has neither rebutters nor neutralizers relative to b.

b is based on g if and only if the belief-forming process that produced b took account of g in producing b.

g is a ground for b only if g is a belief or an experience.

g is adequate if and only if P(b is true/g) is very high.

Alston’s account would not require in the blue-book case that S be aware of his visual experience

of the blue book before him, or that S be aware of the fact that, given that he is having a visual

experience of a blue book before him, the probability is very high that there is a blue book before

him. It would be enough that, in fact, he is having such an experience and that, in fact, such a

probability is very high.

You might think that there is an intuitive sense in which accounts such as Alston’s can

answer the Content Problem. It is not that experiences can serve as reasons for beliefs because of

any logical relation between the former and the latter, the idea would go. After all, experiences

are non-propositional in content, and logical relations stand only between propositions. Rather, it

142

is because experiences can serve as reliable indicators. The rationale for saying that experiences can serve as reasons for beliefs, then, is that experiences can serve as reliable indicators of what the world is like—so that, for instance, in having a visual experience of a blue book before him, S has something that makes it highly likely (and objectively so) that his belief about the book is true.

This line of rationale, notice, could be given for anything you have, such as a smile.

Imagine that, for whatever reason, S smiles pretty much only when x obtains. Imagine also that, on a certain occasion, S forms the belief that x obtains, and that, somehow, S’s smile is the ground of this belief. Then by the rationale given in the previous paragraph it would follow that, in a sense, S has good reason for thinking that x obtains, in that, in a sense, S has something that makes it likely that his belief about x is true. But intuitively, of course, it would not be the case that S has good reason for thinking that x obtains.

There is nothing immediately troubling in this for ground reliabilists such as Alston, since, according to such reliabilists, the only things that can serve as grounds are experiences and beliefs. S’s smile is neither an experience nor a belief, and so S’s smile could not be a ground, let alone an adequate ground, for S’s belief that x obtains.

But why are grounds restricted to experiences and beliefs? Why not include things such as smiles, and blood pressure, and heart rate, and so on? It is not because of the rationale suggested above, given that the rationale suggested above would apply just as easily to things such as smiles and blood pressure and heart rate—were the world such that things like smiles and blood pressure and heart rate reliably indicated what the world is like. But then what is it?

Alston, for one, argues for so restricing the class of grounds by arguing that a ground needs to be the sort of thing that you can fairly directly access on reflection—which includes processes such as introspection, but not processes such as vision. Things such as smiles, blood pressure, and heart rate are accessible not by reflection, but by sense perception and inference.

143

The question now, however, is why things such as hopes and entertainings are excluded from the class of grounds. Surely, things such as hopes and entertainings are every bit as fairly directly accessible on reflection as are experiences and beliefs. If the criterion is just being fairly directly accessibile on reflection, then things such as hopes and entertainings should be included in the class of grounds, so that if the world were such that things such as hopes and entertainings reliably indicated what the world is like, then things such as hopes and entertainings could serve as adequate grounds for beliefs.

But this, of course, would be absurd. Imagine, for example, that Jack lives in a very strange world, a world in which Betty, an evil demon of sorts, tinkers with the world in direct response to certain of Jack’s imaginings, as follows. When Jack reads the nth line in his fiction book A and, as a result, imagines the described scenario, Betty tinkers with the world so that it is

true that nB—whereB nB B is the scenario described in the nth line of Jack’s fiction book B. What is

more, Betty has rigged Jack’s belief-forming processes so that when he imagines the scenario

depicted in the nth line of A, he comes to form the belief that nB.B For instance, suppose that the

5th line in A is the sentence “The biggest building in Chicago is on fire right now.” and that the

5th line in B is the sentence “The President is scratching his elbow right now.”. Then were Jack

to read the 5th line in A and thereby imagine that the biggest building in Chicago is on fire right

now, Jack would come to form the belief that the President is scratching his elbow right now, and

Betty would tinker with the world so that it is true that the President is scratching his elbow right

now. Given the weirdness of Jack’s world, his imaginings pertaining to the contents of book A

would reliably indicate what the world is like, so that, for example, Jack’s imagining that the

biggest building in Chicago is on fire right now would be a reliable indicator of the President’s

scratching his elbow right now. Nonetheless, it would certainly be absurd, even given the

weirdness of Jack’s world, to say that Jack’s imaginings pertaining to the contents of book A give

him good reason for his corresponding beliefs about what the world is like. There is no clear

sense in which, for instance, Jack’s imagining that the biggest building in Chicago is on fire right

144

now makes it such that Jack has good reason for thinking that the President is scratching his elbow right now. But that would be the case were ground reliabilism true, and were the class of grounds to include things such as hopes and entertainings.

The key question for an Alstonian foundationalist, then, is why experiences, but not things such as entertainings, hopings, and imaginings, can serve as reasons for beliefs. To make sense of how experiences can serve as reasons for beliefs, he would need to say more than just that experiences can reliably indicate what the world is like—since so too could things such as entertainings, hopings, and imaginings. He would need to point to something that both differentiates experiences from such states, and that, when coupled with the reliability considerations, sheds light on how experiences can serve as reasons for beliefs.

4.7 Conclusion

The way is finally clear to develop and defend EMV, the brand of moderate explanationist, meta-perspectivalist, veridicalist coherentism sketched in Chapter 1. I argued in

Chapter 2 that the neither the Alternative-Systems Objection nor the Isolation Objection is successful against the coherentist. I argued in Chapter 3 that the explanationist coherentist is immune to the Database Objection. And I argued in this chapter against the Experience

Objection, arguing that—contrary to intitial appearances—an experience is not the right sort of

thing to serve as a reason, either for or against beliefs.

Do keep in mind, however, that my account implies that there is a contingent respect in which experiences are nonetheless required for justification, in that, though they cannot themselves serve as reasons, they can (and contingently do) help in enabling beliefs to serve as reasons. The meta-perspectivalist requirement, which I articulate and defend in Chapter 6, and the veridicality requirement, which I develop and argue for in Chapter 7, together secure this.

1 Though the name is the same, Moser’s objection is very much different from the isolation objections considered in Chapter 2.

145

2 Feldman is yet another fan of the objection. See Feldman 2003, 69.

3 It would invite hard questions, though--questions such as “But why are further beliefs required in all cases, so that fitting with experience is never sufficient for justification?”.

4 This would be similar to arguing against skeptical scenarios (e.g., brain-in-a-vat scenarios) by arguing—a la Putnam—for a theory in the philosophy of language that precludes such scenarios.

5 Steup, for one, makes just this mistake—reading the coherentist as claiming that nothing can count as a source of justification except a belief, and then arguing that the coherentist is in no position (dialectically) to make such a claim. See Steup 1996.

6 The qualifier “at least some” is there to allow for mind-externalist coherentists who think that coherence is necessary but not sufficient for justification.

7 BonJour—who used to be a foe of, but now is a friend of, the view that experiences can serve as reasons for beliefs about experiences—presses just these worries against C. I. Lewis’s account (as expressed in Mind and the World Order and An Analysis of Knowledge and Valuation). See BonJour 1985.

8 Haack, though, says that it is neither foundationalist nor coherentist, but “foundherentist”.

9 Haack, in explicating the notion of a subject’s evidence, distinguishes between evidential states and non- evidential states—restricting a subject’s evidence to his evidential states. She puts beliefs and experiences, on one hand, on the evidential side, and desires and fears, on the other hand, on the non-evidential side. But why? Why not put desires and fears on the evidential side? And why not put beliefs and experiences on the non-evidential side? Haack’s “answer” is that desires and fears, but not beliefs or experiences, belong on the non-evidential side of the evidential/non-evidential divide because intuitively desires and fears, but not beliefs or experiences, belong on the non-evidential side of the evidential/non-evidential divide. See Haack 1993, 76-7.

10 Chalmers further distinguishes between the community-relational concept of phenomenal redness and the individual-relational concept of phenomenal redness, with the first being glossed as “the phenomenal quality typically caused in normal subjects within my community by paradigmatically red things” and the second being glossed as “the phenomenal quality typically caused in me by paradigmatically red things”.

11 See Keith Korcz’s “Recent Work on the Basing Relation” (1997) for a useful survey of the extant theories on the basing relation.

12 This would be compatible with Peacocke’s discussion of the reliabilist account suggested by Sydney Shoemaker. Though he rejects such an account, he rejects it on the grounds that it does not require reasons. This is in no way at odds with the claim that reliability is relevant to whether a reason is good. Think of William Alston’s account of justification—an account which is at odds with the kind of reliablism on which reasons are not required (e.g., Goldman’s reliabilism) and which nonetheless construes the notion of a good reason in terms of reliability.

13 Peacocke suggests that similar things could be said about having the concept of experience, having the concept of sensation, having the concept of imagining, and so on, so that, for example, there is a tight, though non-constitutive, connection between S’s imagining p and S’s believing that he is imagining that p.

14 Given that experiences and sensations are not reason-led states and given that the way in which conscious beliefs are supposed to serve as reasons for appropriately-related self-ascriptions is supposed to be identical to the way in which experiences and sensations are supposed to serve as reasons for appropriately-related self-ascriptions, it follows that—though conscious beliefs are commitment states— this is not what enables them to serve as reasons for appropriately-related self-ascriptions. And given this, it follows that Peacocke’s account of the justification of introspective beliefs is immune to the Terminus

146

Problem; in neither kind of case (i.e., the case in which the reason is an experience or sensation and the case in which the reason is a conscious belief) is there an issue about whether the reason itself is supported by a reason.

15 A further possibility, though not of help to Peacocke, is that Peacocke’s over-all account is inconsistent.

16 He argues that the condition requiring just that the doxastic presumption be true is too permissive, so that there is no intuitive sense in which the satisfaction of the conditions in the account suffices for justification. He argues that the observation requirement, requiring not actual input but just apparent input, is too weak to handle anti-coherentist arguments such as the Alternative-Systems Objection and the Isolation Objection. And he argues that there is a problem with memory beliefs much like the problem with second-order beliefs (i.e., the one that led him to add the condition involving the doxastic presumption), so that the account gives an implausible story about the justification of memory beliefs. See BonJour 1999.

17 For more from BonJour on non-conceptual content, see BonJour 2003 pp.71-2.

18 BonJour is simply being a bit sloppy when, with respect to the experience case, he says that “where such a relation of description exists, the character of the non-conceptual object is what determines whether the conceptual description is correct or true” (Ibid., 72). It is not just the character, or content, of the experience that makes the experiential belief true; the experiential belief involves more than just the claim that the content in question exists. Rather, it is the content’s being the content of one of the subject’s experiences that makes the experiential belief true. Remember: the experiential belief has the form “I am having an experience with content c.”, in the same way that the second-order belief, in the belief case, has the form “I hold the belief that p.”.

19 Timothy McGrew makes the same mistake. He too gives an account on which experiences can justify beliefs. And he too gives a rationale that, though perhaps fine on its own, is in conflict with the metaphysics of the account. See McGrew 2003.

20 Granted, BonJour is explicitly giving a sufficient condition for justification and so, strictly speaking, is not explicitly saying that the second awareness is required. It would be odd, though, for him to bring it up when explaining how his account works for experiential beliefs if it were not essential to the account.

21 Consider: Indeed, such a non-apperceptive, constitutive awareness of content might be said to be strictly infallible in something like the way that foundationalist views have traditionally claimed (but which most have long since abandoned). Since it is in virtue of this constitutive or “built-in” awareness of content that the belief is the particular belief that it is with the specific content that it has, rather than some other belief or some other sort of state, there is apparently no way in which this awareness of content could be mistaken—simply because there is no independent fact or situation for it to be mistaken about. (Ibid., 64) Here, in saying that “there is no independent fact or situation for it to be mistaken about”, he seems to be saying that the built-in awareness is not distinct from the content, a claim, it seems, that would entail that it is also not directed at the content.

147

CHAPTER 5

MODERATE EXPLANATIONISM

I am certainly not alone in backing moderate explanationism, the view, roughly, that all

cogent inductive inferences (or inferential relations) are explanatorily virtuous. Gilbert Harman

defends it in “The Inference to the Best Explanation” (1965), Thought (1973), and Change in

View: Principles of Reasoning (1986). Peter Achinstein argues for it in The Nature of

Explanation (1983) and The Book of Evidence (2001). William Lycan propounds it in Judgement and Justification (1988) and “Explanation and Epistemology” (2002). Peter Lipton, in Inference to the Best Explanation (1991), argues on its behalf. And, interestingly enough, David Hume

endorses it, or at least something close to it, in An Enquiry Concerning Human Understanding

(1777):

All reasonings concerning matter of fact seem to be founded on the relation of Cause and Effect. By means of that relation alone we can go beyond the evidence of our memory and senses. If you were to ask a man, why he believes any matter of fact, which is absent; for instance, that his friend is in the country, or in France; he would give you a reason; and this reason would be some other fact; as a letter received from him, or the knowledge of his former resolutions and promises. A man, finding a watch or any other machine in a desert island, would conclude, that there had once been men in that island. All our reasonings concerning fact are of the same nature. And here it is constantly supposed, that there is a connexion between the present fact and that which is inferred from it. Were there nothing to bind them together, the inference would be entirely precarious. The hearing of an articulate voice and rational discourse in the dark assures us of the presence of some person: Why? Because these are the effects of the human make and fabric, and closely connected with it. If we anatomize all the other reasonings of this nature, we shall find, that they are founded on the relation of cause and effect, and that this relation is either near or remote, direct or collateral. Heat and light are collateral effects of fire, and the one effect may justly be inferred from the other. (Hume 1993, 16-7, emphasis mine)

148

Hume, in saying that the inference would be entirely precarious were it non-causal, seems to be saying that an inferential relation (between propositions about matters of fact) is cogent only if it

is causal—moving either from cause to effect, or from effect to cause, or from effect to effect by

way of a common cause.1

A few clarificatory remarks, as to what I shall be arguing for (in arguing for moderate explanationism), are in order. First, keep in mind that, as I said in Chapter 1, I am not arguing that there is a cogent inductive inferential relation between the claim that e obtains and the claim that h obtains only if h’s obtaining potentially explains e’s obtaining. I mean to allow for cogent inductive inferential relations where e’s obtaining potentially explains h’s obtaining; these would be instances of inference to an explainee. For example, the claim that Jim has a severe chest wound is a cogent inductive reason (or at least a part of a cogent inductive reason) for thinking that he will soon die. Yet it is not Jim’s soon dying that potentially explains his having a severe chest wound. It is Jim’s having a severe chest wound that potentially explains his soon dying.2 I also mean to allow for cogent inductive inferential relations where there is a phenomenon c such that c’s obtaining both potentially explains e’s obtaining and potentially explains h’s obtaining.

For instance, the claim that the street is wet is a cogent inductive reason for thinking that the roof is wet, though the street’s being wet neither potentially explains nor is potentially explained by the roof’s being wet. Its just now raining, though, both potentially explains the street’s being wet, and potentially explains the roof’s being wet.

Second, the kind of moderate explanationism that I am defending is silent on the question of how the man-in-the-street makes inductive inferences. That is a psychological question, which is largely irrelevant to the philosophical question of what makes cogent inductive inferential relations cogent.3

The aim in this chapter, then, is to further the case for the view that all cogent, or good, inductive inferential relations are explanatorily virtuous—in that each such inferential relation is an instance of either inference to the best explainer, or inference to an explainee, or an

149

explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee (and/or one or more valid deductive inferential-relation forms). I shall try to further clarify, and flesh out, the main features of moderate explanationism. For instance, I

shall argue that the moderate explanationist would do well to explicate the notion of bestness in terms of Philip Kitcher’s account of unification. I shall also try to rebut the most pressing objections to moderate explanationism. I shall argue, for example, that though, as Richard

Fumerton points out, statistics are oftentimes quite relevant to cogency, and though neither inference to the best explainer nor inference to an explainee makes explicit mention of any such statistics, there is nothing threatening in any of this to the moderate explanationist. And third, I shall try to improve, mostly in thoroughness, the main positive argument for moderate explanationism. In particular, I shall try to reduce inductive argument forms such as argument

from authority, argument from analogy, and statistical syllogism to moderate explanationism—in

the same way that Harman and Lycan, for instance, try to reduce inductive generalization and enumerative induction to moderate explanationism.

5.1 Bestness, Loveliness, and Likeliness

There is much to be learned from Achinstein’s discussion of moderate explanationism, both in terms of reasons for moderate explanationism and in terms of reasons against possible objections to moderate explanationism. His examples against the leading alternative accounts of evidence are decisive (with one exception).4 He gives the key ingredients to solving one of the

most pressing objections to moderate explanationism—a problem pertaining to the reduction of

statistical syllogism to moderate explanationism. And his discussion of Hempel’s ravens paradox

and Goodman’s grue paradox is careful and insightful, as is his discussion of Thomson’s

discovery of the electron and Perrin’s argument for the existence of molecules.

150

There is a significant problem with Achinstein’s positive account, however. Namely, the explanans involves a notion that is very much related, if not identical, to the explanandum. Let me explain.

Achinstein’s view is this:

e is evidence that h if and only if (1) p(h/e) is higher than ½ and (2) p(there is an explanatory connection between e and h/e) is higher than ½.

Regarding the notion of an explanatory connection, Achinstein has much to say. He distinguishes, as I did above, between three kinds of connections, where in the first connection h’s obtaining potentially explains e’s obtaining, in the second e’s obtaining potentially explains h’s obtaining, and in the third e’s obtaining and h’s obtaining have a common potential explainer.

He also sketches his account of explanation, the details of which he supplies in The Nature of

Explanation. But regarding the notion of probability, Achinstein, at the end of the day, says little.

He starts by appealing to the notion of reasonableness:

According to the present theory, probability, understood as a concept satisfying the probability rules of chapter 3, can be construed as a measure of how reasonable it is to believe a proposition. On this view, reasonableness of belief in a proposition admits of degrees and is subject to the formal rules of probability. So, for example, according to an addition rule, if h1 and h2 are mutually exclusive, then the degree of reasonableness of believing the disjunction h1 or h2 is the sum of the degree of reasonableness of believing each disjunct. Since, in accordance with the rules, probability is measured between 0 and 1, the claim that the probability of a hypothesis h is 1 is understood as saying that believing h has the highest degree of reasonableness. At the other extreme, if the probability of h is 0, then believing that h is false has the highest degree of reasonableness. At the midpoint, saying that the probability of h is ½ means that it is just as reasonable to believe h true as to believe h false, in which case it is reasonable to have no belief either way, that is, it is reasonable to suspend belief. The claim that the probability of h is ¾ entails that believing h true is 3 times more reasonable than believing h false. More generally, if the probability of h is r, so that the probability of not-h is 1-r then believing h is r/1-r as (or more) reasonable as (than) believing not-h. (Achinstein 2001, 95, emphasis mine)

The account, with this, reads as follows:

e is evidence that h if and only if (1) it is more reasonable, given e, to believe that h than to believe that not-h and (2) it is more reasonable, given e, to believe that there is an explanatory connection between e and h than to believe that there is not an explanatory connection between e and h.

151

The question now, though, is of what, according to him, makes it such that a proposition x makes it more reasonable to believe a proposition y than to believe not-y; without an answer to this question, we would be without answers to the questions of what it is for (1) to be true, and of what it is for (2) to be true. Then, in answer to this sort of question, he distinguishes between two possible answers, opting, in the end, for the second:

The question of whether something is reasonable to believe can be considered in one of two senses: in a sense that depends on what (other) beliefs are held by some individual (or type of individual), group, or generally by people, and in a sense that does not. For example, although eating a pound of arsenic is fatal within 24 hours, suppose that the authorities in the community believe that arsenic, even a pound of it, promotes health when put on food. And suppose that people know that Ann ate a pound of arsenic 24 hours ago, and also know what the authorities believe and have no reason to question them. Is it, or is it not, reasonable to believe that Ann is dead or dying? Both answers are possible. We can say that it is not reasonable, at least for members of this community, to believe this, since the authorities preach the benefits of arsenic. We can also say that it is reasonable, even for members of this community, to believe that Ann is dead or dying, since (unbeknownst to the community and its authorities) arsenic is a deadly poison. The latter normative fact about what it is reasonable (even for this community) to believe is determined by, or “supervenes” on, physical facts about arsenic and the human body. It is not determined or affected by facts about what people believe about arsenic and the body. Nor is it relativized to an epistemic situation. The claim is not that it is reasonable for someone in such and such an epistemic situation to believe that Ann is dead or dying. (Ibid., 96)

But what makes the physical facts about arsenic and the human body entail that it is reasonable, given Ann’s eating a pound of arsenic, to think that Ann is dead or dying? Put in general terms, what makes the physical facts about phenomenon f and phenomenon g entail that it is reasonable, given that f obtains, to think that g obtains? Unfortunately, he never says. What he ends up doing, then, is explicating the notion of evidence in terms of the closely related, if not identical, notion of reasonableness, about which he says very little.

You might think that Achinstein would answer by saying that what makes the physical facts about f and g entail that it is reasonable, given that f obtains, to think that g obtains is that there is an explanatory connection between f and g. Or, a bit differently, you might think that he would answer by saying that what does it is the fact that it is probable that there is an explanatory connection between f and g. Neither answer would work, though. The first would stand in direct

152

conflict with his claim that e can be good evidence that h even if, in fact, there is no explanatory connection between e and h. The second, on the other hand, would render the explication circular, since then he would be explicating the probability facts in terms of the reasonableness facts, and the reasonableness facts, in turn, in terms of the probability facts.

There is an important lesson here. In answer to the questions “What makes the best of competing potential explainers the best?” and “What makes the best of competing potential explainers good enough?”, the moderate explanationist should steer clear of notions such as reasonable and likely. For the cogency question—viz., the question of what makes an inductive inferential relation cogent—just is the question of what makes an inductive inferential relation reasonable, or likely. That is, the question of whether an inductive inferential relation between claim X and claim Y is cogent just is the question of whether X makes it reasonable to think that

Y is true, or whether X makes it likely that Y is true. Thus though the claims about an explanatory connection might be of help, we would still be left with questions about reasonableness and likeliness.

Lipton makes a similar point about answering the question of what makes the man-in-the- street count an inductive inferential relation as cogent. He argues that reasonableness and likeliness answers would be of little help, since they would simply invite the questions “What makes it such that the man-in-the-street thinks that Y, given X, is reasonable to accept?” and

“What makes it such that the man-in-the-street thinks that Y, given X, is likely to be true?”.

But how, then, should the moderate explanationist answer the questions “What makes the best of competing potential explainers the best?” and “What makes the best of competing

potential explainers good enough?”? The answer, not surprisingly, is in terms of the notion of explanatory virtue, or explanatory loveliness (to borrow a term from Lipton)—where explanatory virtue and explanatory loveliness are to be understood in terms of things such as simplicity and unificatory power. What makes the best of competing potential explainers the best, the idea goes, is that it is the loveliest. And what makes the best of competing potential explainers good enough

153

is that it is lovely enough. The view, in other words, is that the bestness facts and the good- enough-ness facts are fixed by the loveliness facts, which, in turn, are fixed by facts such as the simplicity facts and the unificatory-power facts.

Lipton, siding even further with loveliness over likeliness, suggests that degree of likeliness and degree of loveliness can come apart, so that loveliness and likeliness pick in favor of different winners in the competition for bestness. The reasons for this, he argues, are that

likeliness, but not loveliness, is a function of the total evidence, and that likeliness, but not

loveliness, is a function of both the extent and the competitiveness of the competition:

One of the reasons that likeliness and loveliness sometimes diverge is that likeliness is relative to the total available evidence, while loveliness is not, or at least not in the same way. We may have an explanation that is both lovely and likely given certain evidence, unlikely given additional evidence, yet still a lovely explanation of the original evidence. Newtonian mechanics is one of the loveliest explanations in science and, ay one time, it was also very likely. More recently, with the advent of special relativity and the new data that support it, Newtonian mechanics has become less likely, but it remains as lovely an explanation of the old data as it ever was. Another reason for the divergence is that the two criteria are differently affected by additional competition. A new competitor may decrease the likeliness of an old hypothesis, but it will usually not change its loveliness. Even without the evidence that favored special relativity, the production of the theory probably made Newtonian mechanics less likely but probably not less lovely. (Ibid., 62)

Two comments. First, that degree of likeliness and degree of loveliness can come apart in no way entails that what is highest in likeliness and what is highest in loveliness can come apart. The claim that Newtonian mechanics is just as lovely as ever is perfectly compatible with the claim that the special theory of relativity is the loveliest. But then the case would be a case in which the likeliest is also the loveliest. Second, that what is likeliest and what is loveliest can come apart would be decisive against moderate explanationism—as it would then allow for cases in which the best explainer, and thus the explainer that should be inferred, is unlikely. If, first, h is the loveliest, h’ is the likeliest, and h’ is unlikely, then, of course, so too would be h. And if, instead, h’ is likely, then, again, h would be unlikely, given that h and h’ are competitors. Thus it is crucial that the moderate explanationist construe the notion of loveliness so that what is likeliest and what is loveliest cannot come apart.

154

To sum up, the moderate explanationist should neither construe bestness and goodness in terms of notions such as reasonableness and likeliness (since then the account would be circular), nor construe bestness and good-enough-ness so that what is best and what is good enough can differ from what is likeliest and what is likely enough (since then the account would be open to counterexample). Rather, he should construe bestness and good-enough-ness in terms of loveliness, and then construe loveliness such that what is loveliest is likeliest, and what is lovely enough is likely enough.

5.2 Deductivism and Potential Explainers

What is it for h’s obtaining to be a potential explainer for e’s obtaining? Or, on the flip

side, what is it for e’s obtaining to be a potential explainee for h’s obtaining?

The answer would seem to be readily forthcoming. The requirements for a potential explainer coincide with the requirements for an actual explainer save the requirement of actuality.

And, likewise, the requirements for a potential explainee coincide with the requirements for an actual explainee save the requirement of actuality.

Consider, by way of illustration, the way in which Carl Hempel distinguishes between

potential explanations and true explanations. A potential explanation, he says, is a valid

argument where the conclusion is a description of the explanandum, and where at least one of the

premises is a law-like sentence that is essential to the argument’s validity. A true explanation, in

contrast, is a potential explanation the premises of which are true. With this distinction in hand,

we could then say that h’s obtaining is a potential explainer for e’s obtaining just in case (a) the

argument “h obtains. Hence, e obtains.” is valid and (b) “h” involves a law-like sentence that is

essential to the argument’s validity.

But consider the rain case from above. That the inferential relation is cogent is supposed to follow, on moderate explanationism, from the fact that the best potential explainer for the street’s being wet is its having just now rained, and the fact that its having just now rained is a

155

potential explainer for the roof’s being wet. The problem, though, is that the arguments “It just now rained. Hence, the street is wet.” and “It just now rained. Hence, the roof is wet.” are invalid—so that, if the Hempel-inspired analysis of a potential explainer were true, it would be

false that the best potential explainer for that the street’s being wet is its having just now rained,

and false that its having just now rained is a potential explainer for the roof’s being wet.

This problem arises not just from Hempel’s account of explanation, but also from any other version of deductivism: the view that h’s obtaining explains e’s obtaining only if the argument “h obtains. Hence, e obtains.” is valid. Philip Kitcher’s unificationist account, for instance, would give rise to the same problem.

Perhaps, then, the moderate explanationist should simply reject deductivism, and thus allow for explanation without entailment. Perhaps he should follow David Lewis and say that explaining why e obtains consists in giving causal information about e—regardless of whether the information involves laws, and thus regardless of whether the information entails that e obtains.

Letting one’s epistemology dictate one’s philosophy of science would not be without precedent. Harman, for example, argues from his account of knowledge to the claim that at least some explanations are invalid, and that at least some such invalid explanations are also incogent.

Suppose, says Harman, that the probability of heads for coin B is .5 while the probability of heads for coin C is .9, and that Sidney, who randomly picked one of the coins and flipped it ten times, got just three heads, and then on that basis inferred that the coin she flipped was B. Then, presumably, Sidney knows that the coin she flipped was B. But that, in turn, requires (given the account of knowledge at play) that Sidney’s flipping B be the best of the available potential explainers for Sidney’s getting so few heads, which—since Sidney’s flipping B neither guaranteed nor made it likely that Sidney would flip exactly three heads—in turn requires that at least some explanations be neither valid nor cogent.

The question, though, is whether the moderate explanationist has to reject deductivism.

Yes, he could circumvent the worry by rejecting deductivism. But is that the only way?

156

Peter Railton’s writings on explanation are quite helpful with this. Railton distinguishes between the notion of an ideal explanatory text and the notion of a piece of explanatory information. The ideal explanatory text for why e obtains is the complete explanation of why e obtains, which involves everything on which e’s obtaining is in any way dependent. In contrast, a piece of explanatory information as to why e obtains is a piece of information that answers at least one question about the ideal explanatory text for e’s obtaining. Railton then uses this distinction in warding off possible objections to his D-N-P model of probabilistic explanation, such as the objection that there are numerous seemingly adequate explanations of indeterministic phenomena that make no explicit mention of statistical laws and that, thus, do not fit his account.

Railton stresses that his account is supposed to answer the question of what the ideal explanatory text for an indeterministic phenomenon would look like—so that, though it entails that no such seemingly adequate explanation is complete (since no such explanation is an ideal explanatory text), it in no way entails that no such explanation is explanatory. The point, to put it another way, is that though the account entails that no such explanation gives a maximal degree of explanatory information, it in no way follows from this that the account entails that no such explanation gives any explanatory information, or even that no such explanation gives a high degree of explanatory information.

This is helpful in that the moderate explanationist could make a similar move. No, the claim that it just now rained does not entail that the street is wet, or that the roof is wet. And so no, its having just now rained is not a potential explainer for the street’s being wet, or for the roof’s being wet. Nonetheless, its having just now rained is a part of—and an essential part of—a larger phenomenon that does, in fact, necessitate both the street’s being wet, and the roof’s being wet. And so its having just now rained, though not itself a potential explainer, is a part of—and an essential part of—a larger phenomenon that is, in fact, a potential explainer for both the street’s being wet, and the roof’s being wet. The moderate explanationist could then account for the cogency of the inferential relation between the claim that the street is wet and the claim that

157

the roof is wet by saying, in part, that its having just now rained is both an essential part of a potential explainer for the street’s being wet, and an essential part of a potential explainer for the roof’s being wet. This is the line that I shall take.

5.3 Explanation

What, if anything, should the moderate explanationist say about explanation? Need he take a stand on explanation, and then use the favored account in explicating the notions of a potential explainer and a potential explainee? Or can he stay neutral on explanation, taking the notions of a potential explainer and a potential explainee as primitives?

For the most part, things are analogous to functionalism (in the philosophy of mind) and causation. The functionalist, for the most part, need not take a stand on the nature of causation, even though he appeals to the notion of causation in cashing out the notion of playing a functional role. I say “for the most part” since, of course, the functionalist should avoid any account of causation construed, even in part, in terms of playing a functional role. The same is true with respect to the moderate explanationist and explanation. He need not—but can—take a stand on the nature explanation, at least for the most part. Any account on which the notion of explanation

(or, better, the notion of a potential explanation) is explicated, even in part, in terms of cogency would spell trouble, since it would make the overall account blatantly circular. But other than that, the moderate explanationist, qua moderate explanationist, is free to choose from among the live options, and is even free not to choose.5

The same, then, holds for the moderate explanationist coherentist—with the added proviso that he avoid accounts of potential explanation that are explicated, even in part, in terms of justification, and that, if (like me) he explicates justification partly in terms of explanation (and not just in terms of potential explanation), he also avoid accounts of explanation that are explicated, even in part, in terms of justification. Otherwise, he would be explicating justification

partly in terms of justification.6

158

It turns out, somewhat surprisingly, that Kitcher’s unificationist account of explanation is taboo for the moderate explanationist coherentist—at least the kind who (like me) explicates justification not just in terms of potential explanation, but also in terms of explanation. The problem, put simply, is that Kitcher explicates explanation partly in terms of the state of science in the limit of rational, or justified, scientific inquiry, so that the explanation facts, on Kitcher’s account, are partially determined by the justification facts.

I shall argue below, though, that the moderate explanationist would do well to use

Kitcher’s account of unification, which is a part of Kitcher’s account of explanation, in

explicating the notion of explanatory virtue, or explanatory loveliness. I shall argue, in particular,

that by doing so he would be able to answer two of the more pressing objections with which he is

faced.

5.4 Inductive Generalization

The positive case for moderate explanationism was started in 1965 by Harman, in his paper “The Inference to the Best Explanation”. Harman argued, in my terms, that inductive generalization and enumerative induction reduce to moderate explanationism—in the sense that all inductive inferential relations that could be construed as instances of either inductive generalization or enumerative induction should instead be construed as instances of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee.7

Harman’s main argument, when charitably construed, has just three immediate premises.

The first is the claim that all inductive inferential relations that could be construed as instances of

either inductive generalization or enumerative induction could also be construed as instances of

either inference to the best explainer, or inference to an explainee, or an explanatory inferential-

relation form built up out of inference to the best explainer and/or inference to an explainee. The

second is the claim that by doing so we could account for when such inferences are cogent. And

159

the third is the claim that at least some inductive inferential relations that could be construed as instances of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee could not also be construed as instances of either inductive generalization or enumerative induction.

There are several problems, of detail, with Harman’s argument. To mention just one,

Harman does nothing to make it plausible that all inductive inferential relations that could be construed as instances of either inductive generalization or enumerative induction could also be construed as instances of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee. Harman simply asserts it, as though its plausibility were transparent.

Nonetheless, Harman’s discussion is quite suggestive of how to argue for moderate explanationism. Suppose, to keep things fully general, that there is a rival account of cogency according to which there are four kinds of inductive inferential relations: IF1, IF2, IF3, and IF4.

Suppose further that, according to this account, an instance of IF1 is cogent just in case condition c1 is satisfied, an instance of IF2 is cogent just in case condition c2 is satisfied, an instance of IF3 is cogent just in case condition c3 is satisfied, and an instance of IF4 is cogent just in case condition c4 is satisfied. The strategy, then, would be to show, by means of examples, that c1, c2, c3, and c4 are satisfied just in case there is a virtuous explanatory inferential relation between the premise (or set of premises) in question and the conclusion in question, so that neither IF1 nor

IF2 nor IF3 nor IF4 is needed to account for the cogency, or lack thereof, of even one inductive inferential relation—thereby making each such argument form entirely superfluous.

I shall focus exclusively on inductive generalization in this section, employing the strategy outlined in the previous paragraph. Other inductive argument forms, such as argument from authority, argument from analogy, and statistical syllogism, will be addressed in the next section.

160

5.4.1 Lycan’s Improvements

Lycan’s discussion of inductive generalization improves on Harman’s discussion in at least two respects. First, Lycan considers both instances in which the sample percentage is 100, and instances in which the sample percentage is less than 100. Harman considers just instances in

which the sample percentage is 100. This is an important improvement, given the ultimate aim—

viz., establishing moderate explanationism (in one of its forms), which is a view about all cogent

inductive inferential relations. Second, Lycan argues (though quickly) where Harman merely

asserts, or suggests:

The besetting fallacies of [inductive generalization] are those of insufficient sample and biased sample. “Insufficient sample” is committed if not enough Xs have been observed. “Biased sample” is committed when we have antecedent reason to think that the proportion of Xs in our sample that are F differs from the proportion of all Xs that are F. I submit that the former fallacy is a fallacy because the suppressed assumption of best explanation fails: When a sample is grossly insufficient, the inductive conclusion is by no means the best explanation of the meager data. If I were to ask just two Carolina students what they thought of President Reagan’s support of the Nicaraguan Contras and the two split 50-50, the hypothesis that “Carolina students generally split 50-50 on the Contra issue” would hardly be a contender. That hypothesis has virtually no explanatory force in that context, even if true. I do not know what would explain our two-person data, depending on the researcher’s interests and purposes; I suspect that without elaborate stage setting and background assumptions, nothing would count as explaining it. But certainly the proposed statistical conclusion is not a candidate. The case of biased sample is even clearer. When we see that a sample is biased, we see precisely a better explanation of its distribution than the hypothesis that the general population is distributed that way. If only registered Republicans are consulted in a poll concerning approval of Reagan’s performance as president, the best explanation of a markedly positive rating is that Republicans tend to support Reagan, not that a marked majority of Americans do. (Lycan 1988, 179-80)

This too is important, in that it perfectly exemplifies the strategy outlined above—a strategy, let me stress, that is suggested quite straightforwardly by Harman’s discussion.

For further support, here are three more cases:

Case 1 Jack randomly picks three numbers from the phone book, calls each number, and asks for the answerer’s favorite color. The first and second answerers say blue. The third says green. On this basis, Jack then infers that roughly two thirds of the people listed in the phone book like blue the best.

161

Case 2 Over a span of ten years, Jack times a vast number of college track athletes from all over North America in the mile. It turns out that 65% of them run it in less than five minutes. Jack then “keenly” notes that they’re people from North America, and that together they’re a sample of the population of people from North America. On this basis, he infers that roughly 65% of all people from North America run the mile in less than five minutes.

Case 3 Jim shakes an opaque jar filled with 4,000 black and red cubes, reaches in without looking, and grabs 500. He then counts the reds, sees that he has 450, and on this basis infers that roughly 90% of the cubes in the jar are red.

In Case 1, the sample is too small, and so, according to the standard line on inductive generalization, the inferential relation is incogent. The same is true on moderate explanationism, since, intuitively, the population percentage’s being roughly two-thirds is in no way the best potential explainer for the sample percentage’s being two-thirds. In Case 2, the sample (though big enough) is blatantly biased, for certain properties positively relevant to running the mile in less than five minutes (e.g., exercising vigorously on a regular basis) are significantly over- represented in the sample, while certain properties negatively relevant to running the mile in less than five minutes (e.g., being a senior citizen) are significantly under-represented in the sample.

For this reason, the standard line on inductive generalization implies that the inferential relation is incogent. The same is true on moderate explanationism, given that, intuitively, the population percentage’s being roughly 65 is not the best potential explainer for the sample percentage’s being 65. In Case 3, the sample is both big enough and unbiased, and hence, on the standard line on inductive generalization, the inferential relation is cogent. But again, the same holds on moderate explanationism, in that, intuitively, the population percentage’s being roughly 90 is the best potential explainer for the sample percentage’s being 90—given the way in which Jim did the sampling.

162

5.4.2 Two Objections

Two potential worries should be addressed at this point, each of which pertains to the claim that the hypothesis that the population percentage’s being roughly 90 is the best potential explainer for the sample percentage’s being 90. The first worry is that what best explains why

90% of the sampled cubes are red is not that the population percentage is roughly 90, but that

90% of the sampled cubes reflect light in a certain way (or are disposed to cause certain experiences in normal observers in normal conditions, or whatever). The second worry is that the population percentage’s being roughly 90 is not even a potential explainer, let alone the best potential explainer, for the sample percentage’s being 90—given that the population percentage’s being roughly 90, even when coupled with the fact that the sample was drawn randomly, is perfectly compatible with a significantly different sample percentage. Even if the population percentage were exactly 90, Jim could have drawn just 100 red cubes (together with 400 black cubes)—in which case the sample percentage would have been just 20.

The proper response to the first worry is that the charge, when clarified, is correct, but

that it in no way casts doubt on moderate explanationism. The moderate explanationist is not

saying that the population percentage’s being roughly 90 explains why the red cubes in the

sample are red. Nor is the moderate explanationist saying that the population percentage’s being

roughly 90 explains why 90% of the cubes in the sample are red. The claim, instead, is that the

population percentage’s being roughly 90 explains why a sample 90% of which is red—in

contrast, for instance, to a sample 70% of which is red—was drawn. In other words, the claim is

that the reason why a sample 90% of which is red—and not a sample 70% of which is red, or a

sample 60% of which is red, etc.—was drawn is that the population percentage is roughly 90.

The second worry is a bit harder to deal with. The non-deductivist on explanation could simply accept the charge that the claim that the population percentage is roughly 90 in no way entails that Jim drew a sample the percentage of which is 90, and then reject the claim that,

163

because of this, the population percentage’s being roughly 90 in not a potential explainer for the sample percentage’s being 90. The deductivst, in contrast, has to invoke a gappy potential explainer involving the population percentage’s being roughly 90, in the same way that, with respect to the rain case in section 2, he has to invoke a gappy potential explainer involving its having just now rained. The immediate question, though, would be of what makes it such that the best potential explainer is the one involving the population percentage’s being roughly 90, and not one involving, say, the population percentage’s being roughly 70. Could the deductivist say anything principled in answer to this? Or would he have to rest content with his intuition?

I shall argue, in sub-section 5.4.5, that the deductivist (moderate explanationist) would benefit greatly by invoking Kitcher’s account of unification in explicating the notion of bestness, or loveliestness. The suggested answer will be that the gappy potential explainer involving the population percentage’s being roughly 90 would make for a higher degree of Kitcher-unification than would any potential explainer involving a different population percentage, so that, thus, the gappy potential explainer involving the population percentage’s being roughly 90 is the best, or the loveliest, of the available gappy potential explainers.

It will prove best to turn first to Fumerton’s objection to moderate explanationism, as it will be easiest to address the two objections together.

5.4.3 Fumerton’s Objection

Fumerton begins by describing a scenario in which a person, who is walking the beach, comes across some footprints and infers, via the argument below, that a person walked the beach recently:

(I) (1) There are footprints on the beach. (2) If a person walked the beach recently, then there would be footprints on the beach. ------(3) A person walked the beach recently.

164

The person’s inference, says Fumerton, is an instance of inference to the best explainer, and, adds

Fumerton, a perfectly legitimate instance at that.

Fumerton then compares (I) with two alternative arguments:

(II) (1) There are footprints on the beach. (2) If Jimmy Carter walked the beach recently, then there would be footprints on the beach. ------(3) Jimmy Carter walked the beach recently.

(III) (1) There are footprints on the beach. (2) If a cow wearing shoes walked the beach recently, then there would be footprints on the beach. ------(3) A cow wearing shoes walked the beach recently.

Each argument, says Fumerton, is like (I) both in being an instance of inference to the best explainer, and in having true premises. But unlike (I), we would reject (II) and (III) as illegitimate, or incogent.

And last, Fumerton uses this putative difference between (I), on one hand, and (II) and

(III), on the other hand, against moderate explanationism—arguing (a) that we think that footprints on beaches are correlated with people’s walking on beaches, (b) that we do not, in contrast, think that footprints on beaches are correlated with either Carter’s walking on beaches or shoe-wearing-cows’ walking on beaches, (c) that, with the correlation, (I) can be reconstrued as a straightforward instance of enumerative induction, and (d) that, since (I) is a paradigm instance of inference to the best explainer, we can safely generalize on what we say about (I):

Since as we just noted we cannot isolate the difference by looking at the form of the arguments and since each argument has true premises, I would suggest that we accept (I) because (I) is an enthymeme. We accept a crucial but unstated premise from which we can legitimately infer the conclusion. That premise is the obvious one:

2a) In the vast majority of known cases footprints are produced by men.

We reject (II) and (III) because we do not accept the relevant implicit premise. It is not true that in the vast majority of known cases footprints on the beach are produced by Jimmy Carter, nor is it true that in the vast majority of known cases such footprints are produced by a cow wearing shoes. But now if 2a) is an essential part of the evidence

165

from which we are willing to infer that a man walked the beach recently, it seems clear that the so-called argument to the best explanation is really just an inductive argument whose form would be put more perspicuously this way:

(1) In all (or most) cases in which we have observed footprints on the beach there were men present just prior to the existence of such footprints. (2) Here is another case of footprints on the beach. ------(3) A man was present just prior to the existence of these footprints.

The argument is a standard inductive argument having the following form:

(1) All or most of the A’s we have observed were immediately preceded by B’s. (2) This is an A. ------(3) It was preceded by a B.

The first premise asserts a discovered correlation between the occurrence of two states of affairs, the second premise asserts one is present and the conclusion asserts that the other is present. (Fumerton 1980, 592-3)

The upshot, argues Fumerton, is that “commonplace, paradigmatic cases of acceptable arguments to the best [explainer] must be considered enthymemes”, and that “when the suppressed premises are made explicit we have all of the premises we need to present either a straightforward

[enumerative-inductive] argument or an argument employing both [enumerative induction] and deduction”.

Fumerton’s conclusion comes much too quickly, in that even if it were true that

“commonplace, paradigmatic cases of acceptable arguments to the best [explainer] must be considered enthymemes” and that “when the suppressed premises are made explicit we have all of the premises we need to present either a straightforward [enumerative-inductive] argument or an argument employing both [enumerative induction] and deduction”, it would not follow that “[it is false that] there is a legitimate process of reasoning to the best [explainer] which can serve as an alternative to either straightforward [enumerative-inductive] reasoning or a combination of

[enumerative-inductive] and deductive reasoning”. For Fumerton has left open the possibility that there is a way of accounting for the role of his correlational premises from within the confines of inference to the best explainer, a way, that is, that in no way involves enumerative-

166

inductive reasoning. Nonetheless, Fumerton has certainly succeeded in foisting a tough question on the moderate explanationist—a question that, if left unanswered, would be a serious problem for moderate explanationism.

The answer, I shall argue, is that yes, the moderate explanationist can make perfect sense of the role of Fumerton-type correlations in explanatory inferences. The thing to see, I shall suggest, is that Fumerton-type correlations serve not as immediate premises in instances of inference to the best explainer, but as immediate premises in support of the immediate premises in instances of inference to the best explainer. Consider:

(1) q obtains. (2) p’s obtaining is a potential explainer for q’s obtaining. (3) With respect to the available potential explainers for q’s obtaining, p’s obtaining is the best. ------(4) p obtains.

I shall suggest, in particular, that a Fumerton-type correlation between phenomena like p and phenomena like q would play a direct role not in support of (4), but in support of (3)—viz., the claim that p’s obtaining is the best of available potential explainers.

5.4.4 Kitcher-Unification

Kitcher starts with the notion of an argument pattern, saying that an argument pattern is an ordered triple the first member of which is a schematic argument, the second member of which is a set of filling instructions, and the third member of which is a classification. Here is an example:

(1) All P’s are Qs. (2) α is a P. (3) α is a Q. Filling Instructions: P must be replaced with a predicate; Q must be replaced with a predicate; α must be replaced with the name of a person. Classification: (1) and (2) are premises; (3) follows from (1) and (2) by universal instantiation and modus ponens.

167

The filling instructions, notes Kitcher, determine its stringency, which, in effect, is a measure of how hard it is to be an instantiation of it.

Kitcher next moves to the notion of an explanatory store, which he unpacks via a series of definitions:

If K is the set of sentences accepted by the scientific community, then S is a systematization of K just in case S is a set of derivations each part of which is a member of K.

If S is a systematization of K, then S is acceptable just in case each derivation in S is valid.

If S is a systematization of K, then G is a generating set for S just in case G is a set of argument patterns such that each derivation in S instantiates an argument pattern in G.

If S is an acceptable systematization of K, if G1 and G2 are the generating sets for S, and if G2, say, better unifies K than does G1, then G2 is the basis for S.

If S1 and S2 are the acceptable systematizations of K, if G1 is the basis for S1, if G2 is the basis for S2, and if G2 better unifies K than does G1, then S2 is the explanatory store over K.

The view, in short, is that the explanatory store over K is the set of derivations that best unifies, or

systematizes, the sentences in K—where the set of derivations that does this is the one the basis

of which is the highest in unification.

But what about unification? What fixes the degree to which a basis G unifies, or is unified? Kitcher points to just three variables:

(1) The size of G’s conclusion set, which is comprised of the conclusions of the derivations in its systematization S of K. (2) The stringency of G’s argument patterns. (3) G’s size (i.e., the number of argument patterns in G).

The degree to which G unifies, according to Kitcher, varies directly with the size of G’s

conclusion set, directly with the stringency of G’s argument patterns, and inversely with G’s size.

It might help to give an example from Kitcher. Consider the following derivation:

Derivation DH (1) For all x, if (a) x was a sample of salt, (b) x was hexed, and (c) x was placed in water, then x dissolved. (2) b was a sample of salt, b was hexed, and b was placed in water. (3) b dissolved.

168

Classification: (1) and (2) are premises; (3) follows from (1) and (2) by universal instantiation and modus ponens.

Suppose that S’ is a systematization of K, that DH is a member of S’, that G’ is the basis for S’, and that DH is an instantiation of argument pattern PH:

Argument Pattern PH (1) For all x, if (a) x was a β, (b) x was hexed, and (c) x was placed in φ, then x χ-ed. (2) α was a β, α was hexed, and α was placed in φ. (3) α χ-ed. Filling Instructions: β must be replaced with the name of a substance; φ must be replaced with the name of a substance; χ must be replaced with a description of an event; α must be replaced with the name of a sample of a substance. Classification: (1) and (2) are premises; (3) follows from (1) and (2) by universal instantiation and modus ponens.

Suppose further that S is an alternative systematization of K, that DH is not a member of S, that G

is the basis for S, and that PS, but not PH, is a member of G:

Argument Pattern PS (1) For all x, if (a) x was a β and (b) x was placed in φ, then x χ-ed. (2) α was a β and α was placed in φ. (3) α χ-ed. Filling Instructions: β must be replaced with the name of a substance; φ must be replaced with the name of a substance; χ must be replaced with a description of an event; α must be replaced with the name of a sample of a substance. Classification: (1) and (2) are premises; (3) follows from (1) and (2) by universal instantiation and modus ponens.

Then, given these suppositions, G would be greater in unification than G’. If, on one hand, PS is

not a member of G’, then the conclusion set of G’ would be smaller in size than the conclusion set

of G—since whereas G (via PS) would generate derivations of both the dissolving of hexed

samples of salt and the dissolving of unhexed samples of salt, G’ (having just PH) would generate

derivations of only the dissolving of hexed samples of salt. And if, on the other hand, PS is a member of G’, then, though G and G’ would have identical conclusion sets, G would be smaller in size than G’—given that whereas PS would be a member of both G and G’, PH would be a

member of just G’.

169

5.4.5 The Solution

My proposal, as I said above, is that the explanationist should use Kitcher’s account of unification as an account of explanatory virtue. The explanationist should say that if S believes that e, if h and h’ are the only available potential explanations of why-e, and if adding the belief that h would make S’s belief system more Kitcher-unified than would adding the belief that h’, then h is the best of the available explanations.

Suppose that you believe (1) that all observed Gs are Fs, (2) that b1 is an F and that its

being a G is due to its being an F, (3) that b2 is an F and that its being a G is due to its being an F,

. . ., (n) that bn is an F and that its being a G is due to its being an F, and (n+1) that bn+1 is a G.

Should you infer that bn+1 is an F? Or, instead, should you infer that bn+1 is an H, which, suppose, would explain why it is a G? My proposal implies that (other things being equal) you should infer that bn+1 is an F, given that adding the belief that bn+1 is an F would make your belief system

more Kitcher-unified than would adding the belief that bn+1 is an H. Why? Because the basis for

your belief system would be smaller in size if you added the belief that bn+1 is an F than if you

added the belief that bn+1 is an H—since, if you added the belief that bn+1 is an H, the basis for your belief system would involve both an argument pattern for the derivations of G-ness from F- ness and an argument pattern for the derivations of G-ness from H-ness, whereas, if you added the belief that bn+1 is an F, the basis for your belief system would involve just an argument pattern for the derivations of G-ness from F-ness. Here, then, is an example in which—using Kitcher’s account of unification as an account of explanatory loveliness—a Fumerton-type correlation (i.e., that all observed Gs are Fs) helps in fixing which of the available potential explanations is the best.

But what about statistical correlations, of the form “Most observed Gs are Fs.”? Can the explanationist give a similar treatment of them? It would seem so. Suppose that you believe (1) that most observed Gs are Fs, (2) that b1 is an F and that its being a G is due to its being an F, . . .,

170

(51) that b50 is an F and that its being a G is due to its being an F, (52) that b51 is an H and that its being a G is due to its being an H, . . ., (56) that b55 is an H and that its being a G is due to its

being an H, and (57) that b56 is a G. Should you infer that b56 is an F? Or, instead, should you infer that b56 is an H? My proposal again implies that (other things being equal) you should infer that b56 is an F. The reason why is that your belief system would better approximate maximal

Kitcher-unification if you added the belief that b56 is an F than if you added the belief that b56 is

an H—given that it would better approximate a belief system according to which b1 is an F and its

being a G is due to its being an F, according to which b2 is an F and its being a G is due to its

being an F, . . ., and according to which b56 is an F and its being a G is due to its being an F. Thus

the explanationist could in this fashion make sense of the importance of statistical correlations in

explanatory inferences.

Imagine, for instance, that your thermostat says that it is 72 degrees, and that your thermostat has had an extremely high, though not perfect, track-record. Should you infer that it is, in fact, 72 degrees, and that this is why your thermostat says that it is 72 degrees? Or should you instead infer, say, that it is 75 degrees, and that your thermostat is malfunctioning? My proposal implies, quite intuitively, that the 72-degrees hypothesis is the better of the two potential explanations, given that your belief system would better approximate maximal Kitcher- unification if you were to accept the 72-degrees hypothesis than if you were to accept the 75- degrees hypothesis. For the closest belief system (to your belief system) with maximal Kitcher- unification says that each of your thermostat’s past readings was correct and that, in addition, each such reading was explained by the actual temperature at the time in question.

Think back, now, to the worry according to which it is not the case that the population percentage’s being roughly 90 potentially explains why you drew a sample the percentage of which is 90. I said that, to answer this worry, the deductivist explanationist has to invoke a gappy potential explanation involving the population percentage’s being roughly 90, and argue that it is the best of the available potential explanations. I then said that, though effective in answering the

171

worry, this line of response would invite the hard question of what it is that makes the gappy potential explanation involving the population percentage’s being roughly 90 the best. And then last, I said that, to answer this question, the deductivist explanationist should adopt Kitcher’s account of unification as an account of explanatory virtue. What I had in mind, in saying this, is perfectly parallel to what I said above about statistical correlations and explanatory inferences.

Given that, in the past, population percentages and sample percentages have almost always been roughly equivalent, and given that (to oversimplify a bit) in each case of success the population percentage figured in why the sampler got a sample the percentage of which is roughly equal to the population percentage, the gappy potential explanation that would best Kitcher-unify your belief system is the one involving the population percentage’s being roughly 90. By adding the belief that the population percentage is roughly 90 and that this is partly why you got a sample the percentage of which is 90, your belief system would better approximate a belief system with maximal Kitcher-unification—viz., a belief system that says that population percentages and sample percentages have always been roughly equivalent, and that in each such case the population percentage figured in why the sampler got a sample the percentage of which is roughly equal to the population percentage.

5.5 Other Inductive Argument Forms

The project, remember, is to undermine the view on which the challenge of explaining what makes cogent inductive inferential relations cogent is answered by appeal to inductive argument forms such as statistical syllogism, argument from authority, and argument from analogy. The strategy is to show, by means of examples, that the cogency (or lack thereof) of a cogent inductive inferential relation can be accounted for by this commonplace alternative account only if it can also be accounted for by moderate explanationism.

Henceforth I will ignore worries analogous to the worry that the population percentage’s

being roughly 90 is not even a potential explainer, let alone the best potential explainer, of the

172

sample percentage’s being 90. The responses to the analogous worries would be analogous to the response I gave in answer to it.

5.5.1 Argument From Authority

Argument from authority (AFAu) can be construed as follows:

(1) S says that p. (2) S is an expert. ------(3) p.

In arguing for p (or in inferring that p), the arguer (or inferrer) appeals to an authority (i.e., an expert who says that p).

Harman says that appeals to authority can be construed as instances of inference to the best explanation. The subject infers that S believes that p on the grounds that S’s believing that p is the best explanation of why he says that p, and from there the subject infers that p is true on the

grounds that p’s being true is the best explanation of why S believes that p. I find this quite

suggestive, and shall defend the thesis that an instance of AFAu is cogent just in case both (a) S’s

believing that p is the best potential explainer for S’s saying that p and (b) p’s being true is the

best potential explainer for S’s believing that p.

First, some (if not all) theorists writing on AFAu require for cogency that S be an expert in the right field.8 If, for example, p is a claim in biology, then S needs to be an expert in biology; otherwise, that he says that p would in no way serve as a good inductive reason, or even as part of a good inductive reason, for thinking that p. I agree with such theorists. When S is not an expert in the right field, it is not the case that p’s being true is the best potential explainer for

S’s believing that p (unless, of course, S’s belief that p is itself based on a cogent appeal to authority).

Second, some theorists require for cogency that there be general agreement on p among

the experts in the field in question.9 I agree, to a certain extent. When there is widespread

173

disagreement on p among the experts in the field in question, then either p’s being true is not the best potential explainer for S’s believing that p, or p’s being true is not good enough as a potential explainer. I say “to a certain extent” because of cases such as the following:

Jim, who is a professional logician, makes up an argument off the top of his head, and then tells his class that it is valid (though weird). His students then infer on this basis (together with the fact that he is a professional logician) that the argument, in fact, is valid. But as it turns out, no other professional logician has ever even entertained the question of whether it is valid—let alone said that it is valid.

The inference from Jim’s testimony (together with his expertise in logic) to the claim that the argument is valid is cogent. However, it is not the case that there is general agreement that the argument is valid among the experts in logic—since only one such expert has ever even considered the claim that the argument is valid. The requirement, then, should be modified, requiring that there not be widespread disagreement on whether p is true, and that there would be general agreement that p is true were the experts to consider the question of whether p is true.10,11

Third, some theorists require for cogency that S be properly situated vis-à-vis whether p is true. Robert Fogelin and Walter Sinnott-Armstrong put the point in terms of evidence:

Even in those cases in which it is clear that the person cited is an expert in the field, we can still ask whether the question is of the kind that can now be settled by an appeal to experts. It is important to raise this question, because sometimes the best experts simply get things wrong. For example, in 1932 Albert Einstein, who was surely an expert in the field, declared that “there is not the slightest indication that [nuclear] energy will ever be obtainable. It would mean that the atom would have to be shattered at will.” Just a year later, the atom was, in fact, split. Even so, a leading British physicist, Ernest Lord Rutherford, insisted that the splitting of the atom would not lead to the development of nuclear power, saying, “The energy produced by the atom is a very poor kind of thing. Anyone who expects a source of power from the transformation of these atoms is talking moonshine.” Given the knowledge available at the time, both Einstein and Rutherford may have been justified in their claims, but their assertions were, after all, more speculations than scientifically supported statements of fact. The lesson to be learned from this is that the best experts are sometimes fallible, and become more fallible when they go beyond established facts in their discipline to speculate about the future. (Fogelin and Sinnott-Armstrong 2001, 397)

Patrick Hurley, in contrast, puts the point in terms of abilities. He considers the argument “Old

Mrs. Furguson (who is practically blind) has testified that she saw the defendant stab the victim

174

with a bayonet while she was standing in the twilight shadows 100 yards from the incident.

Therefore, members of the jury, you must find the defendant guilty.”, commenting as follows:

Here the witness lacks the ability to perceive what she has testified to, so her testimony is untrustworthy. (Hurley 2003, 131)

I agree. When there is a real worry about S’s being properly situated vis-à-vis whether p is true, then either p’s being true is not the best potential explainer for S’s believing that p, or p’s being true is not good enough as a potential explainer; p’s being true either fails to beat out the potential explainer according to which S is speculating (first kind of case), or fails to beat out the potential explainer according to which S is misperceiving (second kind of case).

Fourth, some theorists require for cogency that S be unbiased—that is, that S not have a special interest in p’s being true.12 I concur, for the most part. I agree that when there is a real worry about S’s being biased, then the inferential relation is incogent. For when there is such a worry, p’s being true is not the best potential explainer for S’s believing that p; p’s being true fails to beat out the potential explainer according to which S is biased. I disagree, though, that when S has a special interest in p’s being true, the inferential relation is incogent. If, for example,

S is aware that he is baised and in light of this goes to extra lengths to ensure that the facts support p, then there is no real worry about his being biased and thus about his believing that p regardless of whether p is true; the circumspection trumps the motive. Hence, the requirement shoud be not that S not have a special interest in p’s being true, but that there not be a worry about S’s having a special interest in p’s being true.

And fifth, some theorists require for cogency that S not have a motive to lie.13 Hurley,

for instance, considers the argument “James W. Johnston, Chairman of R. J. Reynolds Tobacco

Company, testified before Congress that tobacco is not an addictive substance and that smoking

cigarettes does not produce any addiction. Therefore, we should believe him and conclude that

smoking does not in fact lead to any addiction.”, arguing that it is incogent since Johnston has a

motive to lie:

175

If Mr. Johnston had admitted that tobacco is addictive, it would have opened the door to government regulation, which could put his company out of business. Thus, because Johnston had a clear motive to lie, we should not believe his statements. (Hurley 2003, 131)

I agree, but (again) with a small qualification. I agree that when there is a real worry about S’s being motivated to lie, the inferential relation is incogent. This follows from the fact that when there is such a worry, S’s believing that p is not the best potential explainer for S’s saying that p;

S’s believing that p fails to beat out the potential explainer according to which S is lying. I disagree, though, that when S has a motive to lie, the inferential relation is incogent. If S has the utmost in integrity, then even if he has a motive to lie, there is no real worry about his being motivated to lie; the integrity trumps the motive. Hence, the requirement shoud be not that S not have a motive to lie, but that there not be a worry about S’s being motivated to lie.

The upshot, therefore, is that argument from authority, like inductive generalization, is unnecessary for accounting for the cogency, or lack thereof, of inductive inferential relations.

When an inferential relation can be construed as a cogent instance of argument from authority, it can also be construed as a cogent explanatory inferential relation. And when, on the other hand, an inferential relation can be construed as an incogent instance of argument from authority, it can also be construed as an incogent explanatory inferential relation.14

5.5.2 Randomized Experimental Studies

A randomized experimental study, simply put, has four key steps. First, the scientist gets a big unbiased sample of the population. Second, he randomly divides the sample into two groups, called “the experimental group” and “the control group”. Third, he exposes the experimental group, but not the control group, to X, which is the suspected cause. And fourth, he looks for differences between the two groups in terms of Y, which is the suspected effect. When

176

and only when there are significant differences, the scientist concludes that X causes Y. (This description is highly simplified, obviously.)

Consider, for example, the following excerpt from a report by the Congressional Office of Technology on a study in Canada in the 70s:

The positive feeding experiments were conducted over two generations. Rats of the first generation were placed on diets containing saccharin at the time of weaning. These animals were bred while on this diet, and the resulting offspring were fed saccharin from the moment of conception until the termination of the experiment. Each animal of the second generation was examined for cancers at its death or at its sacrifice after 2 years on the experiment. Each experiment had appropriate control groups that did not ingest saccharin. Compared to control animals, the saccharin-fed animals showed an excess of bladder tumors. These differences were sufficiently convincing to lead to the conclusion that saccharin caused cancer in rats.

The results, in table-form, are as follows:

Generation Dose Cancer 1st 0% 1/74 1st 5% 7/78 2nd 0% 0/89 2nd 5% 14/94

Consider, for illustration, just the 1st generation rats. Presumably they were an unbiased sample of the population of rats, and presumably the researchers divided them into the experimental group (the group on diets with 5% saccharin) and the control group (the group on diets with 0% saccharin) via a random process. And notice, there is a big difference (between the two groups) in terms of cancer: 9% of the rats in the experimental group contracted cancer, compared with only 1.3% of the rats in the control group.15

This example is typical of randomized experimental studies (in which the scientist concludes that X causes Y). Some of the subjects in the control group have Y, thus showing that

X is not necessary for Y. Some of the subjects in the experimental group do not have Y, thus showing that X is not sufficient for Y. And nonetheless the scientist concludes that X causes Y.

This would be a problem for the deductivist moderate explanationist were he unable to invoke gappy potential explainers. Given that there are subjects with X but not Y, he could not

177

account for the goodness of the scientist’s reasoning by saying, in part, that the scientist thinks that the data is best explained by the hypothesis that X causes Y—where causation is supposed to be a species of explanation. For it is clearly not the case, given deductivism, that X causes Y in that sense, since it is clearly not the case, given deductivism, that X is a potential explainer for Y.

Suppose, for the sake of illustration, that there are just three potential explainers for getting lung cancer:

(1) smoking cigarettes + B + C (2) smoking cigarettes + D + E (3) working in a coal mine + F + G

Suppose further that we get a big unbiased sample of humans, and that we randomly divide it into two groups. And suppose, finally, that we then introduce a difference: we make the subjects in the first group smoke cigarettes. What would the results look like, given these suppositions?

First, it would be highly likely that (a) the groups have roughly the same number of subjects with B and C, (b) the groups have roughly the same number of subjects with D and E, and (c) the groups have roughly the same number of subjects who work in a coal mine and have F and G. After all, we did the dividing randomly. And second, it would be highly likely that there are a lot more subjects in the experimental group with lung cancer than subjects in the control group with lung cancer. Since the subjects in the experimental group smoked cigarettes while the subjects in the control group did not, the subjects in the experimental group with B and C would get lung cancer while the ones in the control group would not. For the same reason, the subjects in the experimental group with D and E would get lung cancer while the ones in the control group would not. And since the two groups have roughly the same number of subjects who work in a coal mine and have F and G, the numbers there would for the most part cancel out. The results, then, would probably be a relatively high percentage of lung cancer in the experimental group and a relatively low percentage of lung cancer in the control group.

Now suppose instead that, not knowing anything about whether smoking cigarettes causes lung cancer, we run such an experiment and get such results. What would follow?

178

The first thing is that the putative fact that smoking cigarettes causes lung cancer—in the sense that smoking cigarettes is an essential part of a larger phenomenon that is a potential explainer for lung cancer—is itself an essential part of a potential explainer for the results. For if smoking cigarettes caused lung cancer in such a sense, and if the other factors in the potential explainer in question were evenly distributed between the experimental group and the control group, then the percentage of lung cancer in the experimental group would be significantly higher than the percentage of lung cancer in the control group.

The second thing that would follow—given that the subjects were randomly divided into the experimental group and the control group—is that the smoking-cigarettes-causes-lung-cancer hypothesis is superior to, or lovelier than, the coincidence hypothesis, according to which the difference in percentage is explained not by the difference in smoking cigarettes but by something else which just happened to be present in the experimental group to a significantly higher degree than it was present in the control group. That the subjects were randomly divided into the experimental group and the control group makes it likely that for every property relevant to lung cancer, the degree to which it is present in the experimental group is roughly equal to the degree to which it is present in the control group.

The overall lesson is that moderate explanationism can make perfect sense of the goodness of the scientist’s reasoning in randomized controlled studies. The putative fact that X causes Y—in the sense that X is an essential part of a larger phenomenon that is a potential explainer for Y—is itself an essential part of a potential explainer for why the percentage of Ys in the experimental group is significantly higher than the percentage of Ys in the control group.

And—given that the subjects were randomly divided into the experimental group and the control group—the X-causes-Y hypothesis is superior to, or lovelier than, the coincidence hypothesis, so that the best potential explainer for the results is that X causes Y.

179

5.5.3 Mill’s Methods

Mill’s method of difference (MD) is often explained via examples such as the following:

Jim and Jerry had lunch together at Bob’s Diner on Friday, and then later Jim, but not Jerry, got food poisoning. They had the same things from the menu, with one exception: Jim had the egg salad, whereas Jerry didn’t.

The only relevant respect in which the cases differ, the idea goes, is the presence of egg salad. In the case in which food poisoning is present, so is egg salad. In the case in which food poisoning in not present, neither is egg salad. Hence, the cause of Jim’s food poisoning is the egg salad.

The analogous case for Mill’s method of agreement (MA) would then be this:

Jim, Jack, Jill, and Jane had lunch together at Bob’s Diner on Friday, and then later got food poisoning. They had different things from the menu, with one exception: they all had the egg salad.

The only relevant respect common to all the cases is the presence of egg salad, and thus, in all probability, the egg salad is the cause of the food poisoning. Can these kinds of inferences be accounted for by the moderate explanationist?

MD, for starters, is fairly easy to deal with. Suppose that there is a potential explainer for food poisoning that involves a certain kind of egg salad, suppose that it was present in the first imagined scenario above, and suppose that no other potential explainer for food poisoning was present. Then it would follow that Jim, but not Jerry, would get food poisoning—since Jim, but not Jerry, had the egg salad. With no other relevant differences, no other putative potential explainers could account for the data. And thus it follows in all probability (if anything follows) that there is a cause of food poisoning that involves a certain kind of egg salad (in the sense that there is a potential explainer for food poisoning of which a certain kind of egg salad is an essential part), and that the cause of Jim’s food poisoning in particular is the egg salad in question

(in the sense that the egg salad in question together with a certain batch of other phenomena brought about the food poisoning).

180

Suppose instead that, in addition to the egg-salad difference, Jim, but not Jerry, had the pasta salad. Then, intuitively, the inference would be incogent. This too makes sense given

moderate explanationism. With this second relevant difference, there would be an alternative,

and equal, hypothesis that can explain the data—viz., that there is a potential explainer for food

poisoning that involves a certain kind of pasta salad, that it was present in the scenario in

question, and that no other potential explainer for food poisoning was present. The egg-salad

hypothesis would no longer be the best potential explainer.

Now for MA. Suppose that there is a potential explainer for food poisoning that involves a certain kind of egg salad, that it was present in the second imagined scenario above, and that no other potential explainer for food poisoning was present. Then it would follow that Jim, Jack,

Jill, and Jane would each get food poisoning—given that Jim, Jack, Jill, and Jane each had the egg salad. But would it follow that the egg-salad hypothesis is the best potential explainer for the data? What about the coincidence hypothesis, or the multiple-cause hypothesis? The egg-salad hypothesis would win, since—intuitively—it is the simplest, or the most unifying. And so, again, moderate explanationism can account for the goodness of the inference.

Now imagine instead that the pasta salad too is common to the cases. Then, intuitively,

the inference would be incogent. This too makes sense on moderate explanationism. With this

second relevant similarity, there would be an alternative, and equal, hypothesis that can explain

the data—viz., that there is a potential explainer for food poisoning that involves a certain kind of

pasta salad, that it was present in the scenario in question, and that no other potential explainer for

food poisoning was present. The egg-salad hypothesis would no longer be the best of available

potential explainers.

Mill’s other methods could be handled in similar fashion, with the end result being that

Mill’s methods are unnecessary for accounting for the cogency, or lack thereof, of inductive inferential relations. When an inferential relation can be construed as a cogent instance of one of

Mill’s methods, it can also be construed as a cogent explanatory inferential relation. When, on

181

the other hand, an inferential relation can be construed as an incogent instance of one of Mill’s methods, it can also be construed as an incogent explanatory inferential relation.

5.5.4 Argument From Analogy

Argument from analogy (AFAn) can be construed as follows:

(1) X1, . . ., Xn and Y are the same in terms of m1, . . ., mn. (2) X1, . . ., Xn have n. ------(3) Y has n.

X1, . . ., Xn and Y are the analogues, with X1, . . ., Xn being the premise analogues and Y being the conclusion analogue. m1, . . ., mn and n, in turn, are the analogies, with m1, . . ., mn being the asserted analogies and n being the inferred analogy.

It is said that the cogency of an instance of AFAn varies directly with things such as the

following:

(a) The degree to which the asserted analogies m1, . . ., mn are relevant to the inferred analogy n. (b) The number of premise analogues X1, . . ., Xn. (c) The number of asserted analogies m1, . . ., mn.

It is also said that the cogency of an instance of AFAn varies indirectly with things such as these:

(d) The number of relevant disanalogies between the premise analogues X1, . . ., Xn and the conclusion analogue Y, as well as the degree of the disanalogies. (e) The number of things with the asserted analogies m1, . . ., mn but not with the inferred analogy n. (f) The number of relevant similarities between the premise analogues X1, . . ., Xn.

Thus when things such as the things referred to (a)-(c) are high enough, and when things such as

the things referred to in (d)-(f) are low enough, then, according to the standard picture, the

inferential relation in question is cogent.

What you would find with cogent instances of AFAn is that there is a lovely potential

explanatory relation between Y’s having m1, . . ., mn and Y’s having n. That is, what you would find is that either Y’s having n explains Y’s having m1, . . ., mn (as in inference to an explainer),

182

or Y’s having m1, . . ., mn explains Y’s having n (as in inference to an explainee), or Y’s having

m1, . . ., mn and Y’s having n have a common explainer. In contrast, what you would find with incogent instances of AFAn is that there is not a lovely potential explanatory relation between

Y’s having m1, . . ., mn and Y’s having n. The other things involved—in the cogent cases—figure

only indirectly in fixing the cogency facts, by helping in fixing the loveliness facts.

Here is an example:

(1) Bob, just like his dog, lives in North America. (2) Bob’s dog barks at the mailman. ------(3) Bob barks at the mailman.

The asserted analogy (i.e., living in North America) is irrelevant to the conclusion analogy (i.e.,

barking at the mailman), and so—given condition (a)—the argument is incogent. And notice,

there is not a lovely potential explanatory connection between Bob’s living in North America and

Bob’s barking at the mailman—at least relative to the facts given. Thus moderate explanationism

gives the right result.

With condition (e), things are similar. When there are lots of things with m1, . . ., mn but not with n, the potential explanatory connection, if there is one, between Y’s having m1, . . ., mn

and Y’s having n is less than lovely. This is why cogency varies indirectly with the number of

things having m1, . . ., mn but not having n.

With (f), things are reversed. When there are few relevant similarities between the premise analogues, the potential explanatory connection between Y’s having m1, . . ., mn and Y’s having n is quite lovely. Think of cases in which Mill’s method of agreement would apply as limiting cases. This is why cogency varies indirectly with the number of relevant similarities between the premise analogues.

Consider, for example, the following arguments:

(1) Jim’s old pair of running shoes, Tom’s old pair of running shoes, and Henry’s old pair of running shoes were made by brand S, had design D, fit well, lasted a long time, and gave excellent support. (2) Jim’s new pair of running shoes was made by S, has D, and fits well.

183

------(3) Jim’s new pair of running shoes will last a long time and give excellent support.

(1) Jack’s old pair of running shoes, Poky’s old pair of running shoes, and Speedy’s old pair of running shoes were made by brand T, had design E, fit well, lasted a long time, and gave excellent support. (2) Jack’s new pair of running shoes was made by T, has E, and fits well. ------(3) Jack’s new pair of running shoes will last a long time and give excellent support.

And consider the following scenario:

Jack’s old pair, Poky’s old pair, and Speedy’s old pair were made of the same material, were used on the same kind of terrain, and were used in the same kind of weather. In contrast, Jim’s old pair, Tom’s old pair, and Henry’s old pair were made of different materials, were used on different kinds of terrain, and were used in different kinds of weather.

Given this further information, there are more relevant disanalogies between the primary analogues in the first argument than between the primary analogues in the second. And because of this, the first argument is better than the second. But, again, notice that the disanalogies support the claim that the reason why, causally speaking, Jim’s old pair, Tom’s old pair, and

Henry’s old pair lasted a long time and gave excellent support is that they were made by S, had

D, and fit well. In other words, they support the claim that if a pair of shoes were to have been made by S, have D, and fit well, then it would last a long time and give excellent support. And this, in turn, further supports, by way of inference to an explainee, the claim that Jim’s new pair will last a long time and give excellent support.

The other conditions can be treated in a similar fashion, so that AFAn, thus, is unnecessary for accounting for the cogency, or lack thereof, of inductive inferential relations.

When an inferential relation can be construed as a cogent instance of AFAn, there is a lovely explanatory connection between the premise and the conclusion. When an inferential relation can be construed as an incogent instance of AFAn, there is no such lovely explanatory connection.

184

5.5.5 Statistical Syllogism

Statistical syllogism (SS), in general terms, is standardly put like this:

(1) XS% of Fs are Gs. (2) b is an F. ------(3) b is a G. Or in the negative case:

(1) XS% of Fs are Gs. (2) b is an F. ------(3) b is not a G.

“S” in “XS” stands for “statistical”, so that X is below 100 and above 0.

Whether an instance of SS is cogent, it is usually said, hinges on two things. The first is whether the percentage is strong enough. In the positive case, it needs to be close enough to 100.

In the negative case, it needs to be close enough to 0. The second thing, in turn, is whether b is a typical F with respect to G—in that an instance of SS is incogent if b has further properties making it such that, even with F, it is not highly likely that he is a G.

Here is an example:

(1) 90% of Americans will net less than $100,000 next year. (2) Tiger Woods is an American. ------(3) Tiger Woods will net less than $100,000 next year.

90 is close enough to 100, and so the percentage is strong enough. The premises, though, do not make it highly likely that the conclusion is true. For Woods is not just any old American with respect to netting less than $100,000 next year. He plays on the PGA Tour, he is ranked #1 in the world golf rankings, he has an endorsement deal with Nike, and so on. And these differences, of course, make a difference.

Lycan considers a slight variant of a measles case from Robert Ennis, and then contrasts his response to it with an analogous response to another case from Ennis—a response that

185

involves SS.16 Lycan’s discussion will thus serve as a nice entry into the difficulties pertaining to

SS.

The slight variant is this:

(1) All observed cases in which an MD says a child will come down with the measles are cases in which the child subsequently comes down with the measles. (2) An MD just said that Jimmy will come down with the measles. ------(3) Jimmy will come down with the measles.

Lycan argues that the reasoning involves a tacit piece of deduction. The reasoning runs from (1) to the claim that all cases in which an MD says a child will come down with the measles are cases in which the child subsequently comes down with the measles (hereafter (4)), and then from (4) and (2) to (3). The inferential relation from (1) to (4) is inductive and explanatory, while the one from (4) and (2) to (3) is deductive. So the variant, concludes Lycan, in no way involves a good but non-explanatory inductive inferential relation.

The other case from Ennis runs as follows:

(1) All observed cases in which a control stick is pulled back are cases in which the air speed subsequently decreases. (2) The control stick has just been pulled back. ------(3) The air speed will decrease.

Lycan starts by modifying the first premise, since, strictly speaking, it is false:

(1) Most observed cases in which a control stick is pulled back are cases in which the air speed subsequently decreases. (2) The control stick has just been pulled back. ------(3) The air speed will decrease.

Lycan then says that, with this modifcation, the tacit-deduction response is not available, as the reasoning, more fully spelled out, would go like this:

(1) Most observed cases in which a control stick is pulled back are cases in which the air speed subsequently decreases. (2) That most cases in which a control stick is pulled back are cases in which the air speed subsequently decreases is the best potential explainer for why most observed cases in which a control stick is pulled back are cases in which the air speed subsequently decreases. ------

186

(3) Most cases in which a control stick is pulled back are cases in which the air speed subsequently decreases. (4) The control stick has just been pulled back. ------(5) The air speed will decrease.

The problem is that the move from (3) and (4) to (5) is an instance of SS, and thus is not deductive. The question, then, is whether the moderate explanationist can account for instances of SS such as this.

The obvious thing to say, as a moderate explanationist, is that the move to (5) is an instance of inference to an explainee. It is right that the decrease in air speed will not explain the observed regularity. And it is right that the decrease in air speed will not explain the pulling. But it is not right that the inferential relation is non-explanatory. The pulling will explain the decrease in air speed (with the stuff about the observed correlation helping to make it such that this potential explanatory relation is lovely enough). Problem solved.

In fact, this is true of all cogent instances of SS. Either b’s being an F is best explained by b’s being a G, or b’s being a G is best explained by b’s being an F, or there is a common explainer for b’s being a G and b’s being an F. And at least part of what makes the potential explanatory relation in question lovely enough is the statistical regularity—in the same way in which, as I argued above, the claim that most observed Gs are Fs can figure in making it such that b56’s being an F is a better potential explainer of b56’s being a G than is b56’s being an H.

Lycan, however, cannot give this response, since he rejects inference to an explainee.

Here is what he says against Harman’s response to the control-stick case, which is essentially the

response given in the previous paragraph:

Harman thinks that we sometimes do explanatory inference in reverse and from A “infer A explains B and then infer B” (p.530). That is, he says, in some cases of “inference to the best explanatory statement” we infer the truth of the explanandum as our final conclusion, rather than (as in the standard case of enumerative induction) the truth of the explanans. And he maintains that this is what is going on in the control-stick example. Fully spelled out, the reasoning about the control stick would take this form: 1. 99 percent (say) of As are Bs. ∴

187

2. The fact that 99 percent of As are Bs explains the next A’s being a B (from step 1, by “inference to the best explanatory statement”). ∴ 3. The next A will be a B (from step 2 by deduction, since “explains,” in step 2, creates a factive context). . . . My main objection to Harman’s ploy . . . concerns the factivity of step 2 (i.e., the fact that 2 entails and/or presupposes the truth of 3). Step 2 does not mean simply that 1 “would explain” the next A’s being a B, in our sense (= “would explain if true”), for then 3 would not follow; “explain,” in 2, must be taken as a success-verb. But then we cannot infer 2 from 1 without already presupposing 3, and in this way the inference begs the question. One might try to say that 3 is not antecedently presupposed but is warranted by 1 at the same time 2 is, but the direct warranting of 3 by 1 is evidently just statistical syllogism, and we are back where we started. (Lycan 1988, 184-6)

The putative shortcoming is that it (i.e., inference to an explainee) is either question-begging, or nothing over and above SS—so that it in no way helps in squaring moderate explanationism with cases such as the control-stick case.17

But this is just not true, at least as I am thinking of inference to an explainee. The first claim in an instance of inference to an explainee has the form “q obtains.”. The second claim, in

turn, has the form “q’s obtaining potentially explains p’s obtaining.”. And then because of these claims and because the potential explanatory relation referred to in the second claim is lovely enough, it is inferred that p obtains. There is nothing question-begging in this, and—though correlations can figure in fixing the loveliness facts—there is more than just statistical reasoning at work.

This is perfectly parallel to inference to the best explainer, notice. The first claim is that q obtains. The second is that p’s obtaining potentially explains q’s obtaining. And then because of these claims and because the potential explanatory relation referred to in the second claim is both loveliest and lovely enough, it is inferred that p obtains. There is nothing question-begging in this (just as there is nothing question-begging in an instance of inference to an explainee), and—though correlations can figure in fixing the loveliness facts (just as in an instance of inference to an explainee)—there is more than just statistical reasoning at work (just as in an instance of inference to an explainee).

188

You might worry, however, that, though the moderate explanationist can account for the cogency of many cogent instances of SS, there are at least some cogent instances of SS the cogency of which cannot be accounted for by the moderate explanationist. Consider:

Sam is a passenger on bus b at time t, along with 99 other people. 90 of the 100 passengers have arthtritis. It is just a coincidence, however, that b at t has 100 passengers 90% of which have arthritis.

It would seem that there is a perfectly cogent instance of SS available here, running like this:

(1) 90% of the passengers on b at t have arthritis. (2) Sam is a passenger on b at t. ------(3) Sam has arthritis.

It would also seem, though, that there is no explanatory connection between the premises and the conclusion—given that it is just a coincidence that b at t has 100 passengers 90% of which have arthritis, and given that, suppose, even if Sam has arthritis, it is not the case that he is riding the bus because of it, or that he has it because he is riding the bus.

Not so—or so I shall argue. Though it is true that the inferential relation is non- explanatory, it is not true that the inferential relation is cogent (initial appearances notwithstanding).

The key ingredients to solving the problem come from Achinstein. He starts by arguing that the inferential relation is incogent:

My reply is to agree that we have here a violation of condition (d), but to claim that it is a mistake to say that e is evidence that h in this case. If it really is a coincidence that, of the 100 passengers on the bus at time t, 90 have arthritis, then I suggest that the fact that Sam happens to be on the bus at t is no evidence whatever that he has arthritis. Suppose that Sam is informed that, by coincidence, at t, 90 of the 100 riders have arthritis (and suppose that Sam has no other reason to believe that he has arthritis and no reason to believe he does not). Should Sam now see his doctor? Would the ridership of the bus at time t provide Sam (or anyone else) with any reason to believe that he has arthritis? I suggest that the answer to these questions is No. Consider the information:

e’: there is a class containing 100 people of whom 90 have arthritis, and Sam is a member of that class.

If we can assume that there are at least 10 non-arthritic people in the world, and at least 90 arthritic people, then we know that e’ must be (trivially) true. (If Sam is non-arthritic,

189

such a class exists in virtue of the existence of at least 9 other non-arthritics and at least 90 arthritics; if Sam is arthritic, it exists in virtue of the existence of at least 89 other arthritics and at least 10 non-arthritics.) The fact that there is a class of the sort described by e’ is not evidence that Sam has arthritis. All that e&b in our previous example adds to e’ above is that by coincidence one such class consists of riders on a certain bus at time t. (Achinstein 1983, 345)

And then he tries to account for the strong intuition to the effect that the inferential relation is cogent, arguing that the intuition is directed at a slightly different case:

What may be tempting the objector to say in the original case that, given b, e is evidence that h is a confusion of this case with another. Let

e1=a person will be selected at random from among the passengers on that bus at time t

b=same as above

h1=the person randomly selected will have arthritis.

It does seem to me reasonable to say that e1 is evidence that h1, in the light of b. But this case satisfies the explanatory connection condition (d): given h1&e1&b, it is very probable that the person randomly selected will have arthritis because that person will be randomly selected from among the passengers on that bus at t (of whom, b tells us, 90 percent have arthritis). (Ibid., 345-6)

I think that he is right on both counts, and thus shall only try to further develop the points.

Suppose, for starters, that there are exactly nine people in the world whose favorite color is c. Suppose further that Sam is a person in the world. Clearly, it would not follow from just the truth of these two suppositions that it is likely that Sam’s favorite color is c.

This is true, notice, in spite of the fact that there is a class of people such that (a) 90% of the members of the class like c the best and (b) Sam is a member of the class. If, as a matter of fact, Sam likes c the best, then the class includes Sam together with (a) the other eight people liking c the best and (b) one person (e.g., me) not liking c the best. If instead Sam does not like c the best, then the class includes Sam together with the nine people liking c the best. This is analogous to Achinstein’s point that “[i]f we can assume that there are at least 10 non-arthritic people in the world, and at least 90 arthritic people, then we know that e’ [i.e., there is a class of

190

people such that 90 of the 100 members of the class have arthritis and such that Sam is a member of the class] must be (trivially) true”.

Just being an instance of SS, then, is not sufficient for being cogent (even coupled with the fact that the percentage is strong enough and the fact that b is a typical F with respect to G):

(1) 90% of the people in class C like c the best. (2) Sam is a person in class C. ------(3) Sam likes c the best.

This is an instance of SS, yet it is incogent. That it is incogent follows from the fact that though

(by hypothesis) we have good inductive reason for thinking that the premises are true, we do not have good inductive reason for thinking that the conclusion is true. What we know so far, remember, is just that there are exactly nine people in the world whose favorite color is c and that

Sam is a person in the world.

Now add to the scenario the following pieces of information:

(a) the people in C are from different countries, with p1 being from c1, p2 being from c2, . . . and p10 being from c10 (b) by coincidence, the people in C end up in New York City on 2006 March 5 (c) by coincidence, the people in C end up on bus b at time t on 2006 March 5 (d) there are no other passengers on b at t on 2006 March 5

It is clear that even with this extra information, we do not yet have good inductive reason for

thinking that Sam likes c the best; this fleeting coincidence would not even increase the

likelihood that Sam likes c the best. This is analogous to Achinstein’s claim that “[a]ll that e&b

in our previous example adds to e’ above is that by coincidence one such class consists of riders

on a certain bus at time t”.

This is supposed to show that the inferential relation in the arthritis case is incogent,

contrary to initial appearances. Just as in the color case, the generalization in the arthritis case

holds only by coincidence. And so just as in the color case, that b is an F in no way makes it

likely that b is a G—even though it is true that a high percentage of Fs are Gs.

191

But what about the initial intuition that it is likely that Sam has arthritis? Why is it that,

though the inferential relation is incogent (as in the color case), it is quite intuitive that it is likely

that Sam has arthritis?

Consider the color case, again. It is clear that if we were to select a passenger at random, we would probably select a passenger liking c the best. And granted, if the passenger we selected were Sam, it would probably be the case that Sam likes c the best. The latter inferential relation, however, would be deductive, and thus not in the scope of moderate explanationism:

(1) The person selected likes c the best. (2) The person selected is Sam. ------(3) Sam likes c the best.

And the former inferential relation, in contrast, would be explanatory, in that the fact that 90% of the passengers like c the best and the fact that a passenger is selected at random together potentially explain why a passenger liking c the best is selected—in the same way in which, in the red-cube case from above, the population percentage’s being roughly 90 is the best potential explainer for the sample percentage’s being 90. It might be that in the arthritis case we are confusing the instance of SS in question with the analogue to this more complicated piece of reasoning.

Now for a slightly different, and seemingly harder, case:

Say a nucleus of some kind explodes and scatters particles randomly in all directions. The nucleus is inside a hollow sphere, and all the particles hit the inside of the sphere; they are more or less evenly distributed over the inner surface of the sphere. Now consider a large subregion of that inner surface (call it R); R is everything that is left over when we remove a small wedge from the sphere. By hypothesis, the vast majority of scattered particles landed in region R. And from this we would seemingly be justified in inferring of any given particle that it (probably) landed in R; the scattering process was random. There is no obvious distinction, in this case, between “normal” particles and deviant ones. (Lycan 1988, 187)

It would seem that there is good inductive reason for thinking that e, which is one of the scattered particles, landed in R, and that what makes it such that there is good inductive reason for thinking this is that e was a particle scattered by the explosion, that the scattering was random, and that the

192

vast majority of the particles scattered by the explosion landed in R. But it would also seem that this is problematic for the moderate explanationist, given that, by hypothesis, there are no hidden variables differentiating between the particles that landed in R and the particles that did not land in R, and given that this, in turn, entails that there are no gappy potential explainers for e’s having landed in R of which e’s being a particle scattered by the explosion is an essential member.

Lycan argues that cases such as the scattered-particle case could not establish that inductive inference forms such as statistical syllogism should be accepted as primitive:

But here is a new argument for the Ferocious view, aimed at those opponents such as van Fraassen who reject explanatory inference, at least taken as primitive, but who rely confidently on some more traditional forms of inductive and statistical inference. Inductive and statistical inference rely on evidence that is in the subject’s possession. The evidence is believed on the basis of memory. But, far from accepting all or even most of the remembered evidence, why should we believe in the reality of any past at all? Why should we not instead accept a Russellian eleventh-hour-creation hypothesis, that the world, whatever it may include besides our present sensations and memory impressions, sprang fully formed into existence half a second ago, though to be sure complete with all the memory impressions and perceptions of apparent traces and records? One cannot resolve the conflict between the hypothesis that our memories are veridical and the Russellian hypothesis by appeal to inductive or statistical argument, for it is the data premises of any such argument that are neutralized by the Russellian hypothesis. Why, then, is it reasonable for us to believe in the reality of the past at all, much less in its statistical details? It is hard to think of any answer that does not invoke explanatory considerations take as primitive. The obvious answer is that the veridicality hypothesis heavily outweighs the Russellian hypothesis at least in simplicity, neatness, and conservativeness, though certainly the details would be hard to settle. (Lycan 2002, 12)

The problem, according to Lycan, is that instances of inductive inference forms such as statistical syllogism involve, or rely on, claims about the past, and that the justification of the claims about the past itself involves, or relies on, explanatory inferences.

The question for the moderate explanationist, however, is whether there is a lovely potential explanatory relation between e’s being a particle scattered by the explosion and e’s having landed in R. For if not, then—regardless of whether the inference involves, or relies on, claims the justification of which itself involves, or relies on, explanatory inferences—moderate

193

explanationism would be false, in that, contra moderate explanationism, there would be at least one cogent but non-explanatory inductive inferential relation.

One thing that the moderate explanationist could say in response is that the objection involves deductivism, and that deductivism is false. He could start by saying that the claim

“There are no gappy potential explainers for e’s having landed in R of which e’s being a particle scattered by the explosion is an essential member.” follows from the claim “There are no hidden variables differentiating between the particles that landed in R and the particles that did not land in R.” only if explanation requires deduction. He could then say that some explanation is statistical (i.e., non-deductive), and that the scattered-particle case is a case in point.18 The

inferential relation would then look like this:

(1) e was a particle scattered by the explosion. (2) e’s being a particle scattered by the explosion is an essential member of a larger phenomenon that potentially explains e’s having landed in R. (3) There were no interfering factors. ------(4) e landed in R.

The larger phenomenon referred to in (2) would include both the fact that the scattering was random and the fact that the vast majority of the particles scattered by the explosion landed in R.

But what about the deductivist moderate explanationist? What could he say about cases such as this? Is there a way for him to account for them, or for him to explain them away? Or must he simply bite the bullet, insisting that, contrary to popular intuition, the cases in question are not, in fact, cases of cogent inductive inference?

It would seem that the second route is available. Compare the particle case to the color case from above. It is true in the color case that if we selected a passenger at random, we would probably select a passenger liking c the best, so that if, in fact, we selected Sam, it would probably be the case that Sam likes c the best. In the same way, it is true in the scattered-particle case that if we selected, by name, a particle at random, we would probably select a particle that landed in R, so that if, in fact, we selected e, it would probably be the case that e landed in R. It

194

could be said, by the moderate explanationist, of the color case that the overall inferential relation is a complex inferential relation involving both inference to the best explainer and deduction:

(1) 90% of the passengers like c the best. (2) We just selected a passenger at random. (3) The fact that 90% of the passengers like c the best and the fact that we just selected a passenger at random together potentially explain why we selected a passenger liking c the best—in the same way in which, in the red-cube case from above, the population percentage’s being roughly 90 is the best potential explainer for the sample percentage’s being 90. (4) There are no interfering factors. ------(5) We selected a passenger liking c the best. (6) The passenger we selected is Sam. ------(7) Sam likes c the best.

In the same way, it could be said, by the moderate explanationist, of the scattered-particle case that the overall inferential relation is a complex inferential relation involving both inference to the best explainer and deduction:

(1) The vast majority of the scattered particles landed in R. (2) We just selected a scattered particle at random. (3) The fact that the vast majority of the scattered particles landed in R and the fact that we just selected a particle at random together potentially explain why we selected a particle landing in R—in the same way in which, in the red-cube case from above, the population percentage’s being roughly 90 is the best potential explainer for the sample percentage’s being 90. (4) There are no interfering factors. ------(5) We selected a particle that landed in R. (6) The particle we selected was e. ------(7) e landed in R.

It would seem, then, that if what I said above about the arthritis case and the color case is correct, the deductivist moderate explanationist could insist that, contra popular intuition, the scattered- particle case is not a case of cogent inductive inference—adding that we might be confusing this case with a slightly different case in which a cogent, but slightly different, inductive inference to the same conclusion is made.

195

5.6 Conclusion

I have tried to better develop and defend moderate explanationism, the view that all cogent, or good, inductive inferential relations are explanatorily virtuous—in that each such inferential relation is an instance of either inference to the best explainer, or inference to an explainee, or an explanatory inferential-relation form built up out of inference to the best explainer and/or inference to an explainee (and/or one or more valid deductive inferential-relation forms). I argued that the moderate explanationist should construe bestness and goodness in terms of loveliness, and then construe loveliness such that what is loveliest is likeliest, and what is lovely enough is likely enough. I argued that the deductivist moderate explanationist can, and should, invoke gappy potential explainers, to handle cases such as the rain case. I argued that

with the exception of Kitcher’s unificationist account of explanation, the moderate explanationist

can stay relatively neutral on the nature of explanation. I argued that the moderate explanationist

would do well to explicate the notion of bestness in terms of Kitcher’s account of unification, in

that, for instance, by doing so he could make sense of the role of Fumerton-type correlations in

inductive inference. And, last, I argued that, in addition to inductive generalization and

enumerative induction, inductive argument forms such as argument from authority, argument

from analogy, and statistical syllogism reduce to moderate explanationism.

1 Strictly speaking, the view should read “an inferential relation (between propositions about matters of fact) is cogent only if it is causal—moving either from putative cause to putative effect, or from putative effect to putative cause, or from putative effect to putative effect by way of a putative common cause”. Hume should not be read as saying that there is a cogent inductive inferential relation between the claim that e obtains and the claim that h obtains only if e and h are, in fact, causally connected, since then he would be saying in part—and quite implausibly—that there is a cogent inductive inferential relation between the claim that e obtains and the claim that h obtains only if h obtains, which itself rules out the possibility of a cogent inductive inferential relation between a true premise claim and a false conclusion claim.

2 I take this example from Achinstein. See Achinstein 1983, 324.

3 This sets apart my brand of moderate explanationism from Hume’s, as well as from Harman’s and Lipton’s.

4 The exception is his Michael-Jordan case against the sufficiency of the high-probability account, which says that e is evidence that h if and only if p(h/e) is higher than k (where k is the threshold for high

196

probability). The problem is supposed to be that the claim that Michael Jordan eats Wheaties and the claim that Michael Jordan is a male basketball star together make it likely that Michael Jordan will not become pregnant, so that the high-probability account has the (allegedly) counterintuitive consequence that, given that Jordan is a male basketball star, Jordan’s eating Wheaties is evidence for his not becoming pregnant. I find this objection weak, since, as even Achinstein admits, the claim that Jordan eats Wheaties and the claim that Jordan is a male basketball star together make it likely that Jordan will not become pregnant, so that, in that sense, there is a perfectly good inductive inferential relation between the first two such claims and the third such claim.

5 For another analogy, consider causal accounts of knowledge (in epistemology) and causation. The causalist (if you will) need not, but can, take a stand on the nature of causation, in order to adequately articulate and defend his account of knowledge—though, of course, he needs to stay away from accounts on which the notion of causation is cashed out, even in part, in terms of knowledge.

6 Keith Lehrer makes just this point against the coupling of explanatory coherentism (construed as a theory of knowledge) and Sylvain Bromberger’s account of explanation. Lehrer argues that Bromberger’s account makes use of the notion of knowledge, and that because of this the resulting account of knowledge would be circular. Knowledge would be construed in terms of explanatory coherence. Explanatory coherence, in turn, would be construed in terms of Brombergerian explanation, which itself is construed in terms of knowledge.

7 The difference between inductive generalization and enumerative induction is that with the former the inference is from a claim about a sample to a claim about the population (of which the sample is a sample), whereas with the latter (i.e., with enumerative induction) the inference is from a claim about a sample to a claim about a member of the population that is not in the sample. “Most observed Americans are left- handed. Therefore (in all probability), most Americans are left-handed.” is an instance of inductive generalization. “Most observed Americans are left-handed. Therefore (in all probability), Paul, who is an American, is left-handed.” is an instance of enumerative induction.

8 See Carter 2004, Russow and Curd 1989, Copi and Cohen 2005, Layman 2005, Hurley 2003, Salmon 2002, and Fogelin and Sinnott-Armstrong 2001.

9 See Russow and Curd 1989, Layman 2005, and Salmon 2002.

10 I take this kind of case from William Melanson.

11 With the stuff about consensus added in, “argument from authorities” better describes the inferential relation than does “argument from authority”.

12 See Russow and Curd 1989 and Hurley 2003.

13 See Hurley 2003 and Fogelin and Sinnott-Armstrong 2001.

14 Some theorists say that when evaluating an instance of AFAu, it is important to consider the question of whether it is true that S in fact says that p. The first worrisome possibility, such theorists suggest, is where the arguer incorrectly paraphrases the expert. The expert says something close to p, but not quite. The arguer misconstrues, either intentionally or not, the expert as saying that p. And, last, the arguer then appeals to this supposed fact as evidence for p. The second worrisome possibility is where the arguer takes what the expert says out of context. The expert says that p, but strictly speaking means something slightly different. The arguer gives the relevant sentence or paragraph, but without the larger context. And then, finally, the arguer uses the expert’s word as a premise for p. I agree. But as this is an issue concerning soundness (not cogency), I shall leave it at that.

15 I take this example from Ronald Giere. See Giere 1984.

197

16 See Ennis 1968.

17 Lycan gives two very different responses to the SS problem. For the first, see Lycan 1988, 186-8. For the second, see Lycan 2002.

18 Railton’s D-N-P model of explanation would do. Railton, in answer to the question of why nucleus u decayed during the interval t0 - (t0 + θ), starts with the following: (1) All nuclei of U238 have a probability r of emitting an alpha-particle during any interval of length θ (unless subjected to environmental radiation). 238 (2) u was a nucleus of U at time t0 (and was subjected to no environmental radiation before or during the interval t0 - (t0 + θ)). ------

(3) u had a probability r of emitting an alpha-particle during the interval t0 - (t0 + θ). Railton argues that, though relevant, this in no way counts as the correct explanation of why u decayed

during the interval t0 - (t0 + θ). The correct explanation involves, in addition, (a) a theoretical account T of the underlying mechanism in question (i.e., the theory of quantum-mechanical tunneling), (b) a derivation of (1) from T, and (c) a parenthetic addendum to the effect that u decayed during the interval t0 - (t0 + θ). So though the claim that u decayed during the interval t0 - (t0 + θ) is not entailed by either the claim that u had a probability r of emitting an alpha-particle during the interval t0 - (t0 + θ), or the claim that u was a 238 nucleus of U at time t0 (and was subjected to no environmental radiation before or during the interval t0 - 238 (t0 + θ)), or the claim that all nuclei of U have a probability r of emitting an alpha-particle during any interval of length θ (unless subjected to environmental radiation), or the claim that T is true, or any combination of such claims, u’s decaying during the interval t0 - (t0 + θ) is nonetheless explained by such claims—on Railton’s D-N-P model of explanation.

198

CHAPTER 6

META-PERSPECTIVALISM

That a coherentist should be a meta-perspectivalist—and say, thus, that (to have good

reason) S needs to have a perspective, or view, of kind K on his first-level perspective, or view, as

to what the world is like—is uncontroversial in the coherentist camp. EMV, remember, is meta-

perspectivalist, in saying, in part, that S’s belief system is coherent only if it involves a view as to

how he (or, better, his belief system) is connected to the outside world, and according to which

the mechanisms involved (e.g., vision) are reliable. So too are Laurence BonJour’s account,

William Lycan’s account, and Keith Lehrer’s account, though in ways differing from EMV.

But why should a coherentist be a meta-perspectivalist? And to what kind of meta- perspectivalism should a coherentist subscribe? The aim in this chapter is to answer these two questions.

I shall start with the why-question, giving a new rationale for meta-perspectivalism.

Then I shall answer the what-question, by articulating and defending a new variety of meta- perspectivalism. I shall argue that it fits with the rationale, that it is free of the troubles plaguing its rivals in the meta-perspectivalist camp, and that it implies moderate explanationism. And then, finally, I shall address the worry that the meta-perspectivalist requirement in EMV is far too stringent—requiring far too much self-reflection and sophistication for animals, and kids, and the man-in-the-street, and even the average epistemologist.

199

6.1 In Search of a Rationale

6.1.1 BonJour

The impetus for BonJour’s observation requirement, which is his meta-perspectivalist component, is the Isolation Objection, specifically the IO 2—which says, remember, that coherentism allows for isolation scenarios (i.e., scenarios in which S’s belief system is justified but entirely isolated from the outside world), and that, since in such scenarios truth is unlikely, coherentism fails to secure a tight connection between justification and truth. BonJour argues that the observation requirement is essential to rebutting the Isolation Objection:

This point [about isolation] is, I believe, essentially sound. What it shows is that any adequate account of empirical knowledge must require putative input into the cognitive system, not merely allow for the possibility of such input. For, as was already argued in the initial statement of objection (II) [i.e., the IO 2], without input of some sort any agreement which happended to exist between the cognitive system and the world could only be accidental and hence not something which one could have any good reason to expect. Thus, as a straightforward consequence of the idea that epistemic justification must be truth-conducive, a coherence theory of empirical justification must require that in order for the beliefs of a cognitive system to be even candidates for empirical justification, that system must contain laws attributing a high degree of reliability to a reasonable variety of cognitively spontaneous beliefs (including in particular those kinds of introspective beliefs which are required for the recognition of other cognitively spontaneous beliefs). (BonJour 1985, 141, emphasis mine)

The problem with this is that the Isolation Objection is about the possibility not of apparent isolation, but of actual isolation. The worry is not that coherentism allows for apparent isolation scenarios (i.e., scenarios in which a belief system is both justified and from its perspective entirely isolated from the outside world), but that coherentism allows for actual isolation scenarios (i.e., scenarios in which a belief system is justified but entirely isolated from the outside world), and that this—i.e., the fact that coherentism allows for actual isolation scenarios—makes it impossible for coherentism to secure a tight connection between justification and truth. So it is not the case, contra BonJour, that BonJour’s brand of meta-perspectivalism is essential to staving off the Isolation Objection—since, in securing just the appearance of non-isolation, it is not even to the point.

200

BonJour, though, is not yet finished with the observation requirement, as he also uses it in arguing against the Alternative-Systems Objection (in particular, the ASO 2):

Thus if it is to constitute a serious objection to a coherence theory of the sort in question here, objection (I) [i.e., the ASO 2] must be interpreted to mean that even in the long run and with the continued impact of (putative) observational beliefs, there will always be multiple, equally coherent empirical systems between which a coherence theory will be unable to decide. But once the possibility of observational input is appreciated, it is no longer clear why this claim should be accepted, or at least why it is thought to be any more plausible in relation to a coherence theory than it is in relation to other theories of knowledge.4 The basic rationale for the original version of the objection was that alternative coherent systems could, at least in principle, be constructed arbitrarily. But such an arbitrarily constructed system will not in general satisfy the Observation Requirement; and if one should be so constructed as to intially satisfy that requirement, there is no reason to think that it would remain coherent as (putative) observations accumulate, even if it were coherent in the beginning. Thus the possibility of arbitary invention seems to provide no real support for the envisaged objection. (Ibid., 144)

It is unlikely, the response goes, that an actual person could construct, at will, an actual belief system that would satisfy the observation requirement and be coherent at the instance of construction, let alone construct, at will, an actual belief system that would satisfy the observation requirement and be coherent over the long run.

But so what? There is there is no claim in the Alternative-Systems Objection (i.e., the

ASO 2) to the effect that an actual person could construct, at will, whichever of the incompatible yet fully coherent possible belief systems that he most likes:

(1) If coherentism is true, then justification is solely a matter of coherence. (2) There are lots of incompatible yet fully coherent possible belief systems such that any possible self-consistent belief whatsoever is a member of at least one such system. ------(3) If coherentism is true, then each such system is justified. (4) That a proposition p is likely to be true entails that no proposition q incompatible with p is also likely to be true. ------(5) If coherentism is true, then justification is not connected to truth. (6) Justification is connected to truth. ------(7) Coherentism is false.

The objection is about possible people, possible belief systems, and possible beliefs. That it is unlikely that an actual person could construct, at will, an actual belief system that would satisfy

201

the observation requirement and be coherent in both the short run and the long run is entirely irrelevant.1

I should note, further, that no such rationale for a variety of meta-perspectivalism could work. After all, I rebutted both the Alternative-Systems Objection and the Isolation Objection in

Chapter 2 without any appeal to meta-perspectivalism. I argued that the coherentist can be an externalist, and that the objections fail against externalist coherentism. I argued that internalist foundationalism is faced with similar objections, so that were internalist coherentism to fall, internalist foundationalism too would fall. And, last, I argued against the objections directly, arguing that each version thereof, when fully fleshed out, has either a false premise or a fallacious sub-argument.

Ernest Sosa, for one, makes just this mistake, in “The Coherence of Virtue and the Virtue of Coherence”.2 Sosa starts with a version of the Alternative-Systems Objection, arguing that the

coherentist, to meet it, needs to appeal to a rather strong (or demanding) variety of meta-

perspectivalism. He then argues that this variety of meta-perspectivalism is far too stringent.

And then last, he concludes that coherentism is false. But again, the coherentist can meet the

Alternative-Systems Objection without recourse to meta-perspectivalism, and thus without

recourse to the variety thereof that Sosa attacks.

6.1.2 Lehrer

Lehrer’s adherence to meta-perspectivalism is seemingly quite principled, in that meta-

perspectivalism seemingly follows directly from his account of coherence. But, as I shall argue

below, initial appearances are deceiving in this case, as Lehrer’s “argument” to meta-

perspectivalism is really just an assertion of meta-perspectivalism.

Lehrer, remember, starts with personal justification, saying that S’s belief that p is personally justified just in case p coheres with the rest of S’s belief system D, and that p coheres

202

with the rest of D just in case all objections to p on D are either answered or neutralized by the other beliefs in D. Lehrer then moves to ultra justification, which requires (in addition to personal justification) that all objections to p on D be either answered or neutralized by the true beliefs in D. And included in Lehrer’s discussion of ultra justification is a discussion of the

Isolation Objection, which is where meta-perspectivalism first comes up. Lehrer writes:

Suppose that someone claims something about the world, that she sees a table in front of her, for example. To be personally justified, all objections must be answered or neutralized. Now, and this is the crucial point, the isolation objection is an objection. Consider the following justification game:

Claimant: I see a table in front of me. Ultracritic: You are isolated from the external world. Claimant: It is more reasonable for me to accept that I see a table in front of me than that I am isolated from the external world. (I am visually connected with the external world and not isolated from it.)

The evaluation system of the claimant enables her to answer the skeptical objection because she accepts that she is appropriately connected with the external world and not isolated from it. Assuming, finally, she is correct in accepting this, then her victory in this round of the justification game will be sustained in other rounds and transformed into a victory in the ultra justification game as well. Her justification will go undefeated. (Lehrer 2000, 163)

The claim that S, the claimant, is isolated from the world, the idea goes, is an objection to S’s claim that p obtains, in that it is more reasonable for S to believe that p obtains on the assumption that he is not isolated from the world than on the assumption that he is isolated from the world.

Because of this, S needs to either answer or neutralize the claim that he is isolated from the world.

S does this, suggests Lehrer, by accepting that he is connected in various ways (e.g., visually) to the world.

I agree, in full. I agree that S would have better reason for thinking that p obtains were he to think that he is connected to the world in various ways than were he not to think that he is connected to the world in various ways. Moreover, I agree that S would not have good reason for thinking that p obtains were he not to think that he is connected to the world in various ways.

The problem, however, is that neither such claim is a rationale for meta-perspectivalism. The

second claim, for instance, is itself a kind of meta-perspectivalism.

203

6.1.3 Lycan

The same is true of Lycan’s discussion, unfortunately. Lycan says some things by way of support for the brand of meta-perspectivalism to which he subscribes. I agree with each point.

But none of the points makes sense of why meta-perspectivalism is needed.

Lycan distinguishes between four kinds, or grades, of coherence (all pertaining just to non-inferential beliefs), the fourth of which is meta-perspectivalist. Consider:

S’s non-inferential belief that p obtains coheres1 just in case it is logically consistent with his other non-inferential beliefs.

S’s non-inferential belief that p obtains coheres2 just in case it is probabilistically consistent with his other beliefs—meaning that it is logically consistent with his other beliefs, and that it is not the case that his other beliefs make it likely that p does not obtain.

S’s non-inferential belief that p obtains coheres3 just in case S believes that q obtains, and S takes q’s obtaining to explain why p obtains.

S’s non-inferential belief that p obtains coheres4 just in case S has an explanatory account as to why he holds the belief that p obtains and according to which the belief-forming process involved is reliable.

Lycan says that, in addition to coherence1 and coherence2, justification (at least for non-inferential

beliefs) requires coherence4. S needs to have a view of (or belief about) his view of (or beliefs about) the world—a meta-view, that is, to the effect that his non-inferential views, or beliefs, are the result of such and such belief-forming processes, and that such and such belief-forming processes are reliable.3 This is the sense in which Lycan is a meta-perspectivalist.4,5

The first thing that Lycan says on behalf of his meta-perspectivalist component is that it helps in beefing up the initial modicum of justification that is conferred on non-inferential beliefs in virtue of being held.6 I do not agree, of course, with either the claim that non-inferential beliefs have such an initial modicum of justification, or the claim that this is needed in order to rebut the Database Objection. I do agree, though, that full coherence, and thus full justification, requires a meta-perspectivalist condition of some kind. Merely saying this, however, in no way suffices for explaining why, or showing that, a meta-perspectivalist constraint is needed.

204

The second thing that Lycan says in support of his brand of meta-perspectivalism is that it helps in ruling out certain counterintuitive cases:

Coherence of the third kind will rule out what we might call “wild” spontaneous beliefs. The spontaneous beliefs that I have focused on so far are of familiar, lovable sorts, produced by perceptual mechanisms, memory, and the like. But we acquire lots of other spontaneous beliefs as well, some of which survive the first two coherence tests and so achieve tenability. For example, we all have primitive, superstitious forebodings of various kinds—eerie feelings that such and such is so, fits of inspiration, tricks of memory, such as déjà vu. Some people unquestionably have religious feelings and experiences. All these sorts of things (and more) can qualify as or produce spontaneous beliefs that may be tenable—logically consistent with all one’s other spontaneous beliefs and with one’s other theories as well, however suspicious they may seem even to the believer. We do not want such beliefs to count as fully justified. I suggest the reason is that typically they do not achieve coherence of the third kind. Either we simply do not find any information-transmitting mechanisms that are producing them, no matter how hard we look or how much we know about our minds, or we find positively discrediting evidence—minor or major disorders whose presence explains the wild beliefs away.6 Thus, although these beliefs are tenable, the ability of other hypotheses to explain them will not count much in favor of those hypotheses. (Lycan 1988, 168-9)

I agree, again. Such beliefs are not justified. Lycan’s meta-perspectivalist requirement accounts for this. And that it does so is to its credit. But none of this even addresses the why-question— viz., the question of why Lycan’s meta-perspectivalist requirement is needed for having good reason.

6.1.4 The Rationale

Let me stress that I am not meaning to criticize BonJour, Lehrer, and Lycan. I am simply pointing out that none of them answers the why-question, at least not in a way that yields philosophical understanding as to why the meta-perspectivalist is right in saying that having good reason requires a good bit of self-reflection.

But BonJour, in a different discussion, hints at the right thing to say:

One reason why externalism may seem initially plausible is that if the external relation in question genuinely obtains, then Norman will in fact not go wrong in accepting the belief, and it is, in a sense, not an accident that this is so: it would not be an accident from the standpoint of our hypothetical external observer who knows all the relevant facts and laws. But how is this supposed to justify Norman’s belief? From his subjective perspective, it is an accident that the belief is true. And the suggestion here is that the

205

rationality or justifiability of Norman’s belief should be judged from Norman’s own perspective rather than from one which is unavailable to him. (BonJour 1985, 43-4)

The suggestion here is that there is an intimate tie between justification (in the sense of having good reason) and perspectival non-accidentality, so that the reason why Norman’s belief about the President is unjustified is that it would be a sheer accident or coincidence—from Norman’s perspective—were his belief about the President true. This suggestion, I shall argue, is the key to why a coherentist should be a meta-perspectivalist.

Imagine, then, that S is entirely isolated from the outside world. Then (in a sense to be specified below) it would not be likely that S’s beliefs are true—as it would be a mere coincidence were they, in fact, true.

The same would be true relative to S’s perspective (or beliefs), if, relative to S’s

perspective (or beliefs), he were entirely isolated from the outside world. That is, if S thinks that

he is entirely isolated from the outside world, or if what S thinks implies that he is entirely

isolated from the outside world, then from his perspective, or relative to the totality of his beliefs,

it would not be likely that his beliefs are true—since from his perspective, or relative to the

totality of his beliefs, it would be a mere coincidence were they, in fact, true. Hence, from S’s

perspective it is unlikely that his beliefs are true if from S’s perspective he is isolated from the

outside world.7

Now imagine not that S is isolated from the outside world, but that it is not likely that S is not isolated from the outside world. That is, imagine that it is not likely that S is connected to the outside world. Then, again, it would not be likely that S’s beliefs are true. And were one such belief the belief that p obtains, it would not be likely that p obtains. Why? Because S could very well be isolated, in which case it would be a mere coincidence were S’s beliefs true. The reasoning, put slightly differently, goes like this:

(1) If it is likely that S’s beliefs are true, then S is not isolated from the world. (2) If S is not isolated from the world, then—trivially—it is likely that S is not isolated from the world. ------

206

(3) If it is likely that S’s beliefs are true, then it is likely that S is not isolated from the world—which means that if it is not likely that S is connected to the world, then it is not likely that S’s beliefs are true.

The underlying claim, in making (2), should be quite intuitive—viz., that if x obtains at t, then, at t, there is no chance that x does not obtain at t, in which case it is at least likely, at t, that x obtains at t.8

By the same reasoning, it follows that if from S’s perspective it is not likely that he is connected to the world, then from S’s perspective it is not likely that his beliefs are true. It would then follow that if one such belief is the belief that p obtains, then from S’s perspective it is not likely that p obtains.

The considerations so far deal just with global isolation, in which S is entirely isolated from the world. But what about local isolation, in which some but not all of S’s beliefs are isolated from the world? Could similar things be said about local isolation scenarios?

It would seem so. Though S might be connected to lots of different phenomena in the world, it would be a sheer coincidence—were no such phenomena connected to p and thus were it not the case that he is connected to p—were his belief that p obtains true. By hypothesis, there would be no connection between (1) p’s obtaining and (2) S’s holding the belief that p obtains.

Thus if it is not the case that S is connected to p, then it is not likely that S’s belief that p obtains is true.9

The same, then, holds for the other three cases. First, it would not be likely that S’s belief that p obtains is true if it were not likely that S is connected to p. Second, from S’s perspective it would not be likely that his belief that p obtains is true if from S’s perspective he were not connected to p. And third, from S’s perspective it would not be likely that his belief that p obtains is true if from S’s perspective it were not likely that he is connected to p.

The overall lesson is twofold:

(1) It needs to be likely from S’s perspective that he is connected to the outside world. (2) It needs to be likely from S’s perspective that he is connected to p.

207

If it is either not likely from S’s perspective that he is connected to the outside world or not likely from S’s perspective that he is connected to p, then S does not have good reason for thinking that p—regardless of what else S believes.

This twofold lesson is the rationale (or at least the bulk of it) for the meta-perspectivalist requirement in EMV. The point of the meta-perspectivalist requirement in EMV is to ensure both that it is likely from S’s perspective that he is connected to the outside world, and that it is likely from S’s perspective that he is connected to p.

6.1.5 A Possible Reply

What if S believed that q obtains, and if, in addition, S believed that q’s obtaining makes it likely that p obtains? Then would not it follow that from S’s perspective it is likely that p obtains, even if from S’s perspective it were not likely that he is connected to the outside world?

After all, it would follow that from S’s perspective q obtains, and that from S’s perspective q’s obtaining makes it likely that p obtains.

It would not do to answer by arguing that since from S’s perspective it is not likely that he is connected to the outside world and since, therefore, from S’s perspective it is not likely that his beliefs are true, it follows that from S’s perspective it is not likely that his belief that q obtains is true, and that, thus, it follows that from S’s perspective it is not likely that p obtains. The problem would be that the non-meta-perspectivalist could simply reply in the same way as above.

What if S believed that r obtains, and if, in addition, S believed that r’s obtaining makes it likely that p obtains? Then would not it follow that from S’s perspective it is likely that q obtains, even if from S’s perspective it were not likely that he is connected to the outside world? After all, it would follow that from S’s perspective r obtains, and that from S’s perspective r’s obtaining makes it likely that q obtains.

208

The right response, instead, is that S’s belief that q obtains, qua reason for thinking that p obtains, is trumped. The claim, that is, is that its not being likely from S’s perspective that he is connected to the outside world precludes S’s belief that q obtains from making it such that S has good reason for thinking that p obtains.

It might be helpful to put this in terms of Matthias Steup’s distinction between undermining defeaters and contradicting defeaters10, which can be put as follows:

S’s belief that d obtains defeats, in the undermining sense, his belief that q obtains as evidence for his belief that p obtains just in case (a) (all else being equal) q’s obtaining makes it highly likely that p obtains, (b) it is not the case that q’s obtaining together with d’s obtaining makes it highly likely that p obtains, and (c) it is not the case that q’s obtaining together with d’s obtaining makes it highly likely that p does not obtain. S’s belief that d obtains defeats, in the contradicting sense, his belief that q obtains as evidence for his belief that p obtains just in case (a) (all else being equal) q’s obtaining makes it highly likely that p obtains, (b) it is not the case that q’s obtaining together with d’s obtaining makes it highly likely that p obtains, and (c) q’s obtaining together with d’s obtaining makes it highly likely that p does not obtain.

The point, in these terms, is not that its not being likely from S’s perspective that he is connected to the outside world is a contradicting defeater, but that its not being likely from S’s perspective that he is connected to the outside world is an undermining defeater. The claim is that its not being likely from S’s perspective that he is connected to the outside world makes it such that none of his beliefs about the outside world—not even the ones for which he has reasons—are beliefs for which he has good reason.

Here is an analogy. Imagine, first, that Jack believes that none of his belief-forming processes are reliable, and that he has a true meta-belief for each of his non-meta-beliefs. Then, quite clearly, it would not be the case that his non-meta-beliefs are justified, in the having-good- reason sense. After all, each of his non-meta-beliefs would be entirely at odds with his other beliefs, given that from his other beliefs he could reason, as follows, against each of his non- meta-beliefs:

(1) I hold the belief that p obtains. (2) None of my belief-forming processes are reliable. ------(3) My belief that p obtains is the result of an unreliable belief-forming process.

209

------(4) It is unlikely that my belief that p obtains is true. ------(5) It is unlikely that p obtains.

Imagine next that, though he has no meta-beliefs, Jack believes that none of his belief-forming processes are reliable. What then? It would seem that each of Jack’s non-meta-beliefs would still be at odds with his other beliefs.11 Why? Because his other beliefs would be saying, in part, that he holds a belief b only if b is the result of an unreliable belief-forming process, and only if, thus, it is unlikely that b is true. Jack—in virtue of holding the belief that none of his belief-forming processes are reliable—is saying, in effect, that whatever he believes, if he believes anything, is probably false. This is perfectly parallel to scenarios in which from S’s perspective it is not likely that he is connected to the outside world, in that S, in such scenarios, is saying, in effect, that whatever he believes, if he believes anything, is probably false (or at least could very well be false). This is because from S’s perspective, in such scenarios, it is not likely that his beliefs are true.

6.2 In Search of the Right Variety

There is no difficulty in finding varieties of meta-perspectivalism that can account for the

lessons from the previous section. The difficulty is in finding varieties that can account for such

lessons without requiring a regress of increasingly complex beliefs, and without conflicting with

the central tenets of coherentism.

6.2.1 Accounts Requiring Justified Meta-Beliefs and Justified Reliability Beliefs

Sosa’s discussion, discussed briefly in the previous section, is a good place to start, as it

is quite suggestive of how not to be a meta-perspectivalist. Consider:

210

yields the incoherence of the set resulting from the transformation in our thought experiment. (Sosa 1991, 205-6)

More importantly, to require that the subject always be aware of his own beliefs would sink us in the quicksand already skirted earlier of requiring acceptable world views to include not only P—the object of an arbitrary object-level belief—but also B(P), and B(B(P)), and B(B(B(P))), etc. (Ibid., 206)

The variety of meta-perspectivalism under consideration says that, in order for S to have good reason for thinking that p obtains, S needs to believe that he believes that p obtains, and believe that his belief that p obtains is the result of a reliable belief-forming process.12 The problem,

argues Sosa, is that the meta-belief requirement starts a regress, as follows. Since presumably S

would also need to have good reason for thinking that he believes that p obtains, S would need to

believe that he believes that he believes that p obtains, and believe that his belief that he believes

that p obtains is the result of a reliable belief-forming process. Since presumably S would then

need to have good reason for thinking that he believes that he believes that p obtains, S would

need to believe that he believes that he believes that he believes that p obtains, and believe that

his belief that he believes that he believes that p obtains is the result of a reliable belief-forming

process. And so on.

The lesson is fully general:

A variety of meta-perspectivalism is too strong if (a) it requires a meta-belief, (b) it is fully general, and (c) it requires that S have good reason for the meta-belief.

The problem with each such variety is that it requires a meta-level regress of increasingly

complex meta-beliefs.13

6.2.2 Accounts Requiring Meta-Beliefs and Justified Reliability Beliefs

Perhaps the coherentist should drop the requirement that S have good reason for the meta-belief. The view, then, would require (a) that S believe that he believes that p obtains, (b) that S believe that his belief that p obtains is the result of a reliable process, and (c) that S have good reason for thinking that the process in question is reliable. Not requiring that S have good

211

reason for thinking that he believes that p obtains, it would then not start a meta-level regress of meta-beliefs.

Think of BonJour’s use of the doxastic presumption, according to which at least most of

S’s beliefs about what he believes are true. BonJour argues that a belief system can be justification-conferring even if its meta-beliefs are not themselves supported by reasons. It is enough that the doxastic presumption is true. The imagined meta-perspectivalist sketched in the previous paragraph could make a similar move, requiring not that S’s meta-beliefs be supported by reasons but just that at least most such beliefs be true.

In fact, this is just what BonJour himself does. He starts by saying that you would have good (provisional) reason for thinking that p obtains if you were to believe that you have a cognitively spontaneous belief of kind K to the effect that p obtains, if you were to believe that

cognitively spontaneous beliefs of kind K in conditions C are highly likely to be true, and if you

were to believe that conditions C obtain. Because then you would be in a position to reason as

follows:

(1) I have a cognitively spontaneous belief of kind K to the effect that p obtains. (2) Cognitively spontaneous beliefs of kind K in conditions C are highly likely to be true. (3) Conditions C obtain. ------(4) It is highly likely that my belief that p obtains is true. ------(5) P obtains.

He then says, quite sensibly, that, in order for this to work, these beliefs too would need to be justified, and in a way consistent with coherentism. But then, last, he simply appeals to the doxastic presumption:

The first subpremise, that I do have the specific belief in question, represents the most obvious and least problematic application of the Doxastic Presumption. As noted earlier in the dicussion of the aprioristic variant of foundationalism (in section 4.5), it is plausible to hold that the existence of the justificandum belief is presupposed, in something like the Strawsonian sense,13 by the very raising of the issue of justification, so that it does not need to be even included as a premise. But whether or not this is so, the grasp of my overall system of beliefs to which a coherence theory must in any case appeal can surely be taken without any additional qualms to include my grasp of the existence of the justificandum belief itself. (BonJour 1985, 128)

212

It is required that you have good reason for thinking that cognitively spontaneous beliefs of kind

K in conditions C are highly likely to be true. And it is required that you have good reason for thinking that conditions C obtain. But it is not required that you have good reason for thinking that you believe that p obtains. It is enough that at least most of your beliefs about what you believe are true.

You might wonder why BonJour does not instead appeal to his account of introspection, which is supposed to make sense of how your introspective beliefs could be supported by reasons.

The problem would be that—since his account of introspection is modeled on the account above and since it itself, thus, involves meta-beliefs—it (i.e., the suggestion) would start a vicious regress. Thus with both perceptual beliefs and introspective beliefs, BonJour appeals, in part, to the doxastic presumption.

The chief problem with requiring both a meta-belief and a justified reliability belief but not requiring a reason for the meta-belief is that it is at odds with a central tenet of coherentism— viz., the claim that a belief is justified only if it is supported by reasons. For it allows for justification-conferring belief systems with meta-beliefs that are not supported by reasons, and

thus allows for justified meta-beliefs that are not supported by reasons. This should be avoided, if

possible.

Perhaps, though, a meta-perspectivalist coherentist of this kind could restrict justification to just the beliefs in a justification-conferring belief system that are supported by reasons. It would then not follow from the fact that there could be justification-conferring belief systems

with meta-beliefs that are not supported by reasons that there could be justified meta-beliefs that

are not supported by reasons. But then the worry would be that, because they are not themselves

justified, the meta-beliefs in question could not help in making other beliefs justified. How is it

that a belief for which you have no reasons could help in giving you good reason for another

belief?

213

6.2.3 Accounts Requiring Justified Reliability Beliefs

Perhaps the lesson should be that it is not required that you have a meta-belief, let alone a meta-belief supported by reasons. It would still be required that you believe that you are reliably connected to the outside world, by, for instance, vision. But it would not be required that you be aware of the beliefs produced by the processes by which you are reliably connected to the outside world.

This would certainly help with the problems facing the views considered above, in that

there would be neither a meta-level regress nor a justified meta-belief that is not supported by

reasons. There would still be difficulties, however.

Suppose that S believes that a is F, that S believes that b is F, . . ., that S believes that n is

F, and that S believes that he is a reliable indicator of Fs. S’s picture of the world, then, would look, in part, like this:

a is F. b is F. . . . n is F. I am a reliable indicator of Fs.

Would it follow that S, in virtue of his belief that he is a reliable indicator of Fs, has good reason for thinking that b, for instance, is F? And would it follow that S, in virtue of his belief that a is

F, his belief that b is F, . . ., and his belief that n is F, has good reason for thinking that he is a reliable indicator of Fs?

It would seem not. For it would that, in addition, S would need to believe that he believes that a is F, and that S would need to believe that he believes that b is F, and so on:

a is F. b is F. . . . n is F.

214

I believe that a is F. I believe that b is F. . . . I believe that n is F. I am a reliable indicator of Fs.

S, then, would have good reason for thinking that b, for example, is F—because of his belief that he is a reliable indicator of Fs together with his belief that he believes that b is F. And S would have good reason for his belief that he is a reliable indicator of Fs—because of his belief that a is

F, his belief that b is F, . . ., his belief that n is F, his belief that he believes that a is F, his belief that he believes that b is F, . . ., and his belief that he believes that n is F.

This is perfectly analogous to the relationship between beliefs about the reliability of measuring devices and beliefs about the things that the devices are supposed to measure. That it is 98 degrees at t, that it is 96 degrees at t’, that it is 102 degrees at t’’, . . . in no way together make it likely that thermometer T is reliable. And that T is reliable in no way makes it likely that it is 98 degrees at t, that it is 96 degrees at t’, that it is 102 degrees at t’’, . . .. But with the fact that T says that it is 98 degrees at t, T says that it is 96 degrees at t’, T says that it is 102 degrees at t’’, . . . together with the fact that it is 98 degrees at t, it is 96 degrees at t’, it is 102 degrees at t’’, . . . it is, in fact, likely that T is reliable. And, conversely, with the fact that T says that it is 98 at t, T says that it is 96 at t’, T says that it is 102 at t’’, . . . together with the fact that T is reliable it is, in fact, likely that it is 98 at t, that it is 96 at t’, that it is 102 at t’’, . . ..

The problem, then, is that, though not requiring a meta-belief avoids certain problems, it, in doing so, leaves a gaping explanatory void. How is it that S has good reason for his non- inferential beliefs? S’s beliefs about reliability are not sufficient. And how is it that S has good reason for his reliability beliefs? S’s non-inferential beliefs are not sufficient.

215

6.2.4 Accounts Requiring Justified Connection Beliefs

There is a radically different means of cashing out the meta-perspectivalist requirement, a means that requires not meta-beliefs but connection-beliefs:

S has good reason for thinking that p obtains only if S has a belief as to the way in which he is connected to p (e.g., by vision).

The requirement is not that S have a belief as to how his belief that p obtains is connected to p, but that S have a picture as to how he, or his cognitive system, is connected to p.

This is a nice suggestion in at least two respects. First, it is not troubled by a meta-level regress of meta-beliefs. Second, it (in contrast to accounts that require reliability beliefs but not meta-beliefs) yields a clear sense in which S has good reason for his reliability beliefs. I shall

make this second point more clear in the next sub-section.

It is not without regress troubles, however. Suppose that S holds the belief that p obtains.

Then, to have good reason for thinking that p obtains, S would need to have a belief as to the way in which he is connected to p—viz., a belief to the effect that he is connected in way w to p.

Then, to have good reason for thinking that he is connected in w to p, S would need to have a belief to the effect that he is connected in way w’ to the fact that he is connected in w to p. Then, to have good reason for thinking that he is connected in w’ to the fact that he is connected in w to p, S would need to have a belief to the effect that he is connected in way w’’ to the fact that he is connected in w’ to the fact that he is connected in w to p. And so on. This is not a meta-level regress of meta-beliefs. Nonetheless, it is troublesome.

6.2.5 The Account

The suggestion in the previous sub-section, though inadequate in the end, is on the right track. It is right in moving away from meta-beliefs to connection-beliefs. It is wrong, though, in requiring such beliefs across the board. The lesson from section 1—i.e., that S has good reason

216

for thinking that p obtains only if from S’s perspective it is likely that he is connected to p—can be accommodated without requiring, across the board, that S have a belief as to how he is connected to p.

Suppose, first, that S believes, non-inferentially, that p obtains, and that S believes, further, that he is connected in way w (e.g., vision) to p. Then, since the latter entails the former,

S would have good reason for the former if he had good reason for the latter.

The suggestion in sub-section 6.2.4 would require that S have a picture of the way in which he is connected to the fact that he is connected in w to p, in order for S to have good reason for thinking that he is connected in w to p. But because this would be the second link in an endless regress, the suggestion in sub-section 6.2.4 should be rejected.

Suppose, next, that S believes that he is having a w experience of p, that S believes that his w is highly reliable, and that S has at least a rudimentary theory, T, as to how his w works— and according to which (in normal conditions) it yields a w experience of x just in case he is connected in w to x. With such beliefs, S would then have good reason for thinking that he is connected in w to p (assuming, of course, that S has good reason for them). For given that his w is highly reliable and that his w works as T says, that he is connected in w to p would be the best explanation of his having a w experience of p.

Now add in that S believes that q obtained at t, that S believes that r obtained at t’, . . ., that S believes that he was connected at t in w to q, that S believes that he was connected at t’ in w to r, . . .. Then (if S has good reason for these beliefs) S would have good reason for thinking that his w is reliable. That his w is reliable would be the best explanation of this spotless track- record.

For S’s belief that he is having a w experience of p, nothing extra is needed—as is evidenced by how his belief that he is connected in w to p would be supported. The line, remember, is that it would be supported by his belief that he is having a w experience of p, his belief that his w is highly reliable, and his belief that his w works as T says; that he is connected

217

in w to p, the idea goes, would be the best explanation of his having a w experience of p. With

S’s belief that he is having a w experience of p, things would be reversed: his having a w experience of p would be explained by his being connected in w to p (together with how his w works).

S’s overall perspective, or world view, would thus look like this (at least in part):

(a1) p obtains. (a2) q obtained at t. (a3) r obtained at t’. . . . (b1) I am connected in w to p. (b2) I was connected at t in w to q. (b3) I was connected at t’ in w to r. . . . (c) My w is highly reliable. (d) My w works as T says. (e1) I am having a w experience of p. (e2) I had a w experience at t of q. (e3) I had a w experience at t’ of r. . . .

(a1) follows from (b1), just as (a2), (a3), . . . follow from (b2), (b3), . . .. (b1) follows from (c),

(d), and (e1), in the same way that (b2), (b3), . . . follow from (c), (d), and (e2), (e3), . . .. (c) and

(d) follow from (b1), (b2), (b3), . . . and (e1), (e2), (e3), . . .. And (e1) follows from (b1), (c), and

(d), just as (e2), (e3), . . . follow from (b2), (b3), . . ., (c), and (d).

Imagine, for instance, that Jim believes, non-inferentially, that there is a blue Great Dane before him. Then the account, using vision as the connecting mechanism, would work like this.

First, Jim would believe that he is visually connected to the fact that there is a blue Great Dane before him, which itself entails that there is a blue Great Dane before him:

I am visually connected to the fact that there is a blue Great Dane before me. ------There is a blue Great Dane before me.

218

Second, Jim would believe that his vision is reliable, believe that his vision works as V says, and believe that he is having a visual experience of a blue Great Dane before him. Together, these beliefs would support his belief that he is visually connected to the fact that there is a blue Great

Dane before him:

My vision is reliable. My vision works as V says. I am having a visual experience of a blue Great Dane before me. ------I am visually connected to the fact that there is a blue Great Dane before me.

Third, Jim would have lots of beliefs of the form “I am visually connected to x.” and lots of beliefs of the form “I am having a visual experience of x.”, which would together support his belief that his vision is reliable:

I am visually connected to the fact that there is a blue Great Dane before me. I was visually connected a few moments ago to the fact that, a few moments ago, it was raining. I was visually connected last Friday to the fact that, last Friday, Sally was crying. . . . I am having a visual experience of a blue Great Dane before me. I had, a few moments ago, a visual experience of its raining. I had, last Friday, a visual experience of Sally’s crying. . . . ------My vision is reliable.

And fourth, Jim’s belief that he is having a visual experience of a blue Great Dane before him

would be supported by his belief that he is visually connected to the fact that there is a blue Great

Dane before him, his belief that his vision is reliable, and his belief that his vision works as V

says:

I am visually connected to the fact that there is a blue Great Dane before me. My vision is reliable. My vision works as V says. ------I am having a visual experience of a blue Great Dane before me.

219

Each belief in this small subset of Jim’s overall belief system would be supported by, and help support, the other beliefs in the set.

The neat thing about this account is that—while not starting a meta-level regress of any kind and while not transgressing the coherentist tenet that a belief is justified only if it is supported by a reason—it fully accommodates, or respects, the lesson from section 1: the point that S has good reason for thinking that p obtains only if from S’s perspective it is likely that he is connected to p. From S’s perspective, he is connected in w to p, q, r, . . .. From S’s perspective, it is likely that he is connected to the fact that his w is highly reliable—in that from S’s perspective his w’s being highly reliable is the best explanation of his w’s spotless track-record.

And from S’s perspective, it is likely that he is connected to the fact that he is having a w experience of p—in that from S’s perspective it is likely that his having a w experience of p is explained by his being connected in w to p. For each of the beliefs in the set, S’s perspective is such that it is at least likely that he is connected to what it is about.

But what about cases in which, though you believe that you are having a w experience of p, you do not believe that p obtains and thus do not believe that you are connected in w to p?

How is it that, in such cases, you have good reason for thinking that you are having a w experience of p? The story above about experiential beliefs is not available in such cases.

True, but something similar to what I say above about S’s belief that p obtains is available. That is, the thing to say is that S would have good reason if his perspective, or world view, looked (at least in part) like this:

(a1) I am having a w experience of p. (a2) I hoped at t that q obtains. (a3) I believed at t’ that r obtains. . . . (b1) I am connected in i (i.e., introspection) to the fact that I am having a w experience of p. (b2) I was connected at t in i to the fact that I hoped, at t, that q obtains. (b3) I was connected at t’ in i to the fact that I believed, at t’, that r obtains. .

220

. . (c) My i is highly reliable. (d) My i works as H says. (e1) I am having an i experience of the fact that I am having a w experience of p. (e2) I had at t an i experience of the fact that I hoped, at t, that q obtains. (e3) I had at t’ an i experience of the fact that I believed, at t’, that r obtains. . . .

(a1) follows from (b1), just as (a2), (a3), . . . follow from (b2), (b3), . . .. (b1) follows from (c),

(d), and (e1), in the same way that (b2), (b3), . . . follow from (c), (d), and (e2), (e3), . . .. (c) and

(d) follow from (b1), (b2), (b3), . . . and (e1), (e2), (e3), . . .. And (e1) follows from (b1), (c), and

(d), just as (e2), (e3), . . . follow from (b2), (b3), . . ., (c), and (d). Problem solved.

“Not so fast.”, you might think. What about the possibility of having (e1) but not having

(a1) and thus not having (b1)? That is, what about the possibility of believing that you are having an i experience of the fact that you are having a w experience of p, but not believing that you are having a w experience of p and thus not believing that you are connected in i to the fact that you are having a w experience of p? Is not this but the start of a vicious regress?

No, since it is not required that you be in such scenarios. It is not required, to focus just on the first case, that you be in a scenario in which you believe that you are having a w experience of p, but do not believe that p obtains and thus do not believe that you are connected in w to p. Nor is it required that you be in a scenario in which you believe that you are having an i experience of the fact that you are having a w experience of p, but do not believe that you are having a w experience of p and thus do not believe that you are connected in i to the fact that you are having a w experience of p. Hence you could satisfy the meta-perspectivalist requirement in

EMV without being in either kind of scenario, and hence without even traversing the initial legs of the imagined regress.

The view, to this point, accounts for perceptual beliefs. But what about non-perceptual beliefs? The view should also account for these beliefs, since the argument in section 1 is fully

221

general; the lesson is supposed to be that S has good reason for believing (whether non- inferentially or inferentially) that p obtains only if from S’s perspective it is likely that he is connected to p.

The answer, not surprisingly, is moderate explanationism. The perspectival connection is secured by explanatorily virtuous inductive inferential relations. For a phenomenon to which, from your perspective, you are not connected via a perceptual faculty, there is an explanatorily virtuous inferential chain tracing back to a phenomenon to which, from your perspective, you are, in fact, connected via a perceptual faculty. There are three basic possibilities here. First, there is the possibility that, from your perspective, you are connected to q via a perceptual faculty, and p explains q. An example would be where you see smoke coming from a chimney and infer, via inference to the best explainer, that there is a fire in the corresponding fireplace. It would not be that from your perspective you are perceptually connected to the fire. Rather, it would be that from your perspective you are perceptually connected to the smoke, and that, from your perspective, the smoke is explained by the fire. Second, there is the possibility that, from your perspective, you are connected to q via a perceptual faculty, and q explains p. Think of the case in which you see a fire in a fireplace and infer, via inference to an explainee, that there is smoke coming from the corresponding chimney. From your perspective you are perceptually connected to the fire, which, from your perspective, explains the smoke. And third, there is the possibility that, from your perspective, you are connected to q via a perceptual faculty, and r explains both q and p. The case in which you infer that the roof is wet on the grounds that the street is wet would be an instance. From your perspective you are perceptually connected to the street’s being wet, which, from your perspective, is explained by its having just now rained, which, from your perspective, explains the roof’s being wet.

222

6.3 The Over-Demandingness Charge

It is clear that BonJour’s account is demanding. It requires not just (a) that S’s belief that

p obtains play an inferential role at t, (b) that S’s belief system be coherent at t, (c) that S’s belief

system satisfy the observation requirement at t, (d) that S’s belief system satisfy the doxastic

presumption at t, (e) that S’s belief system have been coherent, in accord with the observation

requirement, and relatively stable over a relatively long stretch of time, but also (f) that S have

access to, or a reflective grasp of, the facts referred to in (a)-(e) and (g) that S have access to, or a

reflective grasp of, the fact that the facts referred to in (a)-(e) make it highly likely that p

obtains.14

You might worry, though, that it is too demanding. You might worry, for instance, that the man-in-the-street’s grasp of his belief system is too weak to support the belief that his system is coherent. The man-in-the-street’s beliefs, then, would be unjustified—which is absurd.

The same, you might think, should be said about my account, since it too is quite demanding (though less so than BonJour’s). The man-in-the-street, you might argue, does not have beliefs about his experiences, or about the reliability of his belief-forming processes, or about the ways in which he is connected to the world, and thus does not have the kind of perspective on his place in the world required by my account. Nor do animals, or infants, or even epistemologists, you might continue. But surely, all such cognizers have justified beliefs. That is, surely animals and infants and ordinary people and epistemologists have beliefs for which they have good reason. And so my account, like BonJour’s, is too demanding and should thus be rejected.

This sort of worry (call it “the over-demandingness charge”) has two parts: (1) the claim that the man-in-the-street’s beliefs would be unjustified if my account were true; and (2) the claim that, because of this, my account is too demanding and should therefore be rejected. I shall start with the second part, arguing that, even if the first part were true, my account would better fit with, or do more justice to, the everyday epistemic practice than would any alternative theory.

223

BonJour, to his credit, addresses the over-demandingness charge head-on, arguing, in the end, that his account should be seen as a revisionary idealization with respect to which the man- in-the-street is only a rough approximation:

Thus, the account of justification offered here will represent at best an idealization which is only loosely approximated by ordinary cognition. The question is what conclusion should be drawn from this fact. One possibility is to insist that our commonsensical attribution of the status of justification and of knowledge to ordinary cases of cognition is so secure and unproblematic that any epistemological account which calls that status into question is automatically mistaken. This is in effect the position adopted by certain of the externalist views discussed in Chapter 3. I can see, however, no reason for regarding common sense as authoritative to this degree. As also suggested above (see especially sections 1.3 and 2.1), I would be inclined to favor an alternative view, according to which it could well turn out, as it does on the account presented here, that typical commonsensical cases of knowledge are only loose approximations to an epistemic ideal which is seldom if ever fully realized. This amounts to saying that our actual cognitive states, though not without a significant degree of positive epistemic value, would be improved from a purely epistemic standpoint (though not perhaps from a practical standpoint) if the inferences in question and the grasp of one’s own belief system to which they appeal were fully explicit. And I can see no serious implausibility in such a claim. (Something analogous, though stronger to an appropriate degree, would also have to be said about the cognition of young children, animals, mentally defective persons, and so on.) (BonJour 1985, 152-3)

BonJour is arguing that the man-in-the-street’s beliefs would be better off epistemically if they satisfied the conditions in his account, and that, because of this, the mere fact that the man-in-the- street’s beliefs only partially satisfy the conditions in his account in no way casts doubt on its truth.

This is similar to Carl Hempel’s response to a similar objection to his deductive- nomological account of explanation. The relevant part of the account says that e explains why-p only if e entails p, only if e involves a description of a non-statistical law of nature, and only if this description is essential to the entailment. The putative problem is that there are lots of everyday explanations—or, better, things we call “explanations”—that are neither valid nor nomological, such as the claim that the lump of butter melted because it was put in a hot frying pan. Hempel’s response is that his account is supposed to abstract away from the pragmatic aspects of the explanatory practice (which explain why the everyday “explanations” in question

224

are neither deductive nor nomological), in the same way that meta-mathematical proof theory is supposed to abstract away from the pragmatic aspects of the mathematical practice:

Metamathematical proof theory is not intended to give a descriptive account of how mathematicians formulate their proofs. Indeed the formulations that mathematicians actually offer will usually depart to some extent from that called for by rigorous and, as it were, “ideal” metamathematical standards. Yet those standards may be said to exhibit the logical structure and the rationale of mathematical demonstration and to provide criteria for the critical appraisal of particular proofs that might be proposed. (Hempel 1965, 414)

But to call attention to the important pragmatic facets of explanation and to indicate the diverse procedures that may be appropriate in different cases to dispel the perplexity reflected in someone’s quest for an explanation is not to show that a nonpragmatic model of scientific explanation must be hopelessly inadequate, just as analogous arguments concerning the notion of proof cannot show that nonpragmatic models of proof must be sterile and unilluminating. As is well known, the contrary is the case. It is therefore beside the point to complain that the covering-law models do not closely match the form in which working scientists actually present their explanations. Those formulations are generally chosen with a particular kind of audience—and thus with particular pragmatic requirements—in mind. This is true also of the way in which mathematicians present their proofs; but the metamathematical theory of proof quite properly abstracts from these pragmatic considerations. (Ibid., 427-8)

So just as BonJour answers the over-demandingness charge with the claim that the man-in-the- street’s beliefs would be better off epistemically if they satisfied the conditions in his account,

Hempel answers the analogous charge aimed at his deductive-nomological account with the claim that the everyday “explanations” in question would be better off explanatorily if they satisfied the conditions in his account—so that the mere fact that such “explanations” only partially satisfy the conditions in his account in no way casts doubt on its truth.

There is a difference between the two responses, of course. Hempel thinks that the explainers in question, if pressed, would be able to fill in the “explanations” in question so as to better satisfy the deductive-nomological model—if not satisfy it in full. BonJour, in contrast, fully acknowledges that the man-in-the-street, if pressed, would not be able to fill in the justificatory picture so as to fully satisfy his account, or even to satisfy it to a significantly higher degree. In this sense, therefore, Hempel’s account is less at odds with the everyday explanatory practice than is BonJour’s account with the everyday justificatory practice.

225

But what about BonJour’s point that the man-in-the-street’s beliefs would be better off epistemically if they satisfied the conditions in his account? Is it enough to nullify the seemingly relevant point that the man-in-the-street’s beliefs only partially satisfy his account and thus would be unjustified were his account true?

It would seem that, taken just by itself, it is not enough. The mere fact that the man-in- the-street would have better reason for his beliefs were his beliefs to satisfy BonJour’s account in no way shows that the man-in-the-street does not already have good, though less than ideal, reason for his beliefs. It might be that he already has good reason, and that he would have even better reason were his beliefs to satisfy BonJour’s account. Then it would be true that he would

be better off epistemically were his beliefs to satisfy BonJour’s account, but false that his beliefs

are unjustified. In other words, BonJour’s opponents could simply agree with BonJour that the

man-in-the-street would have better reason for his beliefs were his beliefs to satisfy BonJour’s

account, but insist—contra BonJour’s account—that nonetheless the man-in-the-street has good

reason for his beliefs.

In the same way, it would not be enough for me to say that the man-in-the-street’s beliefs would be better off epistemically were they to satisfy EMV. This could be true even if the man- in-the-street already had good reason for them.

The right thing to say, at least initially, is this. The only model on which it is intelligible that S has good reason is EMV, so that any model short of EMV is too permissive. Put another way, the only model (on which there is good reason) for which a philosophical rationale can be given is EMV—such that for any cognizer not satisfying the conditions in EMV, no philosophical sense could be made as to how he has good reason. The extent to which EMV is demanding, then, is perfectly appropriate.

This is perfectly analogous to the discussion in Chapter 4 of whether experiences can serve as reasons. Intuitively, the answer is that they can. But philosophically, the answer is that they cannot. The experientialist foundationalist needs to answer the trilemma challenge, in order

226

to make sense of how experiences could serve as reasons. But so far, his tries at answering it have failed. So though it is quite intuitive that experiences can serve as reasons, no philosophical sense can be made of how they could—and so the mere fact that an account, such as EMV, entails that experiences cannot serve as reasons in no way casts doubt on its truth.

Thus even if (as I am supposing for the sake of argument) the man-in-the-street’s beliefs do no meet the conditions in EMV, EMV should not be rejected. EMV—if the arguments so far given are sound—is the only model for which a satisfying philosophical rationale can be given, and so what should be rejected, if anything should be, is the everyday claims to the effect that so and so has good reason for thinking that such and such.

Let me stress, however, that I am not convinced that the man-in-the-street’s beliefs do not meet the conditions in EMV. It should be kept in mind that my account does not require that the beliefs figuring in the satisfaction of the meta-perspectivalist requirement be conscious, or occurrent. It is fine if they are unconscious, or non-occurrent. And without an account of non- occurrent belief, it is too hard to say, one way or the other, whether the man-in-the-street has the requisite beliefs.

You might be worried that my nonchalance on this matter means that I am singing a different tune (so to speak)—using the word “justification”, or the words “has good reason”, non- standardly. That is, you might think that the man-in-the-street’s perceptual beliefs are paradigmatic instances of justified beliefs, and that any account implying that the man-in-the- street’s perceptual beliefs are unjustified must be either false or not about justification (in the standard sense).

But keep in mind, I agree that pre-philosophically it is quite intuitive that the man-in-the- street’s perceptual beliefs are justified. In other words, I agree that pre-philosophically it is quite intuitive that the man-in-the-street’s perceptual beliefs are beliefs for which he has good reason. I even agree that pre-philosophically it is quite intuitive that perceptual experiences can serve as reasons for beliefs. My point is philosophical. I am saying that, philosophically, experiences

227

could not be what make the man-in-the-street’s perceptual beliefs justified, and that, philosophically, the only beliefs that could do the trick are beliefs satisfying the meta- perspectivalist requirement—so that if it turned out that the man-in-the-street’s perceptual beliefs did not satisfy the meta-perspectivalist requirement, then, philosophically, the man-in-the-street’s beliefs would not be justified. I find nothing objectionable in this.

Consider David Hume’s attack on induction. Would it do to respond to Hume by saying that, in saying that we do not have good reason for thinking that induction is reliable, he is singing a different tune? Surely not. Hume’s point is philosophical. Hume would perfectly well acknowledge that pre-philosophically we have good reason for thinking that induction is reliable, and that pre-philosophically what gives us good reason for thinking this is that we have good reason for thinking that induction has been reliable in the past. Hume’s point is that these pre- philosophical inclinations cannot stand up to philosophical scrutiny.

I would say something similar, if it turned out that the man-in-the-street’s perceptual beliefs did not satisfy the conditions in my account. I would say that though pre-philosophically it is quite intuitive that the man-in-the-street’s perceptual beliefs are justified and though pre-

philosophically it is quite intuitive that the man-in-the-street’s perceptual experiences are what

make his perceptual beliefs justified, these pre-philosophical inclinations cannot stand up to

scrutiny.

6.4 Conclusion

I started by giving a rationale for meta-perspectivalism, arguing that S has good reason for thinking that p obtains only if from S’s perspective it is likely that he is connected to p. I then developed a version of meta-perspectivalism that requires that it be likely from S’s perspective that he is connected to p, and that, unlike other versions of meta-perspectivalism, neither starts a meta-level regress of any kind nor trangresses the coherentist tenet that a belief is justified only if

228

it is supported by a reason. I also rounded out the discussion from the previous chapter, by deriving moderate explanationism from the meta-perspectivalist requirement in EMV.

1 BonJour himself comes to see this. See BonJour 1985, 150-1, and BonJour 1999, 129. In the former, he sticks with coherentism. In the latter, he moves to foundationalism.

2 For the mistake, see Sosa 1991, 203-7.

3 Lycan stresses that it is not required that the meta-belief be true—at least not the parts about reliability. He wants to allow for the demon victim and the brain-in-a-vat, the belief-forming processes of which are unreliable. See Lycan 1996, 22.

4 There are a few passages, such as the one below, in which Lycan seems to deny that coherence4 is required for justification: But I assume Plantinga means his cases to be ones in which the theoretical beliefs are not so impregnably supported and the perception/memory is a good clear one of the sort that does just plain disconfirm a fallible theoretical generalization. If that is so, then as soon as the subject acquires the well-justified perceptual/memory belief, the previous contrary theoretical belief is no longer justified, precisely because it has been disconfirmed. The new perceptual/memory belief is warranted by its tenability and third- or fourth-grade coherence, and it does not lose warrant through its failure to cohere with the theoretical belief, because the latter, having just been refuted by the former, is no longer justified despite its inertial micropersistence. (Lycan 1996, 12, emphasis mine) With the “or”, Lycan seems to be saying both that a non-inferential belief can be justified in virtue of being tenable plus cohering3 (which means that coherence4 is not required), and that a non-inferential belief can be justified in virtue of being tenable plus cohering4 (which means that coherence3 is not required). The key to resolving the apparent conflict between this passage and certain other passages in which Lycan says explicitly say that coherence4 is required is the partial-justification/full-justification distinction. What Lycan is saying, in the passage above, is not that a non-inferential belief can be fully justified in virtue of being tenable plus cohering3, but, instead, that a non-inferential belief can be partially (and perhaps highly) justified in virtue of being tenable plus cohering3.

5 Here Lycan seems to say that a non-inferential belief needs both coherence3 and coherence4: It is the third and fourth kinds of coherence that are not exhibited by what I call ‘wild’ spontaneous beliefs—superstitious forebodings, déjà vu, mild hallucinations, and the like. Such beliefs may be tenable in my sense, but normally they are soon ruled out by their failure to be explained and/or by our having reason to think that they have no reliable source. (Lycan 1996, 8, emphasis mine) With the “and/or”, the claim seems to be that (a) not cohering3 together with not cohering4 is sufficient for not being justified, (b) cohering4 but not cohering3 is sufficient for not being justified (which entails that cohering3 is necessary), and (c) cohering3 but not cohering4 is sufficient for not being justified (which entails that cohering4 is necessary). This is problematic in that there are several passages in which Lycan seems to say that coherence3 is not necessary for justification (even full justification). Here is a suggestion. When Lycan says that “normally they are soon ruled out by their failure to be explained”, he is saying that “normally they are soon ruled out by their failure to be explained in terms of a reliable source”—viz., by their failure to cohere4. The claim, on this reading, is not that normally they are ruled out by their failure to cohere3 and/or their failure to cohere4, but that normally they are ruled out by their failure to cohere4 and/or our having reason to think that they could not cohere4. This reading is nice not just because it resolves the conflict, but also because it frees Lycan of the highly implausible claim that coherence3 is required for full justification. Imagine, for instance, that Jack, whose father is a physicist, sees a vapor trail in a cloud chamber, but that—given his disdain for physics (and science in general)—Jack has no idea as to what explains the vapor trail. The claim that coherence3 is required for full justification gives the wrong result, yielding straight away that Jack’s belief about the vapor trail is unjustified.

229

6 I discussed this aspect of Lycan’s view in Chapter 3. It is the key to Lycan’s attempted rebuttal of the Database Objection.

7 But wait. Suppose that S holds the belief that p obtains, that S is entirely isolated from the world, and that S thinks that he is entirely isolated from the world. From the first result, it would seem to follow that it is not likely that p obtains. And from the second result, it would seem to follow that from S’s perspective it is not likely that p obtains. Yet, to focus just on the first case, it would seem that it is still possible—even given the fact that S is isolated and S holds the belief that p obtains—that it is likely that p obtains. Imagine, for instance, that the claim that p obtains is the claim that it has rained recently, and that, given the fact that the street is wet, it is highly likely that it has rained recently. How could this be? The answer, it seems, is that claims of the form “It is likely that x obtains.” are relational (at least implicitly), saying that it is likely, given y, that x obtains. It would be perfectly intelligible, then, to say both that it is not likely, given just that S is isolated from the world and S holds the belief that it has rained recently, that it has rained recently, and that it is likely, given that the street is wet, that it has rained recently. Think of an instance of argument from authority in which the expert in question is speaking outside of his area of expertise. Then in one sense, it would not be likely that the claim in question is true—viz., the sense in which it is not likely, given just that the expert is speaking outside of his area of expertise, that the claim is true. But in another sense, it might be quite likely that the claim is true, in that there might be other considerations such that, relative to them, it is likely that the claim is true.

8 I am not saying that unlikely things never happen. That is, I am not saying that if it is unlikely that p obtains, then p does not obtain. The claim, rather, is that if, relative to x, it is unlikely that p obtains, then, relative to x, p does not obtain—given that if, relative to x, p obtained, then, relative to x, it would be at least likely that p obtains.

9 The claim “It is not the case that S is connected to p.” should be read so that it is neutral on whether p obtains.

10 See Steup 1996, 12-14.

11 Goldman, for instance, agrees: Millicent has perfectly normal and (we assume) reliable visual powers. She also has a false though justified belief that her visual powers are seriously impaired. Despite this belief, Millicent forms other beliefs that endorse the testimony of her visual appearances. Intuitively, these beliefs are not justified. (Goldman 1986, 110-11) Though she has no meta-beliefs about her vision-based beliefs, Millicent believes that her vision is unreliable, which, according to Goldman, makes it such that her vision-based beliefs are unjustified.

12 Lycan’s meta-perspectivalist requirement differs from this in that Lycan’s is restricted to non-inferential beliefs.

13 You might think that BonJour’s coherentism is faced with this problem, since the observation requirement requires both meta-beliefs and reliability beliefs, and since such beliefs, it would seem, are themselves in need of supporting reasons. Not so, or at least not because of the observation requirement. The requirement is not that there needs to be a meta-belief in order for a non-inferential belief to be justified, but that there needs to be both lots of non-inferential beliefs and reliability beliefs saying that the belief-forming processes involved are reliable. Since this can be satisfied without any meta-meta-beliefs, there is no threat of a regress.

14 (g) itself, BonJour admits, requires that S have access to an a priori argument involving notions such as a priori probability.

230

CHAPTER 7

VERIDICALISM

The veridicality component says, in effect, that, in order for S to have good reason for thinking that p obtains, S’s reasons (for thinking that p obtains) need to be true, and S’s reasons for his reasons need to be true, and so on. This, when coupled with the meta-perspectivalist component, requires (among other things) that S’s view as to the ways in which he is reliably connected to the outside world be correct, which itself, of course, requires that it be the case, in fact, that S is reliably connected to the outside world. And this, in turn, is (at least part of) what makes EMV mind-externalist, since being reliably connected to the outside world is neither mental nor supervenient on the mental.

I argued in Chapter 1 that a coherentist can be a mind-externalist, in that there is nothing, qua coherentist, precluding a coherentist from siding with the mind-externalist. I did not argue, though, that a coherentist should be a mind-externalist. That is the task in this chapter.

I shall start by warding off a rather obvious potential objection to veridicalism, a potential objection that is analogous to the new evil demon problem for reliabilism. Then I shall tout three of veridicalism’s virtues. The first is that it helps in accounting for a second sense in which you can have good reason, a sense that is analogous to soundness in logic. The second is that it helps in illuminating the connection between justification and truth. And the third is that it helps in explaining the difference between epistemic justification and non-epistemic kinds of justification, such as pragmatic justification. And then, last, I shall more fully articulate the version of veridicalism in EMV.

231

7.1 A Potential Objection

The new evil demon problem is initially put forward by Keith Lehrer and Stewart Cohen in “Justification, Truth, and Coherence” (1983), as a problem for reliabilism. Cohen, in a later paper, puts the worry thus:

Imagine that unbeknown to us, our cognitive processes (e.g., perception, memory, inference) are not reliable owing to the machinations of the malevolent demon. It follows on a Reliabilist view that the beliefs generated by those processes are never justified. Is this a tenable result? I maintain not. . . . Now part of what the hypothesis entails is that our experience is just as it would be if our cognitive processes were reliable. Thus, on the demon hypothesis, we should have every reason for holding our beliefs, that we have in the actual world. Moreover since we actually have reason to believe that our cognitive processes are reliable, it follows that in the demon world we would have every reason to believe that our cognitive processes were in fact reliable. We might even imagine that a brilliant philosopher had seemingly demonstrated (a la Descartes of the later meditations) the falsity of the demon hypothesis, to the extent that anyone who could follow the reasoning was (intuitively) justified in accepting the conclusion. It strikes me as clearly false to deny that under these circumstances our beliefs could be justified. (Cohen 1984, 281-2)

You and your counterpart in the demon world are identical mentally—in beliefs, in experiences,

and so on. And so, intuitively, you and your counterpart are also identical justificationally. But

according to reliabilism, there is a difference in justification; your counterpart’s processes, unlike

your processes, are unreliable, and thus your counterpart’s beliefs, unlike your beliefs, are

unjustified. Hence, reliabilism is false. Or so the argument goes.

You might think that, though initially put forward as a problem just for reliabilism, it is

also a problem for any mind-externalist account of having good reason. Cohen, for one, says that

because you and your counterpart in the demon world are identical mentally (i.e., in beliefs, in

experiences, and so on), it follows that you and your counterpart are also identical in terms of

reasons. If, for instance, you have good reason for thinking that there is a tree before you, then

your counterpart too has good reason for thinking that there is a tree before him. But if this is

right, then the facts about having good reason are supervenient on the mental facts—so that all

mind-externalist accounts of having good reason are false.

232

It is certainly true that, in a sense, you are your counterpart are equal in reasons; after all, you and your counterpart have the same reasons (or reason-states). It is also certainly true that the hypothesis that the facts about having good reason are supervenient on the mental facts can perfectly well explain why, in a sense, you and your counterpart are equal in reasons. But it is not true, let alone certainly true, that it follows from any of this that—in all senses—you and your counterpart are equal in reasons. Perhaps there is a sense in which you and your counterpart are unequal in reasons, and perhaps this further sense is mind-externalist. That is, perhaps there is a further sense of having good reason, perhaps in this further sense you are superior in reasons to your counterpart, and perhaps what makes it such that you are thus superior involves a respect in which you differ from your counterpart that is neither mental nor supervenient on the mental.

Consider, for instance, Alvin Goldman’s introduction and use of the distinction between weak justification and strong justification, in answer to Cohen’s charge that the evil demon case shows that reliability is not necessary for justification. The key difference between weakly justified beliefs and strongly justified beliefs is that the latter, but not the former, come from reliable processes:

S’s belief that p, which comes from process M, is weakly justified if (1) M is unreliable, (2) S does not believe that M is unreliable, and (3) S has neither a reliable process nor a process he thinks is reliable that, were he to use it in answering the question of whether M is reliable, would lead him to believe that M is unreliable.

S’s belief that p, which comes from process M, is strongly justified if (1) M is reliable, (2) S does not believe that M is unreliable, and (3) S has neither a reliable process nor a process he thinks is reliable that, were he to use it in answering the question of whether M is reliable, would lead him to believe that M is unreliable.

The answer to Cohen, argues Goldman, is that you and your counterpart in the demon world are equal in weak justification, but not in strong justification:

Under the present theory the treatment of the demon case is straightforward. The victim of the demon fails to have strongly justified beliefs, but he does have weakly justified beliefs. While his beliefs are not well formed, they are blameless and nonculpable. His cognitive processes are not reliable (in his world), but (A) he does not believe that they are unreliable, (B) he has no reliable way of telling this, and (C) there is no method or process he believes to be reliable that would lead him to this conclusion. So on the weak conception of justifiedness, the resulting beliefs are justified. (Goldman 1992, 134)

233

Thus for Goldman there are two senses of justification, one on which you and your counterpart, though different in reliability, are equal, and one on which you and your counterpart, because different in reliability, are unequal.1

Or consider Ernest Sosa’s discussion of internalism and externalism in “Skepticism and the Internal/External Divide”. Sosa explicitly addresses the new evil demon problem, arguing that though there is a justificatory dimension on which you and your counterpart in the demon world are equal, the dimension in question is not exhaustive of the justificatory domain. He starts with a case in which two mental duplicates differ in justification because of differences in how they reasoned to their beliefs:

Compare this: Mary and Jane both arrive at a conclusion C, Mary through a brilliant proof, Jane through a tissue of fallacies. At present, however, they both have forgotten the relevant stretches of their respective reasonings, and each takes herself to have established her conclusion validly. What is more, each of their performances is uncharacteristic, Jane being normally the better logician, while Mary is a normally competent but undistinguished thinker, as they both well know. (Sosa 1999, 152)

The lesson is supposed to be that since, in a sense, Mary’s belief is better justified than Jane’s belief and since the only difference between the beliefs is aetiological, there is an aetiological dimension to justification. He then moves to a case in which two mental duplicates differ in justification because of a difference in the quality of their sources of information:

Suppose a teacher lapses into reasoning that (xn)n = xn+n, and on that basis reports that (22)2 = 24. You, a school child, believe accordingly, just on the teacher’s say-so. (Ibid., 153)

The alternative case differs only in that the teacher reasons correctly. The lesson this time is supposed to be that because, in a sense, your belief in the second case is better justified than your belief in the first case and because the only difference between the beliefs is in the quality of the information sources, there is yet a further dimension to justification—a social aeteliological dimension. And last, he moves to a series of cases in which two mental duplicates differ in justification due to a difference in what produced their beliefs. Here is one such case:

You remember having oatmeal for breakfast, because you did experience having it, and have retained that bit of information through your excellent memory. Your counterpart

234

self-attributes having had oatmeal for breakfast, and may self-attribute remembering that to be so (as presumably do you), but his beliefs are radically wide of the mark, as are an army of affiliated beliefs, since your counterpart was created just a moment ago, complete with all of those beliefs and relevant current experiences. (Ibid., 153-4)

The point here is supposed to be that there is a distinction between subjective justification and objective justification, and that though each belief in each of the latter cases is subjectively justified, only the reliably-produced beliefs are objectively justified.

The question, remember, is whether the new evil demon case poses a problem for mind- externalist accounts of having good reason. The answer, I am suggesting, is that it poses no problem whatsoever, at least not by itself. It could be, as I think, that having good reason in the full, or complete, sense involves having good reason in a weaker sense, and that, though you and your counterpart are equal in reasons in the weaker sense, you and your counterpart are unequal in reasons in the full sense.

7.2 An Analogy to Logic and Practical Reason

There is a distinction in logic between two very different senses in which the premises in an argument can serve as a good reason for the conclusion. I have in mind the distinction between merely valid arguments and sound arguments. Compare:

(1) If Tiger Woods won every tournament on the PGA Tour in 2004, then he won at least one dollar on the PGA Tour in 2004. (2) Tiger Woods won at least one dollar on the PGA Tour in 2004. ------(3) Tiger Woods won every tournament on the PGA Tour in 2004.

(1) If Tiger Woods won every tournament on the PGA Tour in 2004, then he won at least one dollar on the PGA Tour in 2004. (2) Tiger Woods won every tournament on the PGA Tour in 2004. ------(3) Tiger Woods won at least one dollar on the PGA Tour in 2004.

(1) All squares are rectangles. (2) No rectangles are circles. ------(3) No squares are circles.

235

There is a sense in which the premises in the second argument serve as a better reason for the claim that Tiger Woods won at least one dollar on the PGA Tour in 2004 than do the premises in the first argument. Namely, the second argument, but not the first argument, is valid. But, of course, there is more to serving as a good reason, as is evidenced by the third argument. Unlike the second argument, each premise in the third argument is true—making the third argument, but not the second argument, sound.2

But why not stop with validity, in answer to the question of what it would be for an argument to be good? Why require, in addition, that the premises be true? Why is it that having a false premise makes a merely valid argument inferior to a sound argument?

The answer, it seems, is that in giving a merely valid argument, you fail miserably at giving something that makes it likely that the actual world is a world in which the conclusion is true. You give claims that, were they true, would make it likely that the actual world is a world in which the conclusion is true. But since the claims, in fact, are false, what you give has no bearing on whether the actual world is a world in which the conclusion is true. In giving a sound argument, in contrast, you succeed with flying colors. You not only give something that makes it likely that the actual world is a world in which the conclusion is true, you give something that guarantees that the actual world is a world in which the conclusion is true.

There is a similar distinction, I am suggesting, between two very different senses in which you can have good reason. On one hand, there is the sense in which you have something that, were it true, would make it likely that the actual world is a world in which p obtains; your reasons are such that, were they true, it would be likely that p obtains. This is analogous to mere validity. On the other hand, there is the sense in which you have something that, in fact, makes it likely that the actual world is a world in which p obtains; your reasons are such that, because they are true, it is likely that p obtains. This is analogous to soundness. The veridicality requirement in EMV is supposed to secure this second sense of having good reason.

236

There is a similar distinction in the practical-reason literature—similar, that is, to both the distinction in logic and the distinction in epistemology. There are two very different senses in which it can be true that S has good pragmatic reason to X, or in which it can be true that S pragmatically ought to X. One is mind-internalist. The other is mind-externalist.

Geoffrey Grice, for instance, draws the distinction thusly:

The proposition that A ought (objectively) to do x implies that there is a reason for A’s doing x better than . . .; the proposition that A ought (subjectively) to do x implies that A has a reason for doing x better than . . . The distinction between There is . . . and A has . . . has been considered earlier: it is a necessary condition of the truth of the former that a proposition stating that x is in some way in accordance with A’s interest is true; of the latter that A has good reason for judging such a proposition true. The dilemma is resolved by seeing that what has been expressed as ‘He ought to do x and he ought not’ is properly expressed by saying, ‘He ought (subjectively) to do x but (objectively) he ought not to’ with the implication that he has good reason (given the individuality of his situation) for doing x, but there is good reason for his not doing x. (Grice 1967, 25-6, emphasis Grice’s)

On one hand, S has good subjective pragmatic reason to X—or, to put it in Grice’s terms, S has good reason to X—just in case S has good epistemic reason for thinking that X is in his best interests. On the other hand, S has good objective pragmatic reason to X—i.e., there is good reason for S to X—if and only if X is, in fact, in S’s best interests. The difference is that the subjective sense is relative to S’s world-view, while the objective sense is relative not to a world- view (let alone S’s), but to the world itself.

The subjective sense, of course, is the sense analogous to mere validity in logic, and to

the mind-internalist sense of having good reason in epistemology. In giving a merely valid

argument, you give claims that, were they true, would make it likely that the actual world is a

world in which the conclusion is true. In having good mind-internalist epistemic reason for

thinking that p, you believe claims that, were they true, would make it likely that the actual world

is a world in which p obtains. And in having good subjective pragmatic reason to X, you believe

claims that, were they true, would make it likely that X is in your best interests.

The objective sense, in turn, is analogous to soundness in logic, and to the mind- externalist sense of having good reason in epistemology. In giving a sound argument, you give

237

claims that, because they are true, make it likely (by entailing) that the conclusion is true. In having good mind-externalist epistemic reason for thinking that p obtains, you believe claims that, because they are true, make it likely that p obtains. And in having good objective pragmatic

reason to X, there are claims that, because they are true, make it likely (by entailing) that X is in

your best interests.

The analogy could also be put in terms of the mind-internalism/mind-externalism distinction—fully generalized. In the same way that mere validity and having good mind- internalist epistemic reason are mind-internalist, having good subjective pragmatic reason is mind-internalist. If Smith and Jones are identical mentally and thus in terms of beliefs, then

Smith has good epistemic reason for thinking that X is in his best interests just in case Jones has good epistemic reason for thinking that X is in his best interests—so that Smith has good subjective pragmatic reason to X just in case Jones has good subjective pragmatic reason to X.3

And in the same way that soundness and having good mind-externalist epistemic reason are

mind-externalist, having good objective pragmatic reason is mind-externalist. If Smith and Jones

are identical mentally and thus in terms of beliefs, and if, say, in possible world w Smith’s

reaching for the last beer in the fridge would result in much gustatory enjoyment whereas in

possible world w’ Jones’s reaching for the last beer in the fridge would result in death (because in

w’ it contains poison), then in w X would be in Smith’s best interests (or at least might be)

whereas in w’ X would not be in Jones’s best interests—which means that in w Smith would have

good objective pragmatic reason to X whereas in w’ Jones would not have good objective

pragmatic reason to X.

The overall point, to repeat, is just that there is a clear sense in which having good

(epistemic) reason requires having reasons that are true. Yes, there is a sense of goodness—a

sense analogous to mere validity and to having good subjective pragmatic reason—on which truth

is not required. But there is also a sense of goodness—a sense analogous to soundness and

having good objective pragmatic reason—on which truth is required.

238

7.3 Justification and Truth

It is widely thought in epistemology that there is an intimate connection between epistemic justification and truth. It is often said, for instance, that this connection is what distinguishes epistemic justification from other kinds of justification, such as pragmatic justification and moral justification. But what is meant by the claim that there is an intimate connection between justification and truth? Is it that if a belief is justified, then—in virtue of what makes the belief justified—it is highly likely (or guaranteed) that the belief is true? Or is it that, though there could be scenarios in which a belief is justified but not highly likely to be true, having justified beliefs is conducive to having true beliefs? Or is it something else?

Cohen addresses questions such as these in “Justification and Truth” (1984), the paper in which he uses the new evil demon case against reliabilism. He starts with the proposal that if X makes it such that b is justified, then X makes it certain that b is true. The problem, he argues, is that it is far too stringent, not allowing for a sensory experience to justify a belief about the external world. He then moves to the proposal that if X makes it such that b is justified, then X makes it probable that b is true. The evil demon case is supposed to show that this is not the manner in which justification and truth are connected—since, though your counterpart’s beliefs are justified, it is not probable that your counterpart’s beliefs are true. And last, he argues against the proposal that if X makes it such that b is justified, then X entails that S believes that, because of such and such conditions, it is probable that b is true. The alleged shortcoming here is that the man-in-the-street (who has justified beliefs) is lacking in the requisite meta-beliefs—so that the claim is psychologically unrealistic. The upshot, he suggests, is that—suprisingly enough—it is seemingly impossible to make sense of how there could be a tight connection between justification and truth.

I agree that the third construal is a failure, but not for the reason that Cohen gives. The problem is not that it is psychologically unrealistic, but that it requires a regress of meta-beliefs of ever-increasing complexity. S would need to believe that, because of such and such conditions, it

239

is probable that b is true. S would then need to believe that, because of such and such conditions, it is probable that this meta-belief is true. Next, S would need to believe that, because of such and such conditions, it is probable that this meta-meta-belief is true. And so on.

Cohen’s dismissal of the first and second construals, however, is far too quick. The problem, to focus just on his dismissal of the second construal, is that the mere fact that, in one sense, you and your counterpart (in the demon world) are equal in reasons in no way entails that, in all senses, you are your counterpart are equal in reasons, so that the mere fact that, in one sense, you and your counterpart are equal in justification in no way entails that, in all senses, you are your counterpart are equal in justification. It might be that there is a sense in which you and your counterpart are unequal in reasons, and that, thus, there is a sense in which you and your counterpart are unequal in justification. And it might be that this further sense of justification is connected to truth in such a way that if X makes it such that b is justified, then X makes it probable that b is true.

I argued for just this possibility in the previous section, by arguing that there are two very different senses in which you can have good reason, and, thus, two very different senses in which a belief of yours can be justified. On one hand, there is the sense in which you have something that, were it true, would make it likely that the actual world is a world in which p obtains; this is the first sense in which your belief that p obtains can be justified. On the other hand, there is the sense in which you have something that, in fact, makes it likely that the actual world is a world in which p obtains; this is the second sense in which your belief that p obtains can be justified.

There is yet a further oversight on Cohen’s part, an oversight that is pointed out by Earl

Conee in “The Truth Connection” (1992). Conee, like Cohen, begins with the proposal that X makes it such that b is justified only if X makes it certain that b is true, rejecting it, as does

Cohen, on the grounds that it is far too stringent. Next, he considers and attacks the proposal that

X makes it such that b is justified only if X makes it such that b is connected (e.g., causally) to what makes b true. The problem with this, he argues, is that it is open to counterexample—e.g.,

240

both Gettier’s second example, and false but justified beliefs. Then he moves to the proposal that

X makes it such that b is justified only if X makes it such that holding b is a rational means in the pursuit of knowledge. The alleged flaw here is that some justified beliefs are knowledge- defeating, and that some unjustified beliefs are knowledge-promoting. Here is an example of the former:

In some technologically advanced visions of the future, complete brain state detection devices exist. Whatever doxastic states a person is in, these devices can reveal their content to someone else. Such a device might be set by its possessor to link its detection of someone’s having a particular belief to, say, the detonation of an explosion. Suppose that these detection devices are common in the world of our Mr. Jones, whose epistemic goal is to maximize his knowledge. Suppose further that Jones foresees that his believing of a certain proposition, say, the proposition that some basset hounds eat fruit, has been nefariously contrived to cause a deadly explosion in his vicinity. He sees that a brain state detector has been rigged so that this belief will get him blown up. Hence, his believing this proposition is not epistemically justified, given the mild additional assumption that he also sees at least one alternative future for himself that includes his accumulating more knowledge. Yet we may suppose that Jones has just come to possess conclusive evidence from observataion of his own basset hounds for the proposition that some basset hounds eat fruit. Thus, conclusive evidence is not enough to ensure epistemic justification for believing. (Conee 1992, 665)

(Note: I am using “is justified” as he is using “is supported by conclusive evidence”, and I am using “is knowledge-defeating” as he is using “is unjustified”.) And last, he settles on the claim that the truth-connection is evidential—so that X makes it such that b is justified only if X makes it such that b (or, better, the proposition therein) is adequately supported by S’s evidence. This suggestion, which Cohen never even mentions, is exactly right—as I shall explain below.

Consider, again, the two very different senses in which you can have good reason: (1) the sense in which you have something that, were it true, would make it likely that the actual world is a world in which p obtains, and (2) the sense in which you have something that, in fact, makes it likely that the actual world is a world in which p obtains. The veridicality requirement is supposed to account for the second sense. The coherentist requirement, the moderate explanationist requirement, and the meta-perspectivalist requirement, in turn, are supposed to account for the first sense. When your system satisfies the latter such requirements, then from your perspective it is likely that you are connected to p—so that from your perspective it is likely

241

that p obtains, which means that from your perspective it is likely that your belief that p obtains is true.

The overall lesson is that EMV is not only not unable to account for the connection between justification and truth (as Cohen’s arguments would have it), EMV can account for two very different senses in which justification is connected to truth. It yields a mind-internalist connection, in which a justified belief is such that from your perspective it is likely that it is true.

It also yields a mind-externalist connection, in which a justified belief is such that from the world’s perspective (so to speak) it is likely that it is true.

7.4 Non-Epistemic Justification

I said at the start of the previous section that it is widely thought in epistemology that there is an intimate connection between epistemic justification and truth, and that this connection is what distinguishes between epistemic justification and other kinds of justification, such as pragmatic justification and moral justification. I shall now argue that EMV can make perfect sense of this.

Hilary Kornblith makes for a good starting point:

We commonly assume that epistemic evaluation is only one kind of evaluation among many. A candidate belief may fare badly when it comes to epistemic evaluation, but fare well when it comes to various other kinds of evaluation, say, aesthetic or moral. . . . Thus, for example, it seems that I might recognize that having a certain belief would be epistemically ill-advised, and yet have good reason, all things considered, for trying to have the belief. If I could assure world peace by committing some epistemic impropriety, surely it would be worth the price. (Kornblith 2002, 151-2)

The target belief would be non-epistemically justified in some sense (morally, perhaps), given that it would ensure world peace. Yet at the same time, it would be epistemically unjustified, given the impropriety in question—e.g., the fact that he has strong evidence against it, or the fact that he has no evidence in support of it.

With EMV, there are two ways in which epistemic justification differs from other senses of justification—since, with EMV, there are two senses in which epistemic justification is

242

connected to truth. First, there is an internalist way. When your belief that p obtains is epistemically justified, it is likely from your perspective that p obtains. This is not necessarily the case, however, when it is non-epistemically justified. Take the world-peace case, for instance. If, as it might be, the impropriety is that Kornblith has strong evidence against p’s obtaining, then it would not be the case that it is likely from his perspective that p obtains; quite the opposite. Yet

it would still be the case that it is morally justified (pace Clifford). Second, there is an externalist

way in which, with EMV, epistemic justification differs from other senses of justification. If your

belief that p obtains is epistemically justified (in the full sense), it is likely that p obtains. But if

your belief is non-epistemically justified, it might very well not be the case that it is likely that p

obtains. Even if, for instance, Pascal were right in saying that theism is pragmatically justified, it

would not follow that it is likely that, in fact, God exists.

The considerations in favor of the veridicality requirement, to sum up, are quite considerable. First, there is the point that the veridicality requirement accounts for a second sense in which you can have good reason, a sense that is analogous to soundness in logic. Second, there is the point that the veridicality requirement secures a tight connection between justification and truth, a connection over and above the mind-internalist connection that ensures that truth is likely from the subject’s perspective. And third, there is the point that the veridicality requirement secures a second way in which epistemic justification differs from non-epistemic kinds of justification.

7.5 The Veridicality Requirement

The task in this section is to delineate and motivate the specifics of the veridicality requirement, which I glossed in section 1 as saying that, in order for S to have good reason for thinking that p obtains, S’s reasons (for thinking that p obtains) need to be true, and S’s reasons for his reasons need to be true, and so on. This is the last major step in the overall argument for

EMV.

243

The first question that needs to be addressed concerns S’s reason for thinking that p obtains. Is it (supposing that S’s belief that p obtains is both noninferential and satisfies the meta- perspectivalist requirement) S’s belief that he is connected in w to p? Or is it, instead, S’s belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p.

The answer, it would seem, is that S’s reason for thinking that p obtains is his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p. For it would seem that S’s belief that p obtains is involved in his belief that he is connected in w to p, which means that the latter cannot be his reason for the former. But either way, things would end up the same—in that either way, EMV would have the same consequences. Why? Because either way, EMV would require that S be connected in w to p.

Suppose, first, that the right thing to say is that S’s reason for thinking that p obtains is his belief that he is connected in w to p. Then since the veridicality requirement requires that S’s reasons be true, it would be required that S be connected in w to p. Suppose, second, that the right thing to say is that S’s reason for thinking that p obtains is his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p. Then since S’s reason for thinking that he is having a w experience of p is his belief that his w is reliable, his belief that his w works as T says, and his belief that he is connected in w to p, and since the veridicality requirement requires not just that S’s reasons be true but also that S’s reasons for his reasons be true, it would again be required that S be connected in w to p. Thus either way, the veridicality requirement, when coupled with the meta-perspectivalist requirement, requires that S be right that he is connected in w to p, that S be right that his w is reliable, that S be right that his w works as

T says, that S be right that he is having a w experience of p, that S’s reasons for these latter three beliefs be true, that S’s reasons for these reasons be true, and so on.

Why is it required that S’s reasons for his reasons be true, and that S’s reasons for his reasons for his reasons be true? Imagine, for instance, that S is right that his w is reliable, that S

244

is right that his w works as T says, and that S is right that he is having a w experience of p. Then would not it seem that, regardless of whether his reasons for these three beliefs are true, S has good reason for thinking that p obtains? After all, S would have something—viz., his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p—that, because of its truth, makes it likely that p obtains. Why is that not enough?

To see why, consider soundness in logic (again). Though it is sometimes said that sound non-circular arguments are good, it would seem that more is required. Suppose, for instance, that

Jack—with no reason whatsoever—tells Jill that there are at least 43,867 blades of grass in the patch of grass that they are looking at, and that Jack then defends this claim by saying that there are at least 43,868 blades of grass in the patch. Imagine further that, unbeknownst to Jack and

Jill, Jack is right that there are at least 43,868 blades of grass in the patch. Would it follow that

Jack’s argument is good? Yes, in one sense. Jack’s reason is something that, because of its truth,

makes it likely that the actual world is a world in which the conclusion is true. But no, in another sense. Jack’s reason fails in giving either him or her good reason for thinking that the actual world is a world in which the conclusion is true—since neither Jack nor Jill has good reason for thinking that Jack’s reason is true. Imagine next that Jack, in response to Jill’s query as to why she should think that there are at least 43,868 blades of grass in the patch, says that there are at

least 43,869 blades of grass in the patch. Imagine too that this is false, since there are exactly

43,868 blades of grass in the patch. Then the overall argument, though valid, would be unsound,

since the first sub-argument would be unsound. What we want in an argument, it would seem, is

neither soundess supported by nothing nor soundness supported by mere validity, but soundness

supported by soundness supported by soundness . . ..

The same should be said about having good reason. The analogue to soundness supported by nothing would be a scenario in which S’s reasons are true, and in which S’s reasons are themselves without support. The analogue to soundness supported by mere validity would be

245

a scenario in which S’s reasons are true, in which S’s reasons are supported by other beliefs, and in which these other beliefs are false. And the analogue to soundness supported by soundness supported by soundness . . . would be a scenario in which S’s reasons are true, in which these beliefs are supported by other beliefs, in which these other beliefs are true, in which these other beliefs are supported by yet further beliefs (which are true), . . .. This is why the veridicality requirement, when coupled with the meta-perspectivalist requirement, requires not just that S be right that his w is reliable, that S be right that his w works as T says, and that S be right that he is having a w experience of p, but also that S’s reasons in support of these latter three claims be

true, and that S’s reasons in support of these supporting reasons be true, and so on.4

Look at it this way. There are two kinds of good reasons on my account: an internalist kind, and an externalist kind. If S has an internalist reason in support of his belief that p obtains, and if S has no internalist reason in support of this supporting reason, then—given coherentism— the thing to say is that S does not have good internalist reason for thinking that p obtains. In the same way, if S has an externalist reason in support of his belief that p obtains, and if S has no externalist reason in support of this supporting reason, then—again, given coherentism—the thing to say is that S does not have good externalist reason for thinking that p obtains. The motivation in the second case is parallel to the motivation in the first case.

But what if S’s belief that p obtains is not a perceptual belief, and thus is not supported by his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p? What if S's belief that p obtains is an inferential belief, inferred, say, from his perceptual belief that q obtains? What then? Simple. It would need to be the case that q obtains, that S’s reasons for thinking that q obtains are true, that S’s reasons for these reasons are true, and so on. Thus the requirement, more fully put, is that S’s belief that p obtains (whether this belief is non-inferential or inferential) is fully justified only if the beliefs figuring in support of it are true, the beliefs figuring in support of these supporting beliefs are true, and so on—

246

where, because of the meta-perspectivalist requirement, certain beliefs about how S is connected to the outside world need to figure in the picture at some point, and hence also need to be true.

7.6 Is Experience Essential After All?

You might have noticed that the veridicality requirement and the meta-perspectivalist

requirement, as thus stated, together have a rather peculiar consequence. The meta-

perspectivalist requirement is such that (at least in normal perceptual cases) S needs to believe

that he is having a w experience of p. But then the veridicality requirement, in requiring that S’s

reasons be true, implies that it also needs to be the case that S is correct in thinking that he is

having a w experience of p. EMV, then, implies (or at least seems to imply) that full justification

requires the very experiences that the experientialist foundationalist requires. A surprising result,

to say the least.

Three comments. First, there is no sense on EMV in which, as the foundationalist supposes, experiences themselves serve as reasons. On EMV, it is beliefs about experiences—not experiences themselves—that serve as reasons. Second, strictly speaking the meta-perspectivalist requirement does not require that S believe that he is having a w experience of p. The requirement instead is that S have a belief b such that b, together with his belief that his w is reliable and works as T says, makes it likely that he is connected in w to p. I suggested that b be the belief that he is having a w experience of p. But there is no necessity in this; b could be about another part, or product, of w—as long as it is still the case that, relative both to b and to S’s belief that his w (which works as T says) is reliable, it is likely that he is connected in w to p.

This is the point of the qualifier “contingent” in the claim, made in Chapter 4, that EMV imples that there is a contingent respect in which experiences are required for justification. Third, perhaps this is what (or at least part of what) explains the pervasive—though illusory (if I am right)—thought that experiences can, and do, serve as reasons for beliefs. Perhaps it is recognized (correctly) that our experiences are central to the justification of our beliefs. Perhaps

247

it is not recognized that the justification of our beliefs also involves our beliefs about our experiences, and our beliefs about our belief-forming processes, and what not—because, perhaps, these things are somewhat beneath the surface (so to speak). And perhaps, last, it is concluded

(mistakenly) that the only things involved in the justification of our beliefs are our experiences.5

7.7 Is Truth Required for Justification?

You might be worried that with the veridicality requirement EMV has the consequence that justification requires truth, so that S’s belief that p obtains is unjustified if p does not obtain.

Suppose, the worry would start, that S’s belief that p obtains is supported by his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p. Then with the veridicality requirement EMV would require that these supporting beliefs be true, and that, further, the beliefs in support of these supporting beliefs be true. But since the

belief that he is having a w experience of p is supported, in part, by his belief that he is connected in w to p, EMV would further require that, in fact, he be connected in w to p—which itself, of course, would require that p obtain. The connection between EMV-ist justification and truth, then, is too tight—or at least so goes the worry.

It is certainly true, and for the reasons just given, that if S’s belief that p obtains is

supported by his belief that his w is reliable, his belief that his w works as T says, and his belief

that he is having a w experience of p, then (at least in normal cases) EMV has the consequence

that S’s belief that p obtains is justified only if p obtains.6 It is certainly not true, however, that

EMV has the consequence that justification requires truth. Suppose not that S’s belief that p obtains is supported by his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p, but that, instead, it is supported by his belief that q obtains and his belief that p’s obtaining is the best explanation of q’s obtaining. Then though the veridicality requirement would require that q obtain, that the beliefs in support of his belief

248

that q obtains be true, and so on, the veridicality requirement would not require that p obtains. It might be that p’s obtaining is the best explanation, but that p does not obtain.

You might still be worried, though. For it is true, as I acknowledged in the previous paragraph, that if S’s belief that p obtains is a perceptual belief and is thus supported by his belief that his w is reliable, his belief that his w works as T says, and his belief that he is having a w experience of p, then (at least in normal cases) justification requires truth. And this, you might think, is too stringent.

But notice two things. First, EMV still allows for an important sense in which false perceptual beliefs can be justified—viz., the mind-internalist sense. Second, the sense in which false perceptual beliefs are unjustified is perfectly analogous to the sense in which unsound but valid arguments are bad. S’s grounds for thinking that p obtains are that his w is reliable, that his w works as T says, and that he is having a w experience of p. S’s grounds for thinking that he is having a w experience of p are that his w is reliable, that his w works as T says, and that he is connected in w to p. But if p is false, then it is false that he is connected in w to p, so that part of his grounds for believing that he is having a w experience of p is false.

7.8 Is Just One False Belief Enough to Undermine Justification?

I said above that, given the veridicality requirement, S’s belief that p obtains (whether this belief is non-inferential or inferential) is fully justified only if the beliefs figuring in support of it are true, the beliefs figuring in support of these supporting beliefs are true, and so on— where, because of the meta-perspectivalist requirement, certain beliefs about how S is connected to the outside world need to figure in the picture at some point, and hence also need to be true.

But what if just one of the many beliefs involved in this rather large collection of beliefs is false?

Should we really say that S’s belief that p obtains is unjustified?

Imagine, for instance, that S’s beliefs satisfy the meta-perspectivalist requirement, but

that some such beliefs do so partly in virtue of S’s false theory V of how vision works. Suppose

249

that V says that there are five big steps, or stages, in vision (each of which has several sub-steps), and that, though close, V is off the mark on the fourth step. Then would not it follow that S’s vision-based beliefs do not satisfy the veridicality requirement, given that such beliefs—to satisfy the meta-perspectivalist requirement—depend on S’s belief that his vision works as V says, and given that V, in fact, is false? And would not this be bad for EMV, since it would then follow that EMV implies that the scientifically unsophisticated—whose rudimentary theory of how vision works is true—is epistemically superior to the scientifically sophisticated—whose sophisticated theory of how vision works, suppose, is strictly speaking false?

Or imagine, instead, that (1) in satisfying the meta-perspectivalist requirement S believes that his vision is reliable, S believes that q obtained at t and that he was visually connected at t to q, S believes that r obtained at t’ and that he was visually connected at t’ to r, etc., (2) S believes that his vision’s being reliable is what explains this fine track-record, and (3) q did not obtain at t, and so S’s belief that he was visually connected at t to q is false. Would not this isolated falsity mean that S’s belief that his vision is reliable does not satisfy the veridicality requirement?

Would not this, in turn, mean that S’s belief that p obtains does not satisfy the veridicality requirement? And, again, would not this be bad for EMV?

Yes, both such alleged consequences would be bad for my view—were they more than just alleged consequences. But neither such alleged consequence is more than that.

Let me start with an analogous case, which should prove helpful in explaining the veridicality requirement, and, in turn, in explaining why it has neither of the two alleged consequences from above. Suppose that argument G is valid, that every premise in G but the first premise is true, and that the first premise is not essential to G’s validity; G would be valid even if the first premise were removed. Is G a good argument? It would seem so, even though, strictly speaking, G is unsound. For though G has a false premise, G also has true premises that together entail the conclusion.

250

Something similar is true of the two cases above. The scientist has a sophisticated scientific theory V of how vision works, a theory that figures in the satisfaction of the meta- perspectivalist requirement (for his vision-based beliefs). But the scientist did not need a theory of such sophistication in order to satisfy the meta-perspectivalist requirement. It would have been enough if he had believed just that vision is a cognitive process, that the key sensory organ in vision is the eye, and that vision involves visual experiences; he would have still had good reason for thinking that he is visually connected to p. In the same way, S did not need his belief that q obtained at t and that he was visually connected at t to q. It would have been enough if he had believed just that r obtained at t’ and that he was visually connected at t’ to r, that s obtained at t’’ and that he was visually connected at t’’ to s, and so on; he would have still had good reason for thinking that his vision has had a nice track-record, and that the best explanation of this fine track-record is that his vision is reliable. The point, then, is that the veridicality requirement requires not that all the beliefs figuring in the satisfaction of the meta-perspectivalist requirement be true, but just that there be enough true beliefs to satisfy the meta-perspectivalist requirement.

7.9 Conclusion

The veridicality requirement, to sum up, says that S’s belief that p obtains (whether this belief is non-inferential or inferential) is fully justified only if the beliefs figuring in support of it are true, the beliefs figuring in support of these supporting beliefs are true, and so on—where, because of the meta-perspectivalist requirement, certain beliefs about how S is connected to the outside world need to figure in the picture at some point, and hence also need to be true. The first consideration in its favor is that it accounts for a second sense in which you can have good reason, a sense that is analogous to soundness in logic. The second such consideration is that it secures a tight connection between justification and truth, a connection over and above the mind- internalist connection that ensures that truth is likely from the subject’s perspective. And the

251

third is that it secures a second way in which epistemic justification differs from non-epistemic kinds of justification.

1 Goldman’s initial response is much different. He tries to accommodate the intuition that your counterpart’s beliefs are just as justified as are your beliefs, by relativizing the relevant reliability facts to normal worlds—viz., worlds consistent with our general beliefs about the actual world. This is supposed to help since your counterpart’s belief-forming processes are identical to yours, and since your belief-forming processes are reliable in normal worlds. See Goldman 1986, 107-9.

2 There is also the distinction between valid arguments and cogent arguments. It too is a distinction between two very different senses in which the premises in an argument can serve as a good reason for the conclusion.

3 I am assuming here that Grice has in mind the mind-internalist sense of having good reason.

4 Given the nature of the meta-perspectivalist requirement, this regress “stops” fairly quickly—by circling back around. This, as I said in the previous chapter, is one of the chief virtues of the meta-perspectivalist requirement in EMV—especially in comparison to the meta-perspectivalist requirements in rival accounts.

5 Things are exactly parallel for process reliabilism. It allows a role for experiences in justification, in that it allows a role for vision, for example, in justification (given that vision is a reliable belief-forming process), and in that vision involves visual experiences. It does not imply that experiences can serve as reasons, however; it allows a role for experiences in justification because of the contingent fact that experiences are involved in reliable belief-forming processes. And its adherents could point to this contingent fact in explaining away the pervasive thought that experience, not reliability, is the key to justification.

6 The qualifier “at least in normal cases” is needed for cases in which S’s belief that he is having a w experience of p is supported not by his belief that he is connected in w to p, but by a different belief—and one that is not itself supported by his belief that he is connected in w to p. In these cases, the veridicality requirement could be satisfied even if p did not obtain.

252

BIBLIOGRAPHY

Achinstein, P. 1983. The Nature of Explanation. Oxford: Oxford University Press.

______. 2001. The Book of Evidence. Oxford: Oxford University Press.

Adler, J. 1990. “Conservatism and Tacit Belief”. Mind 99 (396): 559-70.

Alston, W. 1983. “What’s Wrong with Immediate Knowledge?”. Synthese 55: 73-95.

______. 1985. “Concepts of Epistemic Justification”. Monist 68 (2): 57-89.

______. 1986. “Epistemic Circularity”. Philosophy and Phenomenological Research 47: 1- 30.

______. 1986. “Internalism and Externalism in Epistemology”. In Epistemology: Internalism And Externalism, edited by H. Kornblith, 68-110. Malden: Blackwell Publishers.

______. 1988. “An Internalist Externalism”. Synthese 74: 227-45.

______. 1993. The Reliability of Sense Perception. Ithaca: Cornell University Press.

Annis, D. 1982. “Knowledge and Inference to the Best Explanation—A Reply”. Philosophia 12: 77-9.

Armstrong, D. 1973. Belief, Truth, and Knowledge. Cambridge: Cambridge University Press.

Bales, R. 1971. “Act-Utilitarianism: Account of Right-Making Characteristics or Decision- Making Procedure?”. American Philosophical Quarterly 8 (3): 257-65.

Bartelborth, T. 1996. “Scientific Explanation”. In Structuralist Theory of Science: Focal Issues, New Results, edited by C. Moulines, 23-43. New York: de-Gruyter.

______. 1999. “Coherence and Explanations”. Erkenntnis 50: 209-224.

Blanshard, B. 1939. The Nature of Thought. London: George Allen & Unwin.

BonJour, L. 1985. The Structure of Empirical Knowledge. Cambridge: Harvard University Press.

253

______. 1986. “A Reconsideration of the Problem of Induction”. Philosophical Topics XIV, 1: 93-124.

______. 1989. “Replies and Clarifications”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 276-92. Dordrecht: Kluwer Academic Publishers.

______. 1989. “Reply To Steup”. Philosophical Studies 55: 57-63.

______. 1992. “Externalism/Internalism”. In A Companion to Epistemology, edited by E. Sosa and J. Dancy, 132-6. Cambridge: Blackwell.

______. 1997. “Haack on Justification and Experience”. Synthese 112: 13-23.

______. 1999. “The Dialectic of Foundationalism and Coherentism”. In The Blackwell Guide to Epistemology, edited by J. Greco and E. Sosa, 117-42. Oxford: Blackwell.

______. 1999. “Foundationalism and the External World”. Philosophical Perspectives 13: 229-49.

______. 2002. Epistemology: Classic Problems and Contemporary Responses. Lanham: Rowman & Littlefield Publishers.

BonJour, L. and Sosa, E. 2003. Epistemic Justification: Internalism vs. Externalism, Foundations vs. Virtues. Malden: Blackwell Publishing.

Bosanquet, B. 1920. Implication and Linear Inference. London: Macmillan and Co.

Brewer, B. 1999. Perception and Reason. Oxford: Clarendon Press.

______. 2001. “Precis of Perception and Reason, and response to commentators (Naomi Eilan, Richard Fumerton, Susan Hurley, and Michael Martin)”. Philosophy and Phenomenological Research 63: 405-64.

Bromberger, S. 1962. “An Approach to Explanation”. In Analytical Philosophy—Second Series, edited by R. Butler, 72-105. Oxford: Blackwell.

______. 1966. “Why-Questions”. In Mind And Cosmos, edited by R. Colodny, 86-111. Pittsburgh: University Of Pittsburgh Press.

Carroll, L. 1995. “What the Tortoise Said to Achilles”. Mind 104 (416): 691-3.

Carter, K. 2004. A First Course in Logic. New York: Pearson Longman.

Chalmers, D. 2003. “The Content and Epistemology of Phenomenal Belief”. In Consciousness: New Philosophical Perspectives, edited by Quentin Smith and Aleksandar Jokic, 220-72. Oxford: Clarendon Press.

Chisholm, R. 1980. “A Version of Foundationalism”. Midwest Studies In Philosophy, 5, 543- 64.

254

______. 1982. The Foundations of Knowing. University of Minnesota Press.

______. 1989. Theory of Knowledge, 3rd edition. Prentice Hall.

Christensen, D. 1994. “Conservatism in Epistemology”. Nous 28 (1): 69-89.

Cohen, S. 1984. “Justification and Truth”. Philosophical Studies 46: 279-95.

Conee, E. 1992. “The Truth Connection”. Philosophy and Phenomenological Research LII (3): 657-69.

______. 1995. “Isolation and Beyond”. Philosophical Topics 23, 1: 129-46.

Copi, I. and Cohen, C. Introduction to Logic, tenth edition. Upper Saddle River: Prentice Hall.

Cornman, J. 1977. “Foundational versus Nonfoundational Theories of Justification”. American Philosophical Quarterly 14 (4): 287-97.

Crumley, J. 1999. An Introduction to Epistemology. Mountain View, California: Mayfield Publishing Company.

Davidson, D. 2000. “A Coherentist Theory of Truth and Knowledge”. In Epistemology: An Anthology, edited by E. Sosa and J. Kim, 154-63. Oxford: Blackwell Publishers Ltd.

Davis, W. and Bender, J. 1989. “Fundamental Troubles with the Coherence Theory”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 52-68. Dordrecht: Kluwer Academic Publishers.

Day, T. 1989. “Circularity, Non-Linear Justification and Holistic Coherentism”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 134-41. Dordrecht: Kluwer Academic Publishers.

Dretske, F. 1999. Knowledge and the Flow of Information. United States: CSLI Publications.

Ennis, R. 1968. “Enumerative Induction and Best Explanation. Journal of Philosophy 65 (18): 523-9.

Ewing, A. C. 1934. Idealism. London: Methuen.

Fales, E. 1996. A Defense of the Given. Lanham, Maryland: Rowman & Littlefield Publishers.

Feldman, R. 1989. “Lehrer’s Coherence Theory of Knowledge”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 69-76. Dordrecht: Kluwer Academic Publishers.

______. 2000. “The Ethics of Belief”. Philosophy and Phenomenological Research 60 (3): 667-95.

255

______. 2003. Epistemology. Upper Saddle River, New Jersey: Prentice Hall.

Feldman, R. and Conee, E. 1985. “Evidentialism”. Philosophical Studies 48: 15-34.

______. 2001. “Internalism Defended”. In Epistemology: Internalism and Externalism, edited by H. Kornblith, 231-60. Malden: Blackwell Publishers.

Firth, R. 1964. “Coherence, Certainty, and Epistemic Priority”. Journal of Philosophy 61 (19): 545-57.

Fogelin, R. and Armstrong, W. 2001. Understanding Arguments: An Introduction to Informal Logic, sixth edition. Wadsworth.

Foley, R. 1979. “Justified Inconsistent Beliefs”. American Philosophical Quarterly 16 (4): 247- 57.

______. 1982. “Epistemic Conservatism”. Philosophical Studies 43: 165-82.

______. 1984. “Epistemic Luck and the Purely Epistemic”. American Philosophical Quarterly 21 (2): 113-24.

Friedman, M. 1974. “Explanation and Scientific Understanding”. Journal of Philosophy 71: 5- 19.

Fumerton, R. 1980. “Induction and Reasoning to the Best Explanation”. Philosophy of Science 47: 589-600.

______. 1995. Metaepistemology and Skepticism. Lanham: Rowman & Littlefield Publishers.

______. 2002. “Bill Brewer, Perception and Reason”. Nous 36 (3): 509-22.

Giere, R. 1984. Understanding Scientific Reasoning, second edition. New York: Holt, Rinehart and Winston.

Goldman, Alan. 1988. Empirical Knowledge. Berkeley: University Of California Press.

______. 1989. “BonJour’s Coherentism”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 125-33. Dordrecht: Kluwer Academic Publishers.

Goldman, Alvin. 1967. “A Causal Theory of Knowing”. Journal of Philosophy 64: 357-72.

______. 1979a. “Varieties of Cognitive Appraisal”. Nous 13: 23-38.

______. 1979b. “What is Justified Belief?”. In Justification and Knowledge, edited by G. Pappas, 1-23. Dordrecht: D. Reidel Publishing Company.

______. 1986. Epistemology and Cognition. Cambridge: Harvard University Press.

______. 1988. “Strong and Weak Justification”. Philosophical Perspectives 2: 51-69.

256

______. 1989. “BonJour’s The Structure of Empirical Knowledge”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 105-14. Dordrecht: Kluwer Academic Publishers.

______. 1992. Liaisons: Philosophy Meets the Cognitive and Social Sciences. Cambridge: MIT Press.

______. 1999. “Internalism Exposed”. In Epistemology: Internalism and Externalism, edited by H. Kornblith, 207-30. Malden: Blackwell Publishers.

______. 2001. “The Internalist Conception of Justification”. In Epistemology: Internalism and Externalism, edited by H. Kornblith, 36-67. Malden: Blackwell Publishers.

Goldman, Holly. 1978. “Doing the Best One Can”. In Values and Morals, edited by Alvin Goldman and Jaegwon Kim. Dordrecht: Reidel.

Goldstick, D. 1971. “Methodological Conservatism”. American Philosophical Quarterly 8 (2): 186-91.

Greco, J. 2000. Putting Skeptics in Their Place: The Nature of Skeptical Arguments and Their Role in Philosophical Inquiry. Cambridge: Cambridge University Press.

Grice, G. 1967. The Grounds of Moral Judgement. Cambridge: Cambridge University Press.

Grobler, A. 2000. “Explanation and Epistemic Virtue”. Poznan Studies in the Philosophy of the Sciences and the Humanities 72: 61-88.

Haack, S. 1993. Evidence and Inquiry: Towards Reconstruction in Epistemology. Oxford: Blackwell Publishers.

______. 1997. “Reply To BonJour”. Synthese 112: 25-35.

Hallsten, H. 1999. “Deductive Chauvinism”. Synthese 120: 49-59.

Harman, G. 1965. “The Inference to the Best Explanation”. Journal of Philosophy 74: 88-95.

______. 1968. “Enumerative Induction as Inference to the Best Explanation”. Journal of Philosophy 65 (18): 529-33.

______. 1973. Thought. Princeton: Princeton University Press.

______. 1975. “Wilfrid Sellars’ Theory of Induction”. In Action, Knowledge, and Reality: Critical Studies in Honor of Wilfrid Sellars, edited by Hector-Neri Casteneda, 265-275. Indianapolis: Bobbs-Merrill.

______. 1978. “Using Intuitions about Knowledge to Study Reasoning: A Reply to Williams”. Journal of Philosophy 75: 433-38.

______. 1986. Change in View: Principles of Reasoning. Cambridge: The MIT Press.

257

______. 1994. “Simplicity as a Criterion for Deciding what Hypotheses to take Seriously”. In Grue: The New Riddle of Induction, edited by D. Stalker, 153-71.

______. 2003. “Skepticism and Foundations”. In The Skeptics: Contemporary Essays, edited by Steven Luper, 1-11. Ashgate: Aldershot.

Heck, R. 2000. “Nonconceptual Content and the ‘Space of Reasons’”. Philosophical Review, 109, 4, 2000.

Hempel, C. 1965. “Aspects of Scientific Explanation”. In Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, 331-496. New York: The Free Press.

Henderson, D. 1995. “One Naturalized Epistemological Argument Against Coherentist Accounts of Empirical Knowledge”. Erkenntnis 43: 199-227.

Henderson, D. and Horgan, T. 2000. “Iceberg Epistemology”. Philosophy and Phenomenological Research 61 (3): 497-535.

Hume, D. 1993. An Enquiry Concerning Human Understanding. Indianapolis: Hackett Publishing Company.

Hurley, P. 2003. A Concise Introduction to Logic, eighth edition. Belmont: Wadsworth.

Josephson, J. 2000. “Smart Inductive Generalizations are Abductions”. In Abduction and Induction, edited by P. Flach and A. Kakas, 31-44. Netherlands: Klewer Academic Publishers.

Kim, K. 1993. “Internalism and Externalism in Epistemology”. American Philosophical Quarterly 30 (4): 303-16.

Kitcher, P. 1976. “Explanation, Conjunction, and Unification”. Journal of Philosophy 73: 207- 12.

______. 1981. “Explanatory Unification”. Philosophy of Science 48: 507-31.

______. 1986. “Projecting the Order of Nature”. In Kant’s Philosophy of Science, edited by R. Butts, 201-35. Dordrecht: D. Reidel Publishing Company.

______. 1989. “Explanatory Unification and the Causal Structure of the World”. In Scientific Explanation, edited by P. Kitcher and W. Salmon, 410-505. Minneapolis: University of Minnesota Press.

Klein, P. 1985. “The Virtues of Inconsistency”. Monist 52: 105-35.

Koppelberg, D. 1998. “Foundationalism and Coherentism Reconsidered”. Erkenntnis 49: 255- 83.

______. 1999. “Justification and Causation”. Erkenntnis 50: 447-62.

______. 2000. “Quine and Davidson on the Structure of Empirical Knowledge”. Poznan Studies in the Philosophy of the Sciences and the Humanities 70: 101-19.

258

Korcz, K. 1997. “Recent Work on the Basing Relation”. American Philosophical Quarterly, 34, 2, 1997: 171-91.

Kornblith, H. 1988. “How Internal Can You Get?”. In Epistemology: Internalism and Externalism, edited by H. Kornblith, 111-25. Malden: Blackwell Publishers.

______. 1989. “The Unattainability of Coherence”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 207-14. Dordrecht: Kluwer Academic Publishers.

______. 2002. Knowledge and its Place in Nature. Oxford: Clarendon Press.

Kvanvig, J. 1995. “Coherentism: Misconstrual and Misapprehension”. Southwest Philosophy Review 11 (1): 159-68.

______. 1997. “In Defense of Coherentism”. Journal of Philosophical Research 22: 299-306.

Kvanvig, J. and Riggs, W. 1992. “Can a Coherence Theory Appeal to Appearance States”. Philosophical Studies 67: 197-217.

Ladyman, J., Douven, E., Horsten, L., and Van Fraassen, B. 1997. A Defence of van Fraassen’s Critique of Abductive Inference”. Philosophical Quarterly 47: 306-21.

Layman, C. 2005. The Power of Logic, third edition. Boston: McGraw-Hill.

Lehrer, K. 1970. “Justification, Explanation, and Induction”. In Induction, Acceptance, and Rational Belief, edited by M. Swain, 100-33. Dordrecht: D. Reidel Publishing Company.

______. 1974. Knowledge. Oxford: Clarendon Press.

______. 1986. “The Coherence Theory of Knowledge”. Philosophical Topics 14 (1): 5-25.

______. 1989. “Coherence and the Truth Connection: A Reply to My Critics”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 253-75. Dordrecht: Kluwer Academic Publishers.

______. 1996. “Proper Function versus Systematic Coherence”. In Warrant in Contemporary Epistemology: Essays in Honor of Plantinga’s Theory of Knowledge, edited by J. Kvanvig, 25-45. Lanham: Rowman & Littlefield Publishers, Inc.

______. 1997. Self-Trust: A Study of Reason, Knowledge and Autonomy. Oxford: Clarendon Press.

______. 2000. Theory of Knowledge. Boulder: Westview Press.

Lehrer, K. and Cohen, S. 1983. “Justification, Truth, and Coherence”. Synthese 55, 191-207.

Lemos, N. 1989. “Epistemic Priority and Coherence”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence

259

BonJour, with Replies, edited by J. Bender, 178-87. Dordrecht: Kluwer Academic Publishers.

Lewis, C. 1929. Mind and the World Order: Outline of a Theory of Knowledge. New York: Dover Publications.

______. 1946. An Analysis of Knowledge & Valuation. La Salle, Illinois: Open Court.

______. 1952. “The Given Element in Empirical Knowledge”. Philosophical Review 61: 168-75.

Lewis, D. 1986. “Causation”. In Philosophical Papers, volume 2, 159-213. Oxford: Oxford University Press.

______. 1986. “Causal Explanation”. In Philosophical Papers, volume 2, 214-240. Oxford: Oxford University Press.

Lipton, P. 1991. “Constrastive Explanation and Causal Triangulation”. Philosophy of Science 58: 687-97.

______. 1991. Inference to the Best Explanation. London: Routledge.

______. 1993. “Is the Best Good Enough?”. Proceedings of the Aristotelian Society 93: 89- 104.

Lycan, W. 1985. “Epistemic Value”. Synthese 64: 137-64.

______. 1988. Judgement and Justification. New York: Cambridge University Press.

______. 1996. “Plantinga and Coherentisms”. In Warrant in Contemporary Epistemology: Essays in Honor of Plantinga’s Theory of Knowledge, edited by J. Kvanvig, 3-23. Lanham: Rowman & Littlefield Publishers, Inc.

______. 2002. “Explanation and Epistemology”. In The Oxford Handbook of Epistemology, edited by P. Moser, 408-33. Oxford: Oxford University Press.

Mackie. J. 1980. The Cement of the Universe: A Study of Causation. Oxford: Clarendon Press.

Martin, M. 1998. “An Eye Directed Outward”. In Knowing Our Own Minds, edited by Crispin Wright, Barry Smith, and Cynthia Macdonald, 99-121. Oxford: Clarendon Press.

McDowell, J. 1994. Mind and World. Cambridge: Harvard University Press.

______. 1994. “The Content of Perceptual Experience”. Philosophical Quarterly 44: 190- 205.

McGrew, L. and McGrew, T. 2000. “What’s Wrong with Epistemic Circularity”. Dialogue 39: 219-39.

260

McGrew, T. 2003. “A Defense of Classical Foundationalism”. In The Theory of Knowledge: Classical and Contemporary Readings, edited by L. Pojman, 194-206. Belmont: Wadsworth.

Meyers, R. 1981. “Sellars’ Rejection of Foundations”. Philosophical Studies 39: 61-78.

Mill, J. 1874. A System of Logic. New York: Harper & Brothers.

Moser, P. 1988. “Foundationalism, the Given, and C. I. Lewis”. History of Philosophy Quarterly 5 (2): 189-204.

______. 1988. “How Not to be a Coherentist”. Analysis 48: 166-7.

______. 1988. “Internalism and Coherentism: A Dilemma”. Analysis 48 (4): 161-3.

______. 1989. “Lehrer’s Coherentism and The Isolation Objection”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 29-37. Dordrecht: Kluwer Academic Publishers.

______. 1989. Knowledge and Evidence. Cambridge: Cambridge University Press.

Nozick, R. 1981. Philosophical Explanations. Cambridge: Belknap Press.

Pappas, G. 1980. “Lost Justification”. Midwest Studies in Philosophy 5: 127-34.

______. 1992. “Explanatory Coherence and Data Sentences”. In Frontiers in American Philosophy, edited by W. Burch and J. Saatkamp, 166-75. College Station: Texas A&M University Press.

Peacocke, C. 1998. “Conscious Attitudes, Attention, and Self-Knowledge”. In Knowing Our Own Minds, edited by Crispin Wright, Barry Smith, and Cynthia Macdonald, 63-98. Oxford: Clarendon Press.

______. 2004. The Realm of Reason. Oxford: Clarendon Press.

Pitt, J. 1980. “Hempel versus Sellars on Explanation”. Dialectica 34 (2): 95-120.

______. 1981. “The Role of Inductive Generalizations in Sellars’ Theory of Explanation”. Theory and Decision 13: 345-56.

Plantinga, Alvin. 1988. “Chisholmian Internalism”. In Philosophical Analysis, edited by D. Austin, 127-51.

______. 1993. Warrant: the Current Debate. Oxford: Oxford University Press.

______. 1993. Warrant and Proper Function. Oxford: Oxford University Press.

Pojman, L. 2001. What Can We Know? Belmont, California: Wadsworth.

261

Pollock, J. and Cruz, J. 1999. Contemporary Theories of Knowledge. Lanham: Rowman & Littlefield Publishers.

Psillos, S. 1996. “On Van Fraassen’s Critique of Abductive Reasoning”. Philosophical Quarterly 46 (182): 31-47.

______. 1997. “How Not to Defend Constructive Empiricism: A Rejoinder”. Philosophical Quarterly 47: 369-72.

______. 1999. Scientific Realism: How Science Tracks Truth. London: Routledge.

______. 2000. “Abduction and Induction”. In Abduction and Induction, edited by P. Flach and A. Kakas, 59-74. Netherlands: Klewer Academic Publishers.

______. 2000. “The Present State of the Scientific Realism Debate”. British Journal for the Philosophy of Science 51: 705-28.

Quine, W. 1953. “Two Dogmas of Empiricism”. In From a Logical Point of View, 20-46. Cambridge: Harvard University Press.

Quine, W. and Ullian, J. 1970. The Web of Belief. New York: Random House.

Railton, P. 1978. “A Deductive-Nomological Model of Probabilistic Explanation”. Philosophy of Science 45: 206-26.

______. 1980. Explaining Explanation: A Realist Account of Scientific Explanation and Understanding. Ph.D. dissertation: Princeton University.

______. 1981. “Probability, Explanation, and Information”. Synthese 48: 233-56.

Rappaport, S. 1996. “Inference to the Best Explanation: Is it Really Different from Mill’s Methods?”. Philosophy of Science 63: 65-80.

Rescher, N. 1974. “Foundationalism, Coherentism, and the Idea of Cognitive Systematization”. Journal of Philosophy 71 (19): 695-708.

______. 1979. “Reply to BonJour”. In The Philosophy of Nicholas Rescher: Discussion and Replies, edited by E. Sosa, 173-4. Dordrecht, D. Reidel Publishing Company.

Resnik, D. 1994. “Epistemic Value: Truth or Explanation”. Metaphilosophy 25 (4): 348-61.

Riggs, W. 1997. “The Weakness of Strong Justification”. Australasian Journal of Philosophy 75 (2): 179-89.

Russow, L. and Curd, M. 1989. Principles of Reasoning. New York: St. Martin’s Press.

Salmon, M. 2002. Introduction to Logic and Critical Thinking. United States: Wadsworth.

Salmon, W. 1970. “Statistical Explanation”. In Nature and Function of Scientific Theories, edited by R. Colodny, 173-231. Pittsburgh: University Of Pittsburgh Press.

262

______. 1984. Scientific Explanation and the Causal Structure of the World. Princeton: Princeton University Press.

______. 1989. “Four Decades of Scientific Explanation”. In Scientific Explanation, edited by P. Kitcher and W. Salmon, 3-219. Minneapolis: University of Minnesota Press.

Schantz, R. 1999. “The Role of Sensory Experience in Epistemic Justification: A Problem for Coherentism”. Erkenntnis 50: 177-91.

Sellars, W. 1963. “Empiricism and the Philosophy of Mind”. In Science, Perception, and Reality, 127-96. Astascadero: Ridgeview Publishing Company.

______. 1973. “Givenness and Explanatory Coherence”. Journal of Philosophy 70: 612-24.

______. 1975. “Epistemic Principles”. In Action, Knowledge, and Reality: Critical Studies in Honor of Wilfrid Sellars, edited by H. Castaneda, 332-47. Indianapolis: The Bobbs- Merril Company.

______. 1979. “More on Givenness and Explanatory Coherence”. In Justification and Knowledge, edited by G. Pappas, 169-82. Dordrecht: D. Reidel Publishing Company.

Senor, T. 1995. “Harman, Negative Coherentism, and the Problem of Ongoing Justification”. Philosophia 24: 271-94.

Silvers, S. 1989. “Coherence, Observation, and the Justification of Empirical Belief”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 168-77. Dordrecht: Kluwer Academic Publishers.

Sklar, L. 1975. “Methodological Conservatism”. Philosophical Review 84: 374-400.

Sosa, E. 1986. “‘Circular’ Coherence and ‘Absurd’ Foundations”. In Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, edited by E. LePore, 387-97. Oxford: Blackwell.

______. 1989. “Equilibrium in Coherence?”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 242-51. Dordrecht: Kluwer Academic Publishers.

______. 1991. Knowledge in Perspective. Cambridge: Cambridge University Press.

______. 1996. “Plantinga on Epistemic Internalism”. In Warrant in Contemporary Epistemology: Essays in Honor of Plantinga’s Theory of Knowledge, edited by J. Kvanvig, 73-83. Lanham: Rowman & Littlefield Publishers, Inc.

______. 1997. “Mythology of the Given”. History of Philosophy Quarterly 14 (3): 275-86.

______. 1997. “Reflective Knowledge in the Best Circles”. Journal of Philosophy 94 (8): 410-30.

263

______. 1999. “Skepticism and the Internal/External Divide”. In The Blackwell Guide to Epistemology, edited by J. Greco and E. Sosa, 145-57. Oxford: Blackwell.

______. 2003. “Privileged Access”. In Consciousness: New Philosophical Perspectives, edited by Quentin Smith and Aleksandar Jokic, 273-92. Oxford: Clarendon Press.

Steup, M. 1989. “BonJour’s Anti-Foundationalist Argument”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 188-99. Dordrecht: Kluwer Academic Publishers.

______. 1989. “The Regress of Metajustification”. Philosophical Studies 55: 41-56.1989.

______. 1996. An Introduction to Contemporary Epistemology. Upper Saddle River: Prentice Hall.

______. 1997. “A Defense of Internalism”. In The Theory of Knowledge: Classical and Contemporary Readings, edited by L. Pojman, 310-21. Belmont: Wadsworth.

Swain, M. 1981. Reasons and Knowledge. Ithaca: Cornell University Press.

______. 1989. “BonJour’s Coherence Theory of Justification”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 115-24. Dordrecht: Kluwer Academic Publishers.

Thagard, P. 1978. “The Best Explanation: Criteria for Theory Choice”. Journal of Philosophy 75: 76-92.

______. 2000. Coherence in Thought and Action. Cambridge: MIT Press.

Tolliver, J. 1989. “The St. Elizabethan World”. In The Current State of the Coherence Theory: Critical Essays on the Epistemic Theories of Keith Lehrer and Laurence BonJour, with Replies, edited by J. Bender, 160-7. Dordrecht: Kluwer Academic Publishers.

Van Cleve, J. 1979. Foundationalism, Epistemic Principles, and the Cartesian Circle. Philosophical Review 88: 55-91.

______. 1984. “Reliability, Justification, and the Problem of Induction”. Midwest Studies in Philosophy 9: 555-67.

______. 1985. “Epistemic Supervenience and the Circle of Belief”. Monist 62: 90-104.

______. 1999. “Epistemic Supervenience Revisited”. Philosophy and Phenomenological Research 59 (4): 1049-55.

van Fraassen, B. 1980. The Scientific Image. Oxford: Clarendon Press.

______. 1989. Laws and Symmetry. Oxford: Clarendon Press.

264

Vinci, T. 1982. “Gettier Examples, Probability and Inference to the Best Explanation”. Philosophia 12: 57-75.

Vogel, J. 1992. “Sklar on Methodological Conservatism”. Philosophy and Phenomenological Research LII (1): 125-31.

______. 2000. “Reliabilism Leveled”. Journal of Philosophy 97 (11): 602-23.

Williams, M. 1977. Groundless Belief: An Essay on the Possibility of Epistemology. Princeton: Princeton University Press.

______. 1978. “Inference, Justification, and the Analysis of Knowledge”. Journal of Philosophy 75: 249-63.

______. 1996. Unnatural Doubts. Princeton: Princeton University Press.

265