Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2004 A Partial Defense of Compatibilism Jason Turner

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected]

THE FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

A PARTIAL DEFENSE OF COMPATIBILISM

By

Jason Turner

A Thesis submitted to the Department of Philosophy in partial fulfillment of the requirements for the degree of Master of Arts

Degree Awarded: Summer Semester, 2004

The members of the Committee approve the Thesis of Jason Turner defended on 25 June 2004.

______Alfred R. Mele Professor Directing Thesis

______Eddy Nahmias Committee Member

______Thomas M. Crisp Committee Member

The Office of Graduate Studies has verified and approved the above named committee members.

ii

ACKNOWLEDGEMENTS

I would like to thank Zac Ernst, Peter Hanowell, Matt James, Jeremy Kirby, Kirk Ludwig, Thomas Nadelhoffer, and Christopher Pynes for helpful suggestions, comments, and criticisms regarding the papers and ideas that led up to this thesis. I am especially grateful to Tom Crisp, Al Mele, and Eddy Nahmias for their extensive and helpful comments both on this thesis itself and on its ancestral papers. I would also like to thank Joseph Keim Campbell for first introducing me to the metaphysics of free will, my wife Starr for putting up with me during the writing process, and my father Ted for his sound advice that this thesis aims to follow.

iii

TABLE OF CONTENTS

List of Figures ...... v Abstract ...... vi

1. INTRODUCTION ...... 1

2. HOW NOT TO BE AN ELIMINATIVIST ABOUT FREE WILL...... 7

2.1. Eliminativism and the Argument For It...... 7 2.1.1. Types of Eliminativism 2.1.2. Intuitional Anarchy 2.1.3. “Free Choice” and Natural Kind Terms 2.1.4. The Argument 2.2. Meaning as Use Plus Eligibility...... 19 2.2.1. Putnam’s Model-Theoretic Paradox 2.2.2. Use Plus Eligibility and the Bottom-Up Argument 2.2.3. An Eliminativist Objection 2.3. Indeterminate Meaning and Contextualism...... 27 2.4. Further Implications...... 33

3. THE INCOMPATIBILITY OF FREE WILL AND NATURALISM...... 40

3.1. The Consequence Argument...... 40 3.2. Naturalism and the Supervenience Argument ...... 43 3.2.1. Choosy Actions 3.2.2. Causal Relations 3.2.3. The Supervenience Argument: A First Pass 3.2.4. The Argument for Trickier Cases 3.3. Implications ...... 54 3.4. Objections and Replies ...... 58

4. PUTTING IT ALL TOGETHER: IN DEFENSE OF COMPATIBILISM ...... 70

4.1. Limitations of an Offensive Defense ...... 70 4.2. Introducing the Use-Plus-Eligibility Argument for Compatibilism ...... 73 4.3. Agent-causation, (β□), and the Use-Plus-Eligibility Argument ...... 75 4.4. Conclusion ...... 82

iv

REFERENCES ...... 86

BIOGRAPHICAL SKETCH ...... 92

v

LIST OF FIGURES

Figure 1: The causal structure of r’s supervenience base ...... 49

Figure 2: “In-the-middle” indeterminism ...... 49

Figure 3: “At-the-end” indeterminism ...... 52

vi

ABSTRACT

Compatibilism is the view that free will can exist even if determinism — the thesis that there is only one physically possible future at any given time — is true. In this thesis, I defend compatibilism by arguing against two of its main rivals. I first argue against necessary eliminativism — the view that free will is impossible — by deploying an attractive view of language (Lewis, 1983, 1984; Sider, 2001) to show that, so long as ordinary folk are liable to experience conflicting intuitions about how to use the term ‘free,’ it will refer to some property which is possibly exemplified. I then argue that libertarians — believers in free will who hold that it is incompatible with determinism — must reject either a naturalistic view of the world with no ontological commitments above and beyond those proscribed by science or the soundness of the best argument in favor of libertarianism (van Inwagen, 1983, ch. 3). Finally, I sketch a way a proponent of compatibilism can use my arguments against necessary eliminativists and naturalistic libertarians to offer a positive argument for compatibilism.

vii

CHAPTER 1 INTRODUCTION

According to my father, the best defense is a good offense. This thesis aims to follow the implicit advice in that adage, defending compatibilism — the view that agents may have free will even if determinism is true — by going on the offensive and attacking some of compatibilism’s main opponents. By determinism, I mean the thesis that there is only one physically possible future given the state of the universe at any given moment. We can, following the lead of Peter van Inwagen (1983, pp. 58-65), make this a little more precise. Say that a proposition φ “expresses a state of affairs at a time, t,” if and only if φ asserts that some state of affairs obtains at t. If φ expresses a state of affairs at t, then call φ “immediate” if and only if φ does not entail any proposition ψ that expresses a state of affairs at t* (where t* ≠ t) and call φ “complete” if and only if, if φ is true, then for every true immediate ψ which expresses a state of affairs at t, φ entails ψ. Determinism is then the thesis that, if L is a proposition that expresses the laws of nature, then any true, immediate, and complete proposition φ that expresses a state of affairs at some time, conjoined with L, entails any true proposition ψ whatsoever — i.e., that □((φ & L) → ψ).1 By free will I mean the kind of freedom that is required in order for agents to count as morally responsible. There are a number of uses of the word “free” in English, and not all of them are related to philosophical issues involving free will. Political and economic freedom, for instance, are desirable things, but they are not the grand prizes philosophers are angling for when they go looking for free will. Rather, “free will” becomes, roughly, a tag for “the kind of freedom that grounds moral responsibility” — i.e., a kind of freedom necessary for moral responsibility not entailed by any other kind of freedom necessary for moral responsibility.

1 There may be other reasons that freedom of this kind is thought desirable. For instance, it is sometimes thought that the very same property that is a necessary condition for being morally responsible is also a necessary condition for being an appropriate “target” of reactive attitudes — attitudes like resentment, gratitude, and love (P. Strawson, 1962) — or for being truly creative, autonomous, unique, and having a certain sort of dignity (Kane, 1996, pp. 81-89). So although we may use moral responsibility to fix the sort of property that free will is supposed to be, there may be a number of other quite valuable things that free will is supposed to gain for us. Compatibilism thus becomes the view that free will, in this sense, is possible even if determinism, as defined above, turns out to be true. Incompatibilism, or the view that determinism precludes free will, is the denial of compatibilism. Incompatibilists divide further into two camps: those who believe that people have free will, and those who do not. Incompatibilist believers in free will are called libertarians. Event-causal libertarians (Ekstrom, 2000; Kane, 1989, 1996, 1999) hold that suitably located indeterministic processes of the sort posited by quantum physics are all that is needed to escape the threat to free will that determinism provides. Agent-causalists (Chisholm, 1964; Clarke, 2003; O’Connor, 2000; Taylor, 1963), however, contend that suitably located indeterminism is not enough. The indeterministic processes postulated by quantum physics are all instances of ordinary (albeit indeterministic) event causation — instances of one event causing another. Agent-causalists insist that free will requires a different sort of causation entirely — agent-causation, in which a substance (an agent), rather than an event, is the cause of an action. It is generally held that ordinary (non- agent-causal) claims that appear to ascribe causal efficacy to an agent (like “John caused the roof to fall”) can be analyzed into causal claims that only involve events (like “John’s pushing on that wall caused the roof to fall”). Agent-causalists hold that genuine agent- causal transactions cannot be analyzed into purely event-causal claims — and that individuals are free only if they agent-cause (at least some) of their actions. Finally, non- causal libertarians (Ginet, 1990, 1997; Goetz 1988) hold that free acts may not be caused at all.

2 Incompatibilists who do not believe in free will have been sometimes called skeptics (Ekstrom, 2000, pp. 3-4) or pessimists (P. Strawson, 1962; G. Strawson, n.d.), among other things. I propose to lump views that deny the existence of free will together under the term eliminativism. An eliminativist may think that the non-existence of free will is a contingent fact, in which case she is a contingent eliminativist, or she may think that it is a necessary truth, in which case she is a necessary eliminativist.2 A further taxonomy of eliminativist positions is provided in §2.1.1. The main arguments of this thesis are targeted primarily at event-causal libertarians and at eliminativists. The view that free will is possible but non-actual is no longer a very popular position. This location in logical space has traditionally been populated by hard determinists, philosophers who think both that determinism is true and that it is incompatible with free will (but see Pereboom, 2001 for a recent defense of a different sort of contingent eliminativism). With the popular acceptance of quantum- mechanical interpretations of subatomic physics, however, determinism has lost favor with the philosophical community in general. Most philosophers tend to think that it stands a strong chance of being false, given our best physics. There thus seems to be little motivation for being a hard determinist. Some readers may wonder what reason anyone could have for being a compatibilist if determinism is most likely false. If there is no threat to our free will from determinism, what reason is there to avoid using indeterministic elements of the world postulated by physics to understand human freedom? There are two responses to this question. The first is that it is not clear whether or not the indeterminism postulated by quantum physics is sufficient to provide the kind of free will incompatibilists think determinism rules out. It may be that the metaphysical commitments that keep (by the incompatibilist’s lights) free will from being compatible with determinism also keep it from being compatible with a physical world where all of the indeterminism percolates up from the microphysical level. Perhaps incompatibilists think that free will is incompatible with determinism because they want a very robust kind of indeterminism — more robust than that provided by quantum physics — to ground free will. In Chapter 3, I argue that at least some reasons for being an

3 incompatibilist are also reasons to think that free will requires more than mere microphysical, event-causal indeterminacy. In this case, compatibilism is interesting not because it can secure our freedom against a (probably false) thesis of determinism but because it can secure our freedom against a (more likely true) thesis that the robust kind of freedom sought by libertarians does not obtain in our world. A second response is that compatibilism is interesting not because it secures human freedom in this world but because of the verdict it issues with respect to our freedom in other, deterministic worlds. Compatibilist believers in quantum indeterminacy may feel that the fortunes of free will should not be held hostage to the truth or falsity of determinism (cf. Fischer, 1994, pp. 6-8). Although compatibilism is not (by their lights) required to secure freedom in our world, it does reflect an interesting and important fact about free will: being free does not require ever having had the ability to do something other than what was done in such a way that doing otherwise would have been physically compossible with some immediate and complete true proposition about the past. If true, this tells us something interesting about the very nature of free will. It tells us that free will is not essentially tied up with a certain sort of ability to actualize futures compossible with the past and laws of nature. As the title of this thesis indicates, the defense found here is only partial. There is a sense in which any philosophical work is partial, since all argumentation must choose an Archimedean point from which to begin. However, in the particular, offensive sense of “defense” used here, it is very appropriate to call this thesis a “partial” defense, for only some of compatiblism’s main rivals are examined. Chapters 2 and 3 each focus on a different specific argument favoring one of compatibilism’s opponents. Chapter 2 focuses on the Bottom-Up Argument (Double, 1991, ch. 5, 2001, p. 115) for a certain sort of eliminativist position. The Bottom-Up Argument, in essence, contends that widely divergent intuitions within individuals about how to use the term “free” preclude that term’s having any sort of meaning that could affect the truth-values of sentences in which it is embedded. I argue in Chapter 2 that this divergence of intuitions need not lead to semantic eliminativism: we can avoid it if we subscribe to a view of language on which the meanings of terms are determined not just

4 by the way we use them but also by metaphysical facts about the world (Lewis, 1984; Sider, 2001a). As an added bonus, on the semantics of “free” produced by the combination of divergent intuitions and this “use-plus-eligibility” view of language, other sorts of necessary eliminativism are false as well. In Chapter 3 I discuss the Consequence Argument (van Inwagen, 1983, pp. 93- 95), perhaps the most prominent argument for incompatibilism to be found in the literature. I do not argue that the Consequence Argument is invalid, but rather that, if it is valid, so is another argument that I call the Supervenience Argument. The Supervenience Argument shows that free will is incompatible with a certain widely accepted naturalistic picture of the world. There are two upshots of the Supervenience Argument and its relationship to the Consequence Argument. First, libertarian theories of free will (such as those of Ekstrom, 2000; Kane, 1989, 1996, 1999) committed both to the Consequence Argument and to naturalism are discredited. Second, insofar as naturalism is a plausible view, incompatibilism becomes difficult to defend, since any such defense must include a defense of the denial of naturalism. Chapter 4 begins by taking stock. In the first section I consider how successful the arguments in Chapters 2 and 3 are as arguments for compatibilism. Finding them wanting, I then suggest a positive argument for compatibilism: the Use-Plus-Eligibility Argument, which draws upon the semantic resources discussed in Chapter 2 and the implications for naturalistic libertarianism drawn out in Chapter 3. However, some of the premises of the Use-Plus-Eligibility Argument require a defense beyond that which I am able to provide within the confines of this thesis. Thus, the argument in Chapter 4 provides another way in which this thesis is only a partial defense of compatibilism: the positive argument marshaled in favor of compatiblism is only partially defended, and further work is required to make it fully compelling. The incompleteness of my defense of compatibilism may put off some of my readers. I readily concede that producing a full defense of compatibilism — considering all of its opponents and leaving no premise undefended — is far more desirable than producing a merely partial defense. It is good to remember, though, that philosophy may proceed incrementally and as a joint effort. Although my defense is merely partial, it may

5 form a crucial plank in a larger and more complete case for compatibilism — which case may be constructed, perhaps, by a multitude of authors. With that hopeful thought for the establishment of grand, sweeping conclusions, I turn to the more humble task I have set for myself.

Notes to Chapter 1

1 This more formal definition is slightly more restrictive, for it entails that there is only one past physically compossible with any given state of the universe as well. This quirk of the definition makes no difference for the arguments found in this thesis.

2 I have suggested that eliminativism is a form of incompatibilism, but one may consistently believe that free will is compatible with determinism but, in actuality, nobody ever acts freely. This, I take it, would be a form of contingent eliminativism. I am not aware of anybody who holds this “hard compatibilist” position, and for the purposes of this thesis it will be largely ignored. Necessary eliminativists, however, are incompatibilists: if it is impossible for people to act freely, free will is incompatible with everything and, a fortiori, incompatible with determinism.

6

CHAPTER 2 HOW NOT TO BE AN ELIMINATIVIST ABOUT FREE WILL

Eliminativism about free will is the view that no agent ever acts freely, and necessary eliminativism is the view that it is impossible that any agent ever act freely. One sort of necessary eliminativist holds that the impossibility of free will stems from an inbuilt conceptual confusion, reflected in the way free-will intuitions diverge and conflict. In this chapter I examine the best argument from this intuitional anarchy to this sort of eliminativism in light of an attractive view of language developed by David Lewis (1984, 1983) and defended by Theodore Sider (2001a, 2001b). Given this picture of language, the argument from intuitional anarchy fails and an alternative to eliminativism can be found. Furthermore, intuitional anarchy plus the view of language in question also spell trouble for those eliminativists who believe that free will is impossible due to the way the world is rather than because of any conceptual confusion.

2.1. Eliminativism and the Argument For It 2.1.1. Types of Eliminativism Eliminativists about free will come in all sorts of varieties. Contingent eliminativists think that, as a matter of contingent fact, people don’t have free will. Hard determinists — incompatibilists who believe in the truth of determinism — are one species of contingent eliminativist, insofar as determinism is taken to be a contingent thesis. The philosopher Derk Pereboom (2001, 2002) is another contingent eliminativist: he holds that agents act freely only if they are agent-causes of their acts and that such agent causation is possible, but almost certainly not actual. In this chapter I am primarily concerned with necessary eliminativists, who hold that the non-existence of free will is a necessary truth. These philosophers think, in short,

7 that nothing can possibly do the sort of work free will is supposed to do. Necessary eliminativists fall into two further camps. Metaphysical eliminativists (Smilansky, 2000, 2001; G. Strawson n.d., 2001) believe that there is no free will due to the way the world is. Semantic eliminativists (Double, 1991, 1996, 2001) believe that there is no free will due to the way our words work. Sentences express propositions, which are the bearers of truth and falsity. Call a proposition the semantic value, or content, of a sentence. The semantic value of a given term is, roughly, the thing a term refers to that affects what proposition the sentence it is embedded in expresses (and thus what truth-value the sentence is associated with). For instance, the semantic value of the term “the Eiffel Tower” may be a particular object located in Paris, the semantic value of “is 300 meters high” may be a certain property of hitting that given height, and, as a result, the proposition expressed by “The Eiffel Tower is 300 meters high” will be true if and only if a particular object in Paris possesses a certain height property. According to semantic eliminativists, free-will terms have no content, or semantic value. In other words, they make no semantic, or truth-value affecting, contribution to sentences in which they are embedded. The truth-values of sentences are never changed by the addition of the modifier “free.” This is a very different position from that of the metaphysical eliminativist, who holds that “free” does have a semantic value, albeit one that is impossible to satisfy. The difference between the metaphysical and semantics eliminativist can also be couched in terms of conceptual analyses: the former thinks “S freely chose to a” can be given necessary and sufficient conditions, but that those conditions are unsatisfiable, while the latter thinks no such conditions can be given. The difference can also be put in terms of perspective: the metaphysical eliminativist thinks the problem with free will lies in the world, whereas the semantic eliminativist thinks it is in our words. Semantic eliminativists may have different views about the non-semantic (i.e., non-truth-value relevant) linguistic function of free-will talk. A free-will subjectivist, like Richard Double (1991, 1996, 2001), contends that the linguistic contribution of free-will terms is analogous to the sort of linguistic contribution non-cognitivists about morality

8 think that terms like “right” and “wrong” make (Double, 2001, pp. 507-508). Other positions are available. One could be an austere eliminativist, for instance, by insisting that there is no useful linguistic role for free-will talk whatsoever — not even a non- cognitivist one — and that we should thus jettison it in favor of some more accurate way of speaking.1 On either view, terms like “free” make no semantic contribution to sentences in which they are embedded, but on one view they have some non-semantic (i.e., non-cognitive) value, while on the other they do not. According to Double (2001, p. 511), there are two routes to semantic eliminativism: the top-down and the bottom-up. The Top-Down Argument, in rough form, argues that free-will terms are relevantly similar to ethical terms and that non-cognitivism is the correct view of ethical terms (Double, 1996, ch. 8-9, 2001, p. 511). Even setting aside a detailed assessment of the Top-Down Argument, we can see that it suffers from two deficiencies. First, its fate is inextricably tied to non-cognitivism. Second, it rules out austere eliminativism. Since the Bottom-Up Argument is consistent with both moral realism and austere eliminativism, I take it to be the more general, and thus the stronger, argument. I propose accordingly to set aside the Top-Down Argument and focus exclusively on the Bottom-Up. There are two main ideas lurking behind the Bottom-Up Argument. The first is that our intuitions about free will conflict with each other. We can find indefinitely many cases in which we each feel the intuitive pull both of calling an agent free and of calling her unfree. The second is that free-will terms are the sorts of words that get whatever meaning they have primarily from our intuitions. These two considerations suggest that free-will terms “do not denote any well-behaved, principled, philosophical ‘natural kinds’ that make discussions of freedom and responsibility truth-valued in even [a] loose sense” (Double, 1991, p. 9). Free-will terms have no semantic value because we do not use them in a way conducive of their having one. 2.1.2. Intuitional Anarchy Call a particular reflective but pre-theoretical judgment about a case an intuitive judgment (cf. Double, 1991, pp. 18-19). Intuitions, as I understand them here, are inclinations to issue a particular judgment about a case such that, if a judgment were

9 made on the basis of that inclination, it would count as intuitive. One may thus have an intuition that φ without judging that φ, so long as one feels an inclination to judge that φ and, if one were to judge that φ on the basis of that inclination, the judgment that φ would be an intuitive judgment. Admittedly, the above account is rough, but it should be good enough for our purposes. The thesis of intuitional anarchy states that we have conflicting intuitions about free will — that is, there are particular cases in which we (or, better yet, pre- theoretical and reflective folk) each feel inclined both to judge the agent in question as free and to judge her as unfree. Furthermore, the conflict is both deep and widespread. This is not a problem involving a few borderline cases where we cannot decide what we want to say. It is, rather, a problem where we feel decisively both that an agent is free and that she is not free — depending, perhaps, on how we view the situation. It is a problem of not being able to decide what it is for a choice or action to count as free. Intuititional anarchy is ultimately an empirical thesis. Philosophers can offer examples and consult their own intuitions (we will do a bit of that), but the only real way to establish whether or not the majority of reflective, pre-theoretical folk respond in the way the thesis of intuitional anarchy predicts is to do some field research on reflective, pre-theoretical folk. There are, however, some preliminary considerations that offer support for the thesis. Begin by considering whether or not the agent chooses freely in the following scenario:

Sally is a teller at a local bank. One afternoon, a robber walks into the bank and pulls a knife on her. He brandishes the knife and says, “Open the vault or you’ll get it.” Sally knows that she is not positioned in a way to be able to alert the police without the robber knowing. She also knows that, if she opens the vault and then alerts the police, the robber will have gotten away before they arrive at the bank. Finally, she knows that, if the robber stabs her, she will probably not die but will likely be severely injured and in a lot of pain. With all this in mind, Sally reluctantly opens the bank vault and ushers the robber inside.

10

Did Sally freely choose to open the vault to the robber? On the one hand, this is a paradigm case of coercion, which in turn is a paradigm of not having free will. When we think about Sally as coerced, we are inclined to judge her choice as unfree. On the other hand, either option — opening the vault or letting the robber stab her — seems to be entirely within Sally’s power. It is not contradictory to think that Sally was able to let the robber stab her, even though she did not. Opening the vault was her choice, and refusing to open the vault appears to be something she could have done. When we think of Sally’s choice in this way, we are inclined to judge her choice as free. For readers unpersuaded about this last point, imagine that, instead of a bank teller, Sally was a nuclear technician, and instead of being threatened by a bank robber, she was threatened by a terrorist who wanted her to cause a nuclear meltdown in a power plant in a highly populated megatropolis many miles away. (Sally would not have been personally physically harmed by the meltdown.) In this case, as opposed to the bank robber case, we would expect people to feel that Sally has an obligation to “take one for the team” and let the terrorist stab her in the interest of saving the lives of millions. We can envision her choosing either option, and whichever way she goes, we tend to think that she chose that option freely. Since the cases are structurally similar, the latter case seems to boost our conviction that Sally could have chosen to be stabbed in the former. It is not difficult to pinpoint the source of the conflict. On the one hand, Sally’s choice seemed entirely in her control. Whether she opened the vault or allowed herself to be stabbed was entirely “up to her,” so to speak — had she decided not to open the vault, there was (we may suppose) nothing the robber could have done to force her to change her mind, however forcibly he might try to persuade her. When we consider Sally’s choice in this light, we are inclined to say that she chose freely. On the other hand, since she was, in essence, manipulated into opening the bank vault, it seems odd to say that she chose freely. It is not as though her action sprang in any way from her own projects and desires. The robber’s actions instead placed Sally in a state of psychological tension: she had to choose between desires she likely takes as constitutive of her personality — desires to not be harmed and to not give money to

11 people to whom it doesn’t belong — and which are usually jointly satisfiable. When we consider her choice in this light, we are inclined to say that she chose unfreely. It looks as though our free-will intuitions are tracking different properties choices might exemplify. In making her choice, Sally seems to have what we might call dual control — roughly, the ability to do either of two (or more) things such that, if she did any of them, they would have been in her control. Our inclination to judge her as free comes upon reflecting on her possession of this property. On the other hand, she was lacking what we might call psychological integration, or harmony in the psychological states that produced her action. When we consider this, we feel inclined to judge her as unfree. One explanation of this phenomenon has to do with how concepts are deployed. There is a considerable amount of empirical data supporting the idea that, when faced with a question like “Did Sally choose freely?” we respond based on how similar Sally’s situation is to some paradigm case or cases (see Ramsey, 1998, pp. 168-170 for a brief but clear discussion). Call terms that function like this paradigm-relative. Consider an example. Suppose, for the sake of illustration, that the paradigm of an intentional action involves an agent consciously and successfully executing a previously settled-on plan in such a way that the success of the execution was all but guaranteed. If “intentional action” is a paradigm-relative term, then whether or not we judge given cases of action intentional will depend on how similar they are to this paradigm. Cases very similar will generate confident judgments of “intentional,” and cases very dissimilar will generate confident judgments of “non-intentional.” Cases somewhere in between will generate unconfident judgments and, perhaps, conflicting judgments in a sample population. The proffered explanation of free-will anarchy is that “free” is paradigm-relative,2 but that it is relative to multiple paradigms. Double (1991, ch. 5) offers an explanation of intuitional anarchy along these lines and suggests that among our free-will paradigms are the Pure Rational Ego — an agent who always acts with complete rationality, unswayed by capricious considerations, for her own maximal utility — and the Non-egocentric Actor — an agent “‘as free as the wind,’ and as unpredictable” (p. 118), broadly rational but capable of doing almost anything at a moment’s notice.

12 It is easy to see how, if “free” is relative to these paradigms, judgments — even pre-theoretical ones — involving it would be sensitive to considerations of dual control and psychological integration. Insofar as we recognize that Sally, should the mood have taken her, would have been able to let the robber shoot her, we see a comparison between her and the Non-egocentric Actor and feel confidence in calling her choice free. On the other hand, the psychological tension engendered in Sally by the situation is incompatible with the tranquility that must befall a Pure Rational Ego, and insofar as we see this, we feel confident in judging her choice unfree. Or so I claim, as does the semantic eliminativist. As indicated above, everyone must examine his or her own intuitions, and there is a limit to how successful philosophical debate can be on this issue. The preliminary considerations are, to my mind, quite weighty, but the question is ultimately empirical. It is a testable matter whether judgments diverge in Sally’s (and similar) cases, it is a testable matter whether “free” is paradigm-relative, and if this claim turns out true, it will be a testable matter whether it is relative to multiple paradigms. There is some empirical evidence available, independent of paradigm relativity, which also suggests that we should expect to see intuitions come apart in cases where dual control and psychological integration come apart. Shaun Nichols (n.d., §5) has suggested that at a young age children begin to develop an obligation system — a cognitive system used to regulate one’s own behavior — to go along with their mindreading system (cf. Nichols and Stitch, 2003) for predicting and explaining the behavior of others (and, perhaps, for giving post hoc explanations for their own actions). Nichols’ data suggest that the obligation system is associated with judgments about agents’ abilities to do otherwise.3 The mindreading system’s role in predicting and explaining behavior, on the other hand, makes it particularly sensitive to psychological considerations. The mindreading system may incline us to make judgments regarding freedom in ways pertinent to psychological integration, whereas the obligation system may produce inclinations which better track dual control. Note that something like dual control has traditionally been in the purview of incompatibilist theories of free will (e.g., Clarke, 1995, pp. 133-134; Kane, 1996, pp.

13 109-111, 133-144, 1999; O’Connor, 2000, pp. 20-22, 81-84), whereas something like psychological integration is often found at the core of compatibilist theories (e.g., Frankfurt, 1971; Bratman 1996. See Kane, 1989, pp. 140-141 for discussion of the role these properties play in compatibilist and incompatibilist theories). There are other properties that philosophers have been concerned about with respect to free actions, too. Incompatibilists Robert Kane (1996, esp. ch. 5) and Galen Strawson (n.d., 2001, esp. pp. 451-453), for instance, each discuss a certain sort of ultimacy that they hold is at the core of free will. On the compatibilist side, John Martin Fischer and Mark Ravizza (1998) highlight responsiveness to reasons as an important component of free will, and both they and Alfred Mele (1995, ch. 9) have argued that compatibilists should think choices must meet certain historical standards in order to count as free. It is an interesting — and open — question as to how many independent properties our free-will intuitions track and how well they line up with philosophical positions on the matter. The existence of these positions gives us at least prima facie reason to think at least some of them are highlighting properties (or features of properties) tracked by at least some free will intuitions. If this were the case, it would provide a natural and unified explanation for Double’s (1996, pp. 103-105, 2001, §2) observation that he feels inclined to assent to each of compatibilism’s and incompatibilism’s positive and negative claims when considered in different lights. For the balance of this thesis, I will content myself with the assumption that the thesis of intuitional anarchy is true and that dual control and psychological integration are the two properties that our free-will intuitions track. The former assumption is necessitated by my own inability to conduct the required empirical research. The latter assumption is mainly for convenience: since I do not have anything more than suggestive data indicating the nature of the properties our “free” intuitions are sensitive to, I will let psychological integration and dual control stand as proxies for whatever (and however many) properties really are at the root of intuitional anarchy. Even without the empirical data, however, it is reasonable to suppose that whatever properties really are responsible for divergent intuitions fall roughly into compatibilist- and incompatibilist-friendly categories. Both the preliminary research cited above and the prevalence of conflicting

14 post-theoretic intuitions in the philosophical debate give ample support to this supposition. 2.1.3. “Free Choice” and Natural Kind Terms The first idea behind the Bottom-Up Argument — that our intuitions about free will conflict — is embodied by the thesis of intuitional anarchy. The second idea, in rough form, is that if free-will terms have any semantic value, it is more-or-less whatever we (i.e., pre-theoretical and reflective folk) think it is. “Free” just means whatever we think it means.4 Intuitional anarchy, though, suggests that we just aren’t sure what “free” means — sometimes we think it means dual control, and other times we think it means psychological integration. There is a wrinkle that needs to be ironed out. On a widely held view of semantics developed and defended by Saul Kripke (1972) and Hilary Putnam (1975), some terms don’t mean what we think they mean. When a natural kind (such as water) has some underlying microphysical composition (such as H2O), then the meaning of the term that names that kind (such as “water”) essentially involves that microphysical composition. More precisely, if K is a natural kind with microphysical composition C, and “T” denotes K, then it is part of “T”s meaning that, necessarily, a substance falls under “T” only if it has microphysical composition C — independent of what we think “T” means or what we think K’s microphysical composition is.5 On Kripke/Putnam semantics, then, a substance falls under the term “water” only if it is composed of H2O. This fact is independent of what we think “water” means.

“Water” was H2O long before anyone had any idea that there were things called molecules or hydrogen atoms. If “free choice”6 is a natural kind term, then its meaning will be independent of what we think it means and the intuitional anarchy may simply reflect a lack of understanding of the real meaning of “free” — the one tied up with the microstructure that all free choices share. Natural kind terms are created (in the simplist and most idyllic cases, at least) when someone points to or describes some individual items or samples and says “let ‘T’ designate this kind of thing.” When most or all of the items or samples share a microphysical composition C, it becomes part of the semantics of “T” that something is T

15 only if it has microphysical composition C (Kripke, 1972, pp. 135-137). This is called the initial baptism of a term. Mark Heller (1996) has suggested that “free choice” is a natural kind term. Thus, at some point (in the simplest and most idyllic case) “free choice” was baptized. Someone described or pointed at some choices and said “Let ‘free choice,’ mean choices like these.” If Heller is right and “free choice” is a natural kind term, then a choice will be free only if it has a microphysical substructure shared by the original set of choices used to introduce the term. If “free choice” is a natural kind term, then the central idea behind the Bottom-Up Argument — that “free choice” means whatever we think it means — is mistaken. “Free choice” means (at least in part) choices with the same microstructure as the choices in the original set used for the initial baptism. But even if “free choice” is a natural kind term, there is hope for the semantic eliminativist: if there is no microphysical substructure shared by the choices in the initial baptism class, then, necessarily (Kripke, 1973, pp. 157-158) nothing falls under the term “free choice” (Heller, 1996, p. 336). This line of argument may be plausible if we think that the divergence of free-choice intuitions suggests that, of the original choices used to introduce the term “free,” some exhibited only dual control and some exhibited only psychological integration. If it can be shown that there is no single microphysical substructure that underlies psychologically integrated choices and choices that exhibit dual control (at least without underlying further choices that do not exemplify any property that our free-will intuitions track), then the eliminativist can argue that there was no substructure unity in the original reference- fixing set and therefore, necessarily, nothing falls under “free choice.”7 The semantic eliminativist need not make this move, however, since there is good reason to think that “free choice” is not a natural kind term at all. If it turned out that every time someone had ever applied the term “free choice” to a decision, that decision was, unbeknownst to the agent, caused by aliens via a device implanted in the agent’s brain, then “free choice” would mean, in part, “caused by such-and-such a device implanted in the brain by an alien.” It seems, though, that in such an eventuality we would conclude that no one had ever chosen freely — not that free choices are much

16 different than we thought they were (Daw & Alter, 2001; van Inwagen, 1983, 106-115). Since whether or not a term is a natural kind one is generally thought to depend on our intuitions about how it functions in these sorts of cases, we have reason to think that “free choice” is not a natural kind term. 2.1.4. The Argument We are now ready for the Bottom-Up Argument. This argument is drawn loosely from Double, 1991 (pp. 130-131); I present it as a reductio to the claim that “free” has a semantic value. For it, and throughout this chapter, I take semantic values to be properties, where it is left open whether properties are universals, similarity classes of tropes, functions from worlds to individuals, or what-have-you. (It is assumed, however, that properties, being the semantic values of terms, will be amenable to truth-functional connection: if there is a property of being P and a property of being Q, then there is a property of being P or being Q.8) The conclusion of the argument is thus equivalent to the claim that no property serves as the semantic value of “free.”

The Bottom-Up Argument (1) If “free” has a semantic value, this value is determined by the way we use the term. (2) The way we use “free” is determined by our intuitions. (3) Thus, if “free” has a semantic value, it is determined by our intuitions (4) So, if “free” has a semantic value, it will preserve most of our intuitions. (5) We can generate an indefinite number of cases where our intuitions pull in opposite directions (i.e., intuitional anarchy is true). (6) Therefore, any property that could serve as the semantic value of “free” will either be too broad, too narrow, or both. (7) So, no property preserves most of our intuitions. (8) Hence, “free” has no semantic value.9

The defense of (1) is that, since “free” is not a natural kind term, its semantic value does not depend on any underlying microphysical structure shared by any set of

17 items (choices or acts or agents) used for an initial baptism. And, says the semantic eliminativist, there seems to be nothing else that could affect its semantic value, so use must do the whole thing. I will grant the semantic eliminativist (2) for the sake of argument; insofar as this is false, it appears to be little help to a foe of semantic eliminativism (see §4.2).10 These two entail (3), the claim that our intuitions determine the semantic value of “free.” The crucial move is from (5) to (6). Here is how it works. If the semantic value of “free” is the disjunctive property of exhibiting either dual control or psychological integration, it will go against intuitions that certain choices that exhibit only one are unfree and therefore be too broad. If the semantic value is the conjunctive property of exhibiting both dual control and psychological integration, it will violate intuitions that certain choices exhibiting only one of these are free and be too narrow. And if it is some intermediate property (i.e., a property exemplified by all things, actual or possible, that exemplify both dual control and psychological integration, and exemplified by all things, actual or possible, that exemplify either dual control or psychological integration), it will not be possessed by some choices that we (in some moods) think are free and will be possessed by some choices that we (in some moods) think are unfree, and therefore be both too narrow and too broad. Thus, no property we identify can do justice to our free-will intuitions — and semantic eliminativists think that no property can even come close due to the deep and widespread nature of intuitional anarchy. If some properties were just slightly too broad or slightly too narrow, the move from (6) to (7) could be blocked. However, since there are an indefinite number of cases in which our intuitions pull in opposite directions, any property chosen will do injustice to an indefinite number of cases favored by opposing intuitions, so it cannot be said even to preserve most of these intuitions. Thus, “free” has no semantic value. Since it is an argument for semantic eliminativism, the conclusion — that “free” has no semantic value — must entail the semantic eliminativist claim that “Our proclaiming choices to be free and persons to be morally responsible for their choices can be nothing more than our venting of non-truth-valued attitudes, none of which is ‘more

18 correct’ or ‘more rational’ that competing attitudes” (Double, 1996, p. 3). In §2.2 and §2.3, I argue that the Bottom-Up Argument fails to establish the truth of semantic eliminativism. In §2.2, I argue that premise (1) of the Bottom-Up Argument is false and, as a result, the truth of intuitional anarchy is consistent with the semantic value of “free” being indeterminate between a number of properties. In §2.3, I contend that the indeterminacy of the semantic value of “free” is consistent with the falsity of semantic eliminativism. Thus, semantic eliminativism does not follow from intuitional anarchy. In §2.4 I discuss some limitations of my argument and then suggest that the considerations adduced here bear upon metaphysical eliminativists as well.

2.2. Meaning as Use Plus Eligibility As already noted, the Bottom-Up Argument for semantic eliminativism relies on the claim that the semantic value of “free” is determined by our intuitions. If (say) “free choice” were a natural kind term, this would be false, but there is good reason to think that “free choice” is not a natural kind term. There is an attractive view of language, however, on which the meaning of “free” may not be entirely determined by the way we use it even if it is not a natural kind term. In this section, I present that view of language and then show that, if it is correct, the thesis of intuitional anarchy entails, at best, only that the semantic value of “free” is indeterminate between multiple properties. 2.2.1. Putnam’s Model-Theoretic Paradox Call the thesis that every term’s semantic value is determined entirely by the way we use it global use theory. The view of language assumed by the Bottom-Up Argument is either a global use theory (on which the function of natural kind terms is subsumed under use since we use natural kind terms in a way that supposes their meaning is parasitic on the substructure of their referents) or a partial use theory according to which the way we use terms and the linguistic conventions we follow are the sole determiners of non-natural-kind terms. Since we are assuming “free” (or any nominal term modified by it) is not a natural kind term, it will not matter which view of language the semantic eliminativist prefers.

19 In order to see how global use theory works, begin by considering the introduction of new terms into an already established language. Suppose scientists discover a new phenomenon, P, which they do not know much about but which they do know is closely related to another phenomenon, Q. They may hypothesize that a pair of previously undiscovered particles is responsible for both P and Q. Further evidence leads them to believe that these two particles annihilate each other upon contact, and that only one of them causes a further phenomenon, R. Scientists may introduce terms to name these (types of) particles — say, “zeppa” and “pezza,” — with the following “Z-P theory”:

Zeppa and pezza cause phenomenon Q when they are in such-and-such circumstances; Zeppa and pezza cause phenomenon P when they are in this-and-so circumstances; Zeppa causes phenomenon R when in such-and-so circumstances; and Zeppa and pezza annihilate each other whenever they come into contact.

Lewis (1970, 1984 pp. 222-224) argues that, if two objects (or types of objects) exist that satisfy “Z-P theory,” then these two (types of) objects are the values of “zeppa” and “pezza.” “Z-P theory” is used to introduce new semantically valuable terms into an existing language. Global use theory suggests that all of our terms get their semantic values in this way. Use provides a theory to be satisfied: a codification of our linguistic conventions and a comprehensive description of how we use terms makes an “ideal linguistic theory.” We can consider different “interpretations,” or assignments of objects and properties to terms, to see which ones make this theory true. The semantic values of our terms are the values they take on the interpretation that makes our ideal linguistic theory true. This is how use determines semantic content. This view comes up against difficulties that arise from Putnam’s (1980) model- theoretic argument against realism. The problem, in its essence, is that if we do not have

20 some already established language ready-to-hand, we run out of “semantic glue” (Lewis, 1984, p. 221) to stick our meanings to their referents. More precisely, provided ideal linguistic theory is consistent, there will not be just one interpretation which makes our ideal linguistic theory true, and there will be no way to decide among the competing interpretations as to which one we really mean.11 Some take this as a refutation of the idea that the values of any of our terms are language-independent objects (Putnam, 1980, 481-482), while others take it as a refutation of the model of language in question. Lewis (1983, p. 371, 1984, p. 226) claims that this shows we need some constraint, in addition to the way we use terms, to decide which of the many interpretations that satisfy ideal use theory is the one that gives our terms their values. Interpretations will satisfy a constraint C not by making some theory about C true, for in that case we will be able to simply add C-theory to our ideal linguistic theory and generate the problem all over again. Rather, since our linguistic behavior cannot do enough to determine the meaning of our terms, the world will have to add something — independent of us — to finish the job. Following Lewis, I take it that what makes one interpretation more eligible to be intended over its fellows is the extent to which it avoids highly disjunctive or gerrymandered meanings for our terms. One way to spell this out is in terms of what Lewis (1983) calls “natural properties.” Natural properties are the ones that “carve nature at the joints” (Sider, 2001b, p. xxi; cf. Lewis, 1984, p. 227). They ground similarity (items that resemble each other do so in virtue of sharing some natural property), they are the properties relevant to causation, and they come in degrees, with some properties being more or less natural than others (Lewis, 1983, pp. 346-347). If a property’s eligibility to be a semantic value just is its naturalness, then more natural properties are more eligible to be semantic values. A property P is more eligible than a property Q just in case P is more natural than Q. In the same vein, one interpretation is more eligible than another if and only if the former has more natural properties serving as the semantic values of terms than the latter. This view of language allows us to avoid Putnam’s model-theoretic worries. We want to know which interpretation of the many that fit our use is the one that determines

21 the semantic value of our terms. These interpretations may be ordered by eligibility. Only a proper subset of them also fit use. The most eligible interpretation in that subset is the one that does the semantic-value-fixing. Whether or not a candidate meaning is eligible to be meant is not a matter of linguistic convention, but a matter of fact supplied by the world. Thus, the fact that use is not up to the task of determining content does not entail that our terms do not have determinate reference: meaning is use plus eligibility (cf. Sider, 2001a, p. 191), and these two together are able to stick our terms onto their referents. 2.2.2. Use Plus Eligibility and the Bottom-Up Argument If the use-plus-eligibility view of language is correct, then the semantic values of terms are not determined solely by the way we use them, for eligibility also gets to play a role. Thus, the first premise of the Bottom-Up Argument is false. In this section, I argue that, on the use-plus-eligibility view of language, the thesis of intuitional anarchy entails that the semantic value of “free” is indeterminate between a number of properties. By this I mean merely that there is no fact of the matter as to whether the semantic value of ‘free’ is one of these properties or another. In §2.3, I outline a way to understand this claim that is consistent with the falsity of semantic eliminativism; for now, however, I leave the claim uninterpreted. Call a property under consideration as the semantic value of a term a candidate meaning. According to Sider (2001a, pp. 189-191), the following claim is an upshot of the use-plus-eligibility view of language:

If (i) a term “T” has two candidate meanings, M1 and M2; (ii) neither M1 nor M2

fits “T”-use better than the other; (iii) neither M1 nor M2 is more eligible than the

other; and (iv) there is no candidate meaning M3 that fits “T”-use or eligibility

better than either of M1 or M2, then the semantic value of “T” is indeterminate

between M1 and M2.

The idea, of course, is that use and eligibility exhaust the list of factors that can determine the semantic value of a term. Thus, if two candidate meanings are tied on both the use

22 and eligibility fronts and are not beaten out by any other candidate meaning, there is nothing to make it the case that the term has one value rather than the other. Dual control and psychological integration — two obvious candidate meanings for the term “free” — both seem to be fairly natural properties. Admittedly, they do not carve nature at joints nearly so fine as, say, the properties of being a neutron or of having negative charge do, but neither are they as gerrymandered as, say, the property of being tired on Tuesdays and happy on odd-numbered days. The idea behind the Bottom-Up Argument — and behind the move from (5) to (6) in that argument — can be cashed out in terms of these two properties. Suppose that properties are sets of their (actual and possible) instances. Thus, for instance, the property of being negatively charged is just the set of all (actual and possible) negatively charged things. Call the property of dual control DC and the property of psychological integration PI. Clearly, both DC and PI are good candidate meanings for “free.” So, however, is PI ∪ DC (the disjunctive property of having either dual control or psychological integration). In fact, according to the semantic eliminativist, every property P ⊆ (PI ∪ DC) is a good candidate meaning for “free,” for they all fit “free”-use equally well. They are each favored by at least some of our intuitions at least some of the time, and none is favored by most of our intuitions most of the time. So no P ⊆ (PI ∪ DC) has claim to be the semantic value of “free.” Note, however, that it is extremely unlikely that every P ⊆ (PI ∪ DC) will be equally natural. In fact, there does not seem to be any prima facie reason to think that any P ⊆ (PI ∪ DC) will be more natural than either of DC or PI themselves. Suppose that, in fact, DC and PI are more natural than any other P ⊆ (PI ∪ DC) and that they are equally natural. It seems plain that no candidate meaning will fit “free”-use better than DC or PI. Thus, by Sider’s principle, the semantic value of “free” is indeterminate between dual control and psychological integration. Whether or not there are such properties P that are at least as natural as both DC and PI will depend, in large part, on exactly what dual control and psychological integration are. I have sketched the contours of these properties, but given no positive account of their nature — in part because there is currently no philosophically

23 uncontroversial account to be given. Furthermore, since these two properties are merely standing in for whatever properties our intuitions in fact track, any such account may prove irrelevant. However, even if there are some natural properties that are at least as eligible to be meant by “free” as dual control or psychological integration are, there won’t be very many. Of the plethora of candidate meanings that fit use equally well, the vast majority are bound to be much more gerrymandered and arbitrary than dual control and psychological integration. The number of candidate meanings with high eligibility will be very low. If there is a single property more eligible than both dual control and psychological integration which also fits use as well as these two do, then it is the semantic value of “free.” If there are a few candidates tied on the eligibility front with dual control and psychological integration, then “free” will be indeterminate among all of them. In any of these eventualities, though, “free” is indeterminate among a relatively small number of candidate meanings. For ease of exposition, I will suppose that dual control and psychological integration are equally eligible and more eligible than any competitors with a reasonably high fit with “free”-use. If this supposition is false, the arguments that follow can be reworked substituting whatever candidate meanings for “free” really are the most eligible. 2.2.3. An Eliminativist Objection In the next section, I will argue that the indeterminacy of the semantic value of “free” is consistent with the falsity of semantic eliminativism. A semantic eliminativist, seeing where my argument is going, may complain before it gets there that I have begged the question. The use-plus-eligibility view of language only applies to terms that are already supposed to have semantic content. Offering interpretations for sets of sentences including “Sam freely chose to a” assumes that “freely” is a modifier that functions in a way amenable to interpretation. Unless we had already assumed that “free” was going to affect the truth-value of sentences in which it appears, we would not have asked the interpretations of our ideal linguistic theory to even try to assign it a semantic value. This assumption, however, is just what the semantic eliminativist rejects.

24 The use-plus-eligibility model was invoked in order to counter the Bottom-Up Argument. That argument, I hold, is best viewed as a reductio: If “free” were to have semantic content, then it would have to refer to a single property which satisfied all (or, at least, most) of our intuitions. This is impossible. Therefore “free” has no semantic value. If the argument is a reductio, though, all I have to do in order to counter it is show that no impossibility follows from the assumption made for reductio — in this case, the assumption that “free” has semantic content. I can therefore make use of this assumption without begging the question. The eliminativist might now object that I have only attacked a straw man. Although the argument I presented may fall to the use-plus-eligibility view of meaning, there is another argument, which is not a reductio, which does not. It runs as follows. Since “free” does not pick out any natural-kind property, the only reason we have for thinking it has a semantic value is general pre-theoretical agreement about the conditions under which claims involving “free” are true (Double, 1991, p. 131). Intuitional anarchy precludes this agreement, however, so there is no reason to think “free” has a semantic value. I offered the Bottom-Up Argument as a reductio not because I wanted to attack a straw man but because I think that it is stronger than the alternative just proposed. This alternative relies on the supposition that, at least for non-natural kind terms, a reason needs to be given for thinking that a term has semantic content. This supposition seems strange. “Free” wears its reason for thinking it has content on its face, so to speak. The way it functions in language reveals a presupposition towards thinking it semantically valued. Its syntactic role is shared by terms that are largely so-valued, and the presupposition is that terms that fit into our language in this manner have semantic content. Of course, this presupposition is defeasible. Non-cognitivists about moral terms, for example, agree that the role these terms play in language suggests that they have content. These philosophers then argue that this prima facie support for a cognitivist view of moral terms is undercut by more decisive considerations (cf. Mackie, 1977, pp. 30-35,

25 38-42). Were these considerations absent, however, it would be reasonable to suppose that terms like “good” or “wrong” had semantic value. The same seems true for “free.” Absent some positive reason for thinking it has no semantic value, we have reason to think that it does. The semantic eliminativist thus does better by claiming that intuitional anarchy undercuts the presupposition that “free” has content than he does by claiming both that (i) “free” has no content unless we give reasons to think otherwise and that (ii) intuitional anarchy suggests that we won’t be able to find such reasons. My version of the Bottom-Up Argument is intended to embody the claim that intuitional anarchy undercuts the presupposition that “free” has content. The use-plus-eligibility response defeats intuitional anarchy’s (alleged) presupposition- defeating power. Double may reply that the burden is on those who think that “free” has a semantic value after all, grammatical presuppositions notwithstanding. He favors a “metaphilosophy” that is ontologically austere, on which “we must not let unjustified entities into our picture of what exists” (1996, p. 114). Furthermore, he seems to think that every time we allow a new referential term into our language, we allow a new entity into our ontology. Since entities need justification before we should countenance their existence, they need justification before we should countenance terms that purport to refer to them. We have to justify the existence of free acts and choices before we can countenance the suggestion that “free” has semantic content. One problem with this line is that it endorses a strange view about what counts as adding items to an ontology. Double claims, for instance, that parsimony considerations bar “book-women” — entities which are either books or women — and “emeroses” — entities which are either emeralds prior to a time t or roses otherwise (1996, p. 114) — from our ontology. Every book-woman is either a book or a woman, though, and so long as we have either books or women in our ontology we have not added any new thing by allowing that there are some book-women. Similar remarks apply for emeroses. To say otherwise would be as ludicrous as saying that, although we believe in both the existence of men and the property of being unmarried, we won’t allow bachelors into our ontology for reasons of parsimony. But if we reject this odd view of ontological commitment, so

26 long as we believe in the existence of choices and the properties which our free-will intuitions track, the supposition that “free” has a semantic value does not translate into any additional ontological commitment. Even if we were persuaded by this idiosyncratic understanding of ontological commitment, however, there would still be a difference between a term’s having semantic content and the existence of an item that it refers to. “Married bachelor” has, I think, a semantic value, but there aren’t any in my ontology. Thus, considerations of ontological parsimony seem to have no weight in deciding whether or not a term has a semantic value, so there should be no presupposition against a term’s content on the basis of parsimony. The presupposition is that terms have content, and metaphilosophical considerations of the sort Double alludes to provide no reason to doubt this presupposition.

2.3. Indeterminate Meaning and Contextualism So the semantic value of “free” is indeterminate between dual control and psychological integration. In this section, I argue that this indeterminacy is consistent with the falsity of semantic eliminativism. Specifically, I outline a picture of how free- will terms may function in different contexts such that they make real semantic contribution to discussions of free will, which discussions are themselves in turn truth- valued. Since semantic eliminativism precludes discussions of free will being truth- valued, and since the thesis of intuitional anarchy is consistent with the semantic value of “free” being indeterminate between dual control and psychological integration, this consistency undercuts the Bottom-Up Argument for semantic eliminativism. There is a sense, however, in which at least some of the semantic eliminativist’s claims may be vindicated if “free” is indeterminate between dual control and psychological integration. The claim that “free” has no semantic value remains true, if it is understood as the claim that “free” has no single, unique, univocal semantic value. It is false, however, if it is read as the claim that “free” is, semantically speaking, valueless. “Free” has multiple semantic values, so it has a semantic value — just not a unique one.

27 The semantic eliminativist also may retain the metacompatibilist thesis, which asserts that neither compatibilists nor incompatibilists have the correct theory of free will (cf. Double, 1991, p. 134). If the semantic value of “free choice” is indeterminate between dual control and psychological integration, and if the former is inconsistent with determinism whereas the latter is not, then there is no unique, univocal answer to the question, “Are free choices possible in a deterministic world?” Since each of compatibilism and incompatibilism licenses a unique, univocal answer to this question, neither one can be correct.12 This is a far cry from the full thesis of semantic eliminativism, though, which holds that individual tokens of “free” do not function in a way that could “make discussions of freedom and responsibility truth-valued” or that their content “for various speakers will skew wildly, and there will be no truth to the matter of whose usage is correct” (Double, 1991, p. 9). There is a way for discussions of freedom and responsibility to be truth-valued and there to be a truth of the matter as to whose usage of “free” is correct if “free” is indeterminate between a relatively few candidate meanings. Consider a vague but contextually sensitive term like “flat.” There is arguably no such thing as a single, absolute class of all flat entities, because “flat” is sensitive to contextual considerations. I may, for instance, claim that the lanes in the local bowling alley are flat. If I am discussing what virtues a bowling lane should have, my statement may very well be true. If, however, we are discussing Euclidean geometry, it will probably be false. The microscopic bumps and crevices in the surface of the bowling lane that keep it from describing a Euclidean plane are irrelevant to someone worrying about whether or not his bowling ball will bounce into the gutter (cf. Lewis, 1979, p. 353). Relative to one context of utterance, the bowling lane is flat; relative to another, it is not. Thus, there cannot be any single, context-independent class of flat things: if there were, it both would and would not include this bowling lane. In this respect, “flat” seems somewhat analogous to “free.” There is no univocal, context-independent fact of the matter as to the truth of “a surface can be flat without describing a Euclidean plane,” just as there is no univocal fact of the matter as to the truth of “Choices can be free even if determinism is true.” Furthermore, just as there is no

28 single property that serves as the context-independent semantic value of “free,” there is no single property that serves as the semantic value of “flat.” Sometimes (in, say, Euclidean-geometry contexts) “flat” expresses the property of having zero extension in one of three dimensions; in others, it does not. A bowling lane has the property that is the semantic value of “flat” in bowling-discussion contexts but not the property expressed by “flat” in Euclidean-geometry contexts. Furthermore, there are (arguably) principled reasons for its being in the former but not in the latter that go above and beyond my “unprincipled impressions.” The bowling lane is not flat (relative to a discussion of bowling) simply because I feel it is; rather, it is flat because the purpose of the conversation is such that the lane’s (context- independent) properties make it count as flat. My suggestion for the semantics of “free” should by now be obvious. There is nothing in the thesis of intuitional anarchy that suggests that conflicting intuitions are not evoked in part by different contexts or by different background conditions that would be well suited to evoking different contexts. Double’s discussion of intuitional anarchy is in fact quite amenable to the view that contextual considerations are a part of the explanation of intuitional anarchy (cf. 1991, pp. 123-124, 1996, pp. 106-107). If there is a systematic division of our intuitions that follows contextual features, we can associate the property tracked by each intuition with the context that tends to evoke that intuition. In this case, each context will be well-suited to determining which property serves, in that context, as the semantic value of “free.” Discussions of free will are thus truth-valued, at least in context. For example, it may be that temporal perspective tends to have an important effect on how our free-will intuitions function. Perhaps when we are faced with a decision and ask ourselves whether or not our future choice will count as free we tend to evoke intuitions affected by dual control. We ask ourselves, “Am I free to choose either option?” And perhaps when we look back on our choices, wondering whether or not we are free, we want to know whether our actions were reasonable and rational. In such cases, our intuitions are likely to be affected by psychological integration. Thus, it may be that in deliberative contexts, where we are looking forward to our future act, the

29 semantic value of “free” is dual control, whereas in explanatory contexts, where we are looking back at a past act and trying to understand it, the semantic value of “free” is psychological integration.13 If this contextualist suggestion is correct, then the semantic values of our terms depends on more than simply what we think we mean by them. Contexts are determined by much more than our “unprincipled impressions” (Double, 1991, p. 9). On any contextualist picture, the reasons we want to know why someone is free and the considerations that have been made salient may do much to determine the truth value of some given free-claim regardless of what the speaker’s (or hearer’s) impressions are when the claim was made. If I call a bowling alley “flat” in order to give an example of a Euclidean plane, then what I say is false even if I think it is true. Likewise, if I call a choice that does not exhibit dual control “free” in (say) a deliberative context, then what I say may be false, regardless of whatever unprincipled psychological-integration-favoring impressions I might have. Even if speakers’ unprincipled impressions were the only factor that determined contexts (which, I would contend, is false), it may still not be the case that each speaker’s free-will impressions are the sole determiner of the truth-values of his or her free-will claims. According to a single scoreboard semantics (DeRose, 2004), the truth-values of speakers’ claims are determined not only by the speakers’ own intentions but by the intentions of the other conversational participants. Thus, if five people are discussing free will and four of them mean by “free” something like psychological integration, the fifth participant, when making claims driven by an understanding of “free” in terms of dual control, will likely be making false claims.14 Double (1996, pp. 116-117) may object to the contextualist view outlined here on the grounds that a context-sensitive understanding of free choice cannot ground ascriptions of moral responsibility. On his view, the purpose of free-will terms is to justify moral responsibility ascriptions, where being an appropriate recipient of a moral responsibility ascription is supposed to entail being an appropriate recipient of the following:

30 (1) reactive attitudes, (2) deserved moral praise and moral censure, (3) deserved reward and punishment, (4) membership in a moral community, and (5) certain sorts of ascriptions of autonomy

(Double, 1991, p. 116). Furthermore, moral responsibility is a package deal: no account of free will is any good, by Double’s lights, unless it can ground all five of these. Double claims this would mean that “the sort of free will that agents enjoy would support at most some, but not other, moral properties… Any concept of a free person that presupposes the conjunction of these properties would be unsatisfiable” (Double, 1991, p. 117). It is not clear, however, why this would be the case. Suppose that an agent S performed an action, a, and that in a context C1, “S freely a-ed” is true, but in another context C2, “S freely a-ed” is false. Clearly, “S is morally responsible” had better not be true in C2. But why not say that it is true in C1, so that in that context S is an appropriate target of the entire package of “morality,” whereas in a context C2 “S is morally responsible” is simply false? If there is intuitional anarchy about the conditions under which an agent may be called “morally responsible,” and if some candidate meanings for “morally responsible” are equally eligible to be meant, then “morally responsible” should be indeterminate between these meanings and could, presumably, be sensitive to the same contextual vectors that generated intuitional anarchy in the case of free-will terms in the first place. For those who think that a context-sensitive moral responsibility is implausible, there is another option. Perhaps “morally responsible” is univocal across contexts but Double is too insistent in holding that (1)-(5) stand or fall together. It may rather be that (for example) only psychological integration is required for receiving (1) and (5), whereas only dual control is necessary for receiving (2)-(4). Thus, contexts in which we are interested in (1) or (5) would likely be contexts in which the content of “free” is psychological integration, and ones where we are concerned with (2)-(4) are ones where it is dual control. It is not implausible to think that the standards we must meet for being

31 appropriate targets of reactive attitudes or ascriptions of autonomy are very different than the ones we must meet for being blameworthy and punishable. Nor does this suggestion force us to think that nothing can ground all of (1)-(5), either. There is no a priori reason to think that dual control and psychological integration are mutually exclusive — depending on what, exactly, these properties are, there may be some overlap. Since any choice that counts as “free” on all contexts would be one in which the agent could justifiably receive all of (1)-(5) — i.e., be “morally responsible” as Double understands it — if there is some overlap, agents can be morally responsible in this “full” sense. A semantic eliminativist might object that, since an agent turns out to be “fully” morally responsible (i.e., an appropriate recipient of (1)-(5)) only if she exhibits both psychological integration and dual control, we should think that the “true” meaning of “free” is the conjunctive property of exhibiting both dual control and psychological integration. By “free” we mean “whatever it takes to ground ‘full’ moral responsibility.” However, intuitional anarchy makes it clear that we regard some choices that do not exhibit dual control or that do not exhibit psychological integration to be free in this sense. Thus, goes the objection, the contextualist suggestion is inconsistent with intuitional anarchy. The objection, however, has forgotten the crucial point behind invoking the use- plus-eligibility view of language: the semantic value of our terms is not entirely determined by the way we use them. It may be true that people sometimes hold agents that exhibit only dual control or only psychological integration “fully” morally responsible, but (according to this suggestion) they are wrong to do so. It is false that “free” means “whatever it takes to ground ‘full’ moral responsibility” in all and every context in which we think it means that. Semantic eliminativists may complain that this is not good enough. If the thesis of intuitional anarchy is true and there are some candidate meanings which fit the conditions of Sider’s principle, though, this is as good as it gets. Of course, I cannot forbid anyone from giving up on free will on the grounds that “free” does not mean, in every single instance, exactly what we think it should mean, or ground, in every single instance,

32 exactly the sort of moral responsibility that we think it should ground. For those willing to accept the world’s best efforts to make their sentences come out true, however, a semantically-indeterminate context-sensitive free will offers an attractive alternative to semantic eliminativism.

2.4. Further Implications I have assumed throughout this chapter that both psychological integration and dual control are properties equally eligible to be the semantic values of our terms. This may be false. If these two properties really do both fit the way we use free-will terms equally well (and no worse than any other candidate meanings), then if one is more eligible to be meant than another, it is the semantic value of “free.” In this case, there would be a fact of the matter about the content of “free,” there would be a unique, context-independent fact of the matter as to the truth of “free will is compatible with determinism,” and there would be a fact of the matter as to whether or not people are free in the “fullest” sense. There are a few upshots of this observation. First, the eligibility of the different candidate meanings is a purely metaphysical matter. Thus, if intuitional anarchy is true, the various properties our intuitions track all fit use equally well, and the use-plus- eligibility view of language is correct, then the correct theory of free choices will be settled by pure metaphysics, not conceptual analysis. Intuitions and counterexamples, or at least those at the level of freedom and moral responsibility, lose much of their import, for they all lie on the use side of the equation, and the factors that will tip the balance in favor of one theory or another will all lie on the eligibility side. Of course, even the claim that multiple candidate meanings each fit use equally well is contestable. First, does the thesis of intuitional anarchy entail that all (or many) candidate meanings fit use equally well, or just that none fits use decisively better than any other? Second, we do not even have a clear idea of what counts as use and what does not. Perhaps intuitions about “free” are only one part of use and whatever folk- psychological theory people have is another. In this case, something more than intuitional anarchy would be needed to support the claim that all candidates fit use equally well.

33 Third, we have no clue as to how the use-plus-eligibility view of language evaluates candidate meanings in certain cases where use and eligibility are not tied (Sider, 2001b, p. 186). In the case where one candidate meaning M1 is marginally better suited to “T”- use than another, M2, how much more eligible must M2 be than M1 in order to be the semantic value of “T”? At this point, we do not even have a hint of how to answer this question. Despite this current lacuna in our understanding of the use-plus-eligibility view, there are a few things we can say. If a candidate meaning for a term “T” both fits better with use and is more eligible than all of its fellows, it is the semantic value of “T.” If a number of candidate meanings for “T” are tied on both use and eligibility fronts and not surpassed on either of these by any other “T”-candidate, “T” is indeterminate in meaning among them all. The case of metaphysical eliminativism is especially interesting when viewed in the light of the use-plus-eligibility model of language. Metaphysical eliminativism holds that “free” has as its semantic value a property that cannot be exemplified. In rough outline, metaphysical eliminativists hold that the semantic content of “free” is something like the property dual control, that the possession of dual control entails agent-causation, and agent-causation is impossible. Thus, no (actual or possible) choice can ever be free (cf. Smilansky, 2000; G. Strawson n.d., 2001). Impossible properties are extremely ineligible to be meant (cf. Sider, 2001b, p. 196). Eligible properties are natural ones, and natural properties are properties that “carve nature at the joints.” Impossible properties do not carve anything out of nature — actual or possible. They do not identify any joints in nature or demarcate any lines of real metaphysical distinction. They are thus not very natural, and so not very eligible. It is tempting to say that this fact means impossible properties cannot be the semantic value of terms, but this would be too strong. Some terms do indeed have impossible properties as their meanings. Call a set “m-complete” if and only if its members comprise some complete and finite axiomatization of arithmetic. It is not possible that any set be m-complete — there cannot be a finite axiomatiziation of

34 arithmetic — but we should not say that the property of being a set comprising a finite axiomatization of arithmetic is not the semantic value of “m-complete.” The reason for this should be clear. “M-complete” is defined in terms of other well-behaved and perfectly possible terms. The only way we could make “m-complete” not refer to an impossible property would do great injustice to the terms “set,” “finite,” “axiomatization” and “arithmetic.” It would violate the way we use these words. The reason “m-complete” gets to have an impossible property as its semantic value is that this property is highly favored by use. This tremendous fit with use (in this case, a fit parasitic on the definition, since “m-complete” has probably never been used by anyone before) offsets its extremely low eligibility. Impossible predicates can be meant, but only insofar as this is justified by a fit with use that far outstrips the fit had by any more eligible candidate meaning. Metaphysical eliminativists must therefore support their position by showing that their conception of “free” possesses a far higher fit with the way we use the term than does any competitor conception. Disputants in the free will debate disagree about how to define “free,”15 so it is unlikely that metaphysical eliminativists can rest their case on some potential disruption to other well-behaved terms that would result if “free” were to get some semantic value other than the impossible one they prefer. This means that the metaphysical eliminativist must argue that the way we use the term “free” itself provides the evidence for thinking it has a high fit with use.16 Intuitional anarchy, I take it, precludes any candidate meaning for “free” from having the sort of fit with use an impossible property would need in order to be its semantic value. The metaphysical eliminativist may try to argue that all candidate meanings for “free” are impossible, but this route seems problematic. If our free-will intuitions track different properties, then any candidate meaning for “free” with a reasonably high fit with use will entail that everything that exemplifies it also exemplifies (at least) one of the intuitions that our free-will intuitions track. (For instance, if the intuitions track dual control and psychological integration, then any property that something can have without having at least one of dual control or psychological integration will not fit well enough with use to be countenanced as the semantic value of

35 “free.”) However, it is unlikely that none of the properties tracked by our free-will intuitions are compatibilist-friendly properties. Even those who find compatibilism counterintuitive often admit that they can see some relevant sense of “free” in which “free choices are compatible with determinism” is true, even if they do not think that sense good enough. This suggests that even these opponents to compatibilism feel the slight tug of intuitions that track compatibilist-favoring properties. And there seems to be no reason whatsoever to think that these compatibilist-friendly properties are themselves impossible. Thus, there should be at least some possible candidate meanings for “free.” Metaphysical eliminativists are therefore committed to the rejection of intuitional anarchy. Intuitional anarchy, however, is an empirical thesis: how the vast majority of folk use the term “free” is a question better suited for sociologists than for the armchair speculations of philosophers. Furthermore, even if full-blown intuitional anarchy is false — that is, even if it is false that a number of candidate meanings for “free” are just about tied on how well they fit with use — it may be that no candidate meaning has a high enough fit with use to justify its receiving an impossible property as its semantic value. The considerations adduced to give pre-empirical plausibility to the thesis of intuitional anarchy may at least be taken to show that a single, univocal use of “free” is unlikely, even if one use of “free” occurs substantially more frequently than others. Either of these theses — intuitional anarchy or, barring that, mere intuitional competition — when combined with the use-plus-eligibility view of language results in the falsity of metaphysical eliminativism.

Notes to Chapter 2

1 Paul Churchland (1981) advocates a similar response to mental states, which he thinks should be eliminated. His justification for this view, however, is not closely allied with semantic eliminativism: his position would be best classified (under the current taxomony) as contingent eliminativism about mental states.

2 Note that this suggestion is distinct from Anthony Flew’s (1955) Paradigm Case Argument. The Paradigm Case Argument, in rough form, holds that terms like “free” get their meaning by ostentation (similar to the initial baptism of natural kind terms discussed

36

in §2.1.3) in such a way that a choice counts as free if and only if it is sufficiently like the original choices used to fix the meaning of the term. Since there were original choices used to fix the meaning of the term, free will must be possible, and the odds are good that there could still be choices sufficiently like those ones to count as free. The paradigm- relativeness of concepts is distinct in that there need not be any actual — or even merely possible — paradigm instance of free-ness that fits the mental paradigm used to deploy the concept. (It is unclear, for instance, if the Pure, Rational Ego discussed below is even possible.) Thus there is no reason to conclude, on the basis of paradigm-relative concept deployment, that free will is actual or even possible.

3 He suggests, in fact, that the data support the claim that children have a concept of agent-causation, where that is (minimally) understood as entailing indeterminism. As far as I can tell, his data are consistent with a compatibilist conditional understanding of “could have done otherwise.” This, however, has little bearing on the present discussion.

4 I do not say that a term’s meaning is just its semantic value, but I do intend, by “meaning,” something that has a term’s semantic value (provided the term has one) as a part. I will thus sometimes use “meaning” to refer to a term’s semantic value, but only in places where it is clear by context that I am concerned with the semantic component of the term’s meaning.

5 Generally, on Kripke/Putnam semantics, necessarily, a substance falls under “T” if and only if it has microphysical composition C. There are reasons to be wary of the sufficiency of the substructure C for falling under “T,” however: at least some people think that a world where the laws of nature are such that H2O forms a mushroom-like substance poisonous to humans is a world where H2O is not water (Barnett, 2000). Only the weaker necessity claim is required to raise the worry discussed in this section.

6 It is easier to discuss nominals than modifiers as natural kind terms, so for the balance of this section I will concern myself with the semantic value of “free choice” rather than the semantic value of “free.”

7 This is perhaps too quick. It turns out that the semiprecious stone called “jade” can have one of two microphysical substructures, called “jadeite” and “nephrite.” Presumably, had there been no nephrite on Earth, “jade” would have been a natural kind term that meant “jadeite.” However, despite the lack of a substructure shared by the original samples used to fix the reference of “jade,” it has semantic value. Something is jade only if it is either jadeite or nephrite. Likewise, if there are unifying substructures for each of psychological integration and dual control, the semantic value of “free choice” might simply entail that a choice is free only if it has either the underlying substructure associated with psychological integration or the one associated with dual control.

37

8 I do not say that every meaningful predicative phrase has some unique property as its semantic value. If there is a property which serves as the semantic value of “being a property that is not exemplified by itself,” we will be forced into Russell-style paradoxes about properties. (Suppose P is the semantic value of the above expression. Does P exemplify P? If yes, then no, but if no, then yes.) However, the above property cannot be generated truth-functionally out of properties that do not themselves give rise to paradoxes, so this fact does not forbid us from allowing truth-functionally connected compound properties.

9 Double (1991) makes the argument extensional, with the conclusion that “there is no class of entities that constitutes free choice” (p. 131), but this is too weak for his purposes. Metaphysical and contingent eliminativists will agree that there is no class of entities that constitutes (or is designated by) free choice but reject the claim that “Our proclaiming choices to be free… can be nothing more than our venting of non-truth- valued attitudes, none of which is ‘more correct’… than competing attitudes” (Double, 1996, p. 3). They think there is a correct, truth-valued attitude about the matter: all choices are unfree.

10 One source aside from intuitions that affects the way we use our terms surely does not deserve to be included. Theoretical commitments can incline us to make different judgments, but these inclinations do not count as intuitions since they are post-theoretical. Philosophers committed to some theory of free will or another will likely have the way they use the term affected by this commitment. However, insofar as these philosophers are trying to discover (rather than create) the semantic value of “free,” their post- theoretical commitments should get no place in determining what this semantic value is. Insofar as they are trying to create such a value, their use of the term is irrelevant to the project at hand, which is one of discovery, not of creation.

11 I take it that the problem is just as poigniant for partial use theory. In partial use theory, only interpretations where values of natural kind terms are properly related to microphysical substructures are eligible to fix the reference of the rest of our language. So long as sufficiently many terms are neither natural kind terms nor definable using only natural kind terms, there will still be an abundance of interpretations that satisfy ideal linguistic theory and no way to choose between them. We thus run out of the “semantic glue” needed to stick referents onto these natural-kind-independent terms.

12 Readers anticipating the contextualist suggestion below may wonder whether metacompatibilsm is entailed by a context-sensitive “free.” Assuming that dual control is incompatible with determinism, in contexts where the value of “free” is dual control, tokens of “free will is compatible with determinism” will be false; in contexts where the value of “free” is psychological integration, tokens of this sentence will be true. If compatibilism and incompatibilsm each entail that there is a context-invariant answer to

38

the question “is free will compatible with determinism?” then neither compatibilism nor incompatibilism will be correct.

13 See Graham & Horgan, 1994 and Hawthorne, 2001 for suggestions that free will judgments might be contextually sensitive. Both of these suggestions differ from mine in that they suggest “free” is a “graded” term — like “flat” — where what varies by context is either (i) how similar possible worlds must be to the actual in order for their contents to be relevant to actual agents’ freedom (Graham and Horgan, 1994, pp. 240-241) or (ii) how many kinds of “determiners” we are ignoring (Hawthorne, 2001). My suggestion is not that the standards for freedom changes with context — i.e., that how psychologically integrated or how much dual control an agent needs in order to count as “free” varies with context — but rather that the nature of the property designated by “free” changes by context.

14 DeRose prefers what he calls an “exploding scoreboard semantics” on which, when speakers disagree, the claims they disagree about lack truth values. He concedes, however, that there is no decisive reason to reject other single scoreboard semantics on which, in such a situation, one speaker or the other “wins” by having their claim turn out true. On such a semantics, all claims may have a determinate truth value.

15 At least in any interesting way — it is trivial to define (say) “free choice” as a choice that exhibits free will, but that does not address the root issue.

16 The contingent eliminativist rejection of free will is analogous to that of the metaphysical eliminativist, except that the property serving as the semantic value of “free” is not taken to be impossibly satisfied, but simply not as a matter of fact satisfied. It is not clear what the use-plus-eligibility theory would say about this view; does the non-actual exemplification of a possible property count against its eligibility or not?

39

CHAPTER 3 THE INCOMPATIBILITY OF FREE WILL AND NATURALISM

The Consequence Argument has long been a staple in the defense of libertarianism, the view that free will is incompatible with causal determinism and that humans have free will. It is generally (but not universally) held that libertarianism is consistent with a certain naturalistic view of the world — that is, that libertarian free will can be accommodated without employing ontological commitments beyond those provided by the indeterminism posited by our best (quantum) physics. In this chapter, I argue that libertarians who support their view with the Consequence Argument are forced to reject this naturalistic worldview, since the Consequence Argument has a sister argument — I call it the Supervenience Argument — which cannot be rejected without threatening either the Consequence Argument or the naturalistic worldview in question.

3.1. The Consequence Argument The Consequence Argument purports to show that free will is incompatible with causal determinism, where the latter thesis is understood as the claim that the laws of nature, conjoined with any proposition accurately describing the entire state of the world at some given time, entail any other true proposition.1 An informal version of the argument runs as follows:

If determinism is true, then our acts are the consequences of the laws of nature and events in the remote past. But it is not up to us what went on before we were born, and neither is it up to us what the laws of nature are. Therefore, the consequences of these things (including our present acts) are not up to us (van Inwagen, 1983, p. 56).

40

If the argument is sound, determinism is incompatible with free will. This argument can be clothed in formal garb. This garb makes use of a modal operator, “N,” where “Nφ” means “φ, and no one has, or ever had, any choice about whether φ” (van Inwagen, 1983, p. 93).2 (I will follow Alicia Finch and Ted A. Warfield (1998, p. 516) in understanding “someone has a choice about φ” as “someone could have acted so as to ensure the falsity of φ,” and I take falsifying φ to be the same as ensuring the falsity of φ.) Originally, the Consequence Argument used two inference rules:

(α) From “□φ,” deduce “Nφ,” and (β) From “Nφ” and “N(φ → ψ),” deduce “Nψ”

(van Inwagen, 1983, p. 94), where “□” represents broad logical necessity. Unfortunately, these two rules are jointly invalid. Taken together, they entail that the N-operator is agglomerative. That is, if (α) and (β) are valid, then so is

(Agg) From “Nφ” and “Nψ,” deduce “N(φ & ψ).”3

(Agg) is demonstrably invalid, as has been shown by Thomas McKay and David Johnson (1996, p. 115). In their example, we consider an agent who does not flip a coin, but could have. In this case, “N(the coin does not land heads)” is true, and “N(the coin does not land tails)” is true. To ensure, for instance, the falsity of “the coin does not land heads,” one would have to ensure that the coin does land heads. This, presumably, is not something anyone could do. Yet “N(the coin does not land heads & the coin does not land tails)” is false. The agent could have falsified the embedded conjunction by flipping the coin.4 In response to this and other problems, proponents of the Consequence Argument have rejected (α) and (β) in favor of another inference rule which does not entail (Agg):

(β□) From “Nφ” and “□(φ → ψ),” deduce “Nψ,”

41

(cf. Finch & Warfield, 1998, pp. 521-522; Widerker, 1987, p. 41). If “P” is a proposition that expresses the state of a world at a remotely early time (before there were any human agents, say), “L,” a conjunction of all the laws of nature, and “F,” any true proposition whatsoever, then the Consequence Argument is as follows:

The Consequence Argument (1) N(P & L) Premise (2) □((P & L) → F) Assumption of Determinism (3) NF β□: 1, 2

Thus, if determinism is true (and if no one has, or ever had, a choice about the truth of the conjunction of the laws of nature with a proposition expressing the state of the world in the remote past), then no one has ever had a choice about anything. What of the first premise? It is highly intuitive that we cannot do anything to change the laws of nature — i.e., we cannot do anything that would ensure the falsity of the laws (and hence we “have no choice” about them). It is likewise intuitive that we cannot do anything to change the past (van Inwagen 1983, p. 96). It seems intuitive that, as a result, we have no choice about the conjunction of these two propositions. Since, as we have already seen, the N-operator is not agglomerative, we cannot rely on some sort of entailment to get “N(P & L)” from our assent to both “NP” and “NL.” But while this is formally correct, it may not be much of an obstacle: (1) does not seem to be a plausible candidate for rejection even given the general invalidity of N- agglomeration. Finch and Warfield describe it this way:

[T]he core intuition [described above] motivates the acceptance of [the first] premise. This core intuition is, we maintain, the intuition that the past is fixed and beyond the power of human agents to affect in any way. P describes the state of the world at some time in the distant past (before any human agents existed). L is a conjunction of the laws of nature which, we presume, in addition to being

42 inalterable by human agents, do not change over time. Thus the conjunction (P & L) offers a description of what might be called the “broad past” — the complete state of the world at a time in the distant past including the laws of nature. We maintain, in asserting our premise, that the broad past is fixed [in a way that justifies N(P & L)] (1996, p. 523).

Thus we need not appeal to agglomeration to justify “N(P & L)” given our intuitions that “NP” and “NL” are true, because those intuitions directly support “N(P & L)” without any formal mediation.

3.2. Naturalism and the Supervenience Argument There is a view about the nature of reality, which I will tag with the over-worked name of naturalism, which in rough form holds that everything around us eventually boils down to fundamental physics. This is not necessarily a reductionistic view (although global reductionism is one variant of naturalism), but rather a supervenience thesis which holds that all events and causal relations supervene on microphysical events and causal relations. As I understand it here, naturalism consists of two main claims:

(N1) All events supervene on microphysical events, and (N2) All causal relations supervene on microphysical causal relations.

Supervenience is understood as a merely logical relation: if an event a supervenes on an event b, then it is impossible that b occur and a fail to occur. Stronger relations are also sometimes called “supervenience,” including relations of dependence and constitution, but for our purposes we need only this weaker supervenience relation.5 There are stronger and weaker versions of each of naturalism’s claim. The strong form of (N1) holds that every event logically supervenes on the microphysical — that is, any two possible worlds which differ with respect to which events occur in them differ with respect to which microphysical events occur as well. The weak form requires only

43 that the supervenience between events be nomic, so that any two possible worlds with divergent events have either divergent microphysical events or divergent laws of nature.6 Claim (N2) also comes in a variety of strengths, the strongest holding that for any events c and e, c caused e if and only if c’s microphysical supervenience base caused e’s microphysical supervenience base (cf. Kim, 1984, p. 99). The weak version I will concern myself with here holds merely that if c causes e, then there is a microphysical causal chain running from some events in c’s supervenience base to some events in e’s supervenience base which forms part of a complete microphysical causal chain for e’s supervenience base. Call the version of naturalism generated by combining this diluted sort of causal supervenience with merely nomic supervenience of events on the microphysical weak naturalism. Weak naturalism is weaker than it might be, but it is stronger than some other views that have gone under the title of “naturalism.” Of especial interest to us here are the self-proclaimed “naturalistic” agent-causal theories of Timothy O’Connor (2000) and Randolph Clarke (2003). Insofar as both of these theories are consistent with something like (N1), there is a sense in which they deserve to be called “naturalistic.” Neither, however, claims to be consistent with (N2), since they both posit forms of substance causation not meant to supervene on microphysical event-causation.7 As I am using the term “naturalism” here, I mean to rule agent-causal theories, including O’Connor’s and Clarke’s, out. The Supervenience Argument is supposed to undercut libertarian theories that attempt to build free will without adding ontological commitments that go beyond those of current science. It is not intended to target agent-causal or other theories that “postulate unusual forms of… causation” (Kane, 1996, p. 115), no matter how friendly to current science those theories may try to be.8 The Supervenience Argument is designed to show that, if weak naturalism is true, the class of actions about which someone has or ever had a choice is empty. Since stronger forms of naturalism entail the weak form, it follows that any form of naturalism is inconsistent with the existence of actions of this sort. As with the Consequence Argument, there is an informal version of the Supervenience Argument:

44 If weak naturalism is true, then our acts are the consequences of the laws of nature, events in the remote past, and the outcomes of undetermined microphysical events. But it is not up to us what went on before we were born, what the laws of nature are, or how undetermined microphysical events turn out. Therefore, the consequences of these things (including our present acts) are not up to us.

This argument, or at least something very much like it, has already found favor in the eyes of some. Trenton Merricks, for instance, gives what I take to be a version of it in his book Objects and Persons (2001, pp. 155-161). Others have given related arguments against the compatibility of the sort of free will libertarians demand with a broadly naturalistic worldview (e.g. Bishop, 2003; Loewer, 1996; Unger, 2002). However, no one has taken the trouble to clothe the Supervenience Argument in the same formal garb worn by the Consequence Argument. As a result, we do not yet have a clear idea of how closely the Supervenience Argument is tied to the Consequence Argument and whether or not there is some move to be made that will undermine the former while leaving the latter intact. The remainder of this chapter provides this formal clothing and considers whether there is a way to reject the Supervenience Argument that leaves the Consequence Argument intact. 3.2.1 Choosy Actions Call an event a choosy if and only if “A & ~NA” is true, where “A” is a proposition expressing the occurrence of a. The Consequence Argument is meant to show that, if determinism is true, “NF” is true for any true F. Presumably, though, if there is any true “F” for which “~NF” is true — i.e., if the conclusion of the Consequence Argument is avoided — then there will be a choosy action. Recall that we are reading “Nφ” as “φ, and no one could have acted so as to ensure the falsity of φ” (Finch & Warfield, 1998, p. 516). Thus, if “φ & ~Nφ” is true for some φ, then there is some possible action, a, which nobody performed but someone could have performed and, had a occurred, it would have ensured the falsity of φ. If omissions are actions, then the actual

45 action of not-a-ing is a choosy action, since it did occur and someone could have ensured the falsity of the proposition that expresses this occurrence. I would like to avoid committing myself to the view that omissions are actions. Plausibly, though, if someone could have a-ed at a time t but did not, then there will be some other action the agent performed at or before t which she would not have performed had she a-ed instead. Consider a case of an omission that is plausibly not an action. Jane, who works night shifts, wants to avoid voting in a local election. In order to do this, she simply sets her alarm for 7:00 p.m. on election day the way she does every other day, goes to sleep before the polls open, and wakes up after they are closed (Mele, 2003, p. 151). Jane, it seems, could have voted at (say) 3:00 p.m. But Jane was sleeping at 3:00 p.m., so it sounds strange to say she performed any actions then at all, including any negative actions of not voting. However, it is clear that in Jane’s case there were actions she did perform before 3:00 p.m. — setting her alarm for 7:00 p.m., and, more importantly, deciding not to vote, both of which happened before she went to bed — that she would not have performed if she had voted at 3:00 p.m. instead. It appears that Jane could have ensured the falsity of “Jane decided not to vote,” so her decision is a choosy action. It appears that, in general, if there are true propositions φ such that φ & ~Nφ, then there are some choosy actions. If this is correct, then there are choosy actions if and only if the conclusion of the Consequence Argument is false. Since the conclusion of that argument is supposed to preclude the existence of free will, the existence of choosy actions must be (at least by the lights of proponents of the Consequence Argument) a necessary condition for free will. If there are any choosy actions, there is a first one.9 There are two minor wrinkles, both of which deal with ties. To begin with, on one popular account of action- individuation, a single bodily movement may comprise a large number of actions, and an individual may perform multiple actions simultaneously. Thus there would be no first choosy action because the first time someone acts in a choosy manner there will be many actions that all begin simultaneously. We may accommodate this fine-grained account of action-individuation by a little formal maneuvering. Suppose that, at a time t, an agent S’s

46 behavior counts as multiple actions on a fine-grained account of action but as a single action on a coarse-grained account. Clearly, even for the fine-grained theorist, the actions S performed at t bear some sort of similarity to each other that they do not bear to other actions performed by S at t, or to actions performed by other agents, whether at t or not.10 So we can shift our discussion from that of actions to that of equivalence classes of actions (using this similarity relation) and argue that the set of equivalence classes of choosy actions must have a first member. For stylistic reasons, the coarse-grained way of speaking will be used in the balance of this chapter whenever possible; for my purposes, talking of equivalence classes of actions will add complication without enlightenment. Considerations of action-individuation aside, there is still the potential for genuine ties for the first choosy action. If two choosy actions a and b are performed simultaneously, and there are no choosy actions that occur before a and b, then the class of choosy actions will not have a first element. For our purposes, though, we will be able to get by with an artificially restricted notion of “first.” We can stipulate, for instance, that whichever of a or b is highest and closest from the northeast to the intersection of the international date line, the equator, and sea level, is the “first” element of the class. It is not possible that there be any ties in this competition. Our purpose in locating a “first” element of the class of choosy actions is to allow us to argue that the class of choosy actions is empty. We do this by showing that the first element in the class of choosy actions is not choosy. (The argument has the form of a reductio: Suppose the class is non-empty. Then it has a first element, which is choosy. But that element is not choosy. Thus the class is empty.) In order to make the argument work, we need only an ordering with the following properties: (i) for every set S of choosy actions, S has a first element, and (ii) if a occurs “before” b, then b cannot be such that its occurrence or non-occurrence could affect the occurrence of a. Clearly, our artificial ordering satisfies both of these properties, so it is suited to do the work we need it to do.11 3.2.2 Causal Relations The Consequence Argument is supposed to show that, if determinism is true, people do not act freely. It does this by showing that, if determinism is true, then “NF” is

47 true for any true “F.” So, by the lights of the Consequence Argument, if anyone ever acts freely, there are true propositions “F” such that “~NF.” Given considerations of the previous section, if there are some free actions, then by the lights of the Consequence Argument, there are some choosy ones. If there are some choosy actions, there is a first choosy one. Call it r, and suppose it was performed by an agent S. For illustration, suppose that the causal theory of action is true. Then r, by virtue of being an action, will have been caused by some particular pair of desires and beliefs (or their neural realizers), which I will call db. But db probably will not encompass all of the causes of r. Causal theorists seldom think that a desire/belief pair alone is nomically sufficient for an action. Other inner states of the agent, as well as external, environmental factors, etc., will likely figure into the causal story of most actions. So let db+ represent the sum total of what we would call the causes of r if we knew enough about r’s production. If weak naturalism is true, both db+ and r will supervene on microphysical events. For the sake of exposition, I will suppose they each have only four events in their supervenience base (db1, …, db4 and r1, …, r4, respectively); a more realistic count would be unworkable. Since db+ caused r and causal relations supervene (in a weak sense) on microphysical causal relations, there will be some microphysical causal chain running between db+’s supervenience base and r’s supervenience base. Of course, not everything in r’s supervenience base will have been caused by something in db’s supervenience base; let us suppose that r1 and r2 are caused by db1, …, db4, whereas r3 and r4 are not.

Suppose for illustration that the causal chains work in the following way: db1 causes an event, e1, which in turn causes another event, d1. Meanwhile, db2 and db3 jointly cause an event e2, and db4 causes an event e3. Then e2 causes d2 and e3 causes r2.

Finally, d1 and d2 jointly cause r1. Suppose further that c1 and c2 are the causes of r3 and + r4, respectively. The causal chain between db ’s supervenience base and r’s supervenience base will then look as it does in Fig. 1.

Now, if this is a deterministic universe, the occurrence of db1, …, db4, c1, and c2 will be nomically sufficient for the occurrence of r’s supervenience base (and therefore nomically sufficient for r). If we are talking about choosy events, however, libertarians

48 will hasten to remind us that the universe — and this causal chain in particular — had better not be deterministic. So we shall suppose it is not.

db1 e1 d1 r1

db2 r2 e2 d2

c1 r3 db3

db e r 4 3 c2 4 Figure 1 The causal structure of r’s supervenience base

If r is not going to be deterministically caused, then some link in the microphysical causal chain will have to be indeterministic. There are two places indeterminism could crop up. First of all, one of the events in the causal chain “in between” db+ and r may have been only indeterministically caused by its antecedents. Or the indeterminism could occur “at the end” of the chain: one of the events in r’s supervenience base may have been only indeterministically caused. Let us begin by supposing that the indeterminism is of the first kind — the kind that only crops up “in the middle.” Suppose, for example, that e2 is the indeterministic culprit, and call it in from here on out to emphasize its indeterministic nature.

db1 e1 d1 r1

db2 r2 in d2 db3 c1 r3

db4 e3 c2 r4 Figure 2 “In-the-middle” indeterminism 49 Events db2 and db3 cause in, but only indeterministically: there are worlds with the same laws of nature in which db2 and db3 occur but in does not. In this case, the occurrence of the db’s and the c’s will not be nomically sufficient for the occurrence of the r’s. However, the occurrence of the db’s, the c’s, and in will be sufficient for the occurrence

of the r’s. Therefore, if “DB” is a proposition expressing the occurrence of db1, …, db4,

“C” a proposition expressing the occurrence of c1 and c2, “IN” a proposition expressing the occurrence of in, and “R1,” …, “R4” propositions expressing the occurrence of each of r1, … r4, respectively, then

□((DB & C & IN & L) → (L & R1 & R2 & R3 & R4)).

Since r supervenes nomically on r1, …, r4,

□((L & R1 & R2 & R3 & R4) → R).

Combining these two facts, we get

□((DB & C & IN & L) → R).

3.2.3 The Supervenience Argument: A First Pass We are now in a position to offer a formal version of the Supervenience Argument. The first premise is the nomic sufficiency of the db’s, the c’s, and in for r defended above. The second is that no one has, or ever had, a choice about whether DB & C & IN & L. It should be clear how the argument is supposed to go:

The Supervenience Argument (1) □((DB & C & IN & L) → R) Premise (Naturalism) (2) N(DB & C & IN & L) Premise (3) NR β□: 1, 2

50 So r is not a choosy act. Recall, though, that r was supposed to be the first choosy act; it follows that there are no choosy acts at all. If the argument is sound, no one has, or ever had, a choice about anything. The second premise seems to follow from the “broad past” principle appealed to by Finch and Warfield with respect to the Consequence Argument. In that instance, the intuitions supporting “N(P & L)” were that both “P” and “L” were true long before there were any humans around and that the past is fixed. Apparently, the idea is that, since “P & L” was true before anyone could have done anything to falsify it, and since we cannot now do anything to falsify what has gone on before, nothing we can now do could falsify “P & L.” Similar reasoning lends support to “N(DB & C & IN & L).” The proposition “DB & C & IN & L” is made true before r occurs, and r is the first choosy act. Thus, nobody could have done anything to falsify it at the time it was made true. (Suppose they could have. Then what they did instead would be or be preceded by a choosy act. But r is the first choosy act, so they could not have.) By the time r comes around, “DB & C & IN & L” is already a fixed part of the past. Of course, one may object that “DB & C & IN & L” is not part of the remote past, since it occurs very soon before r. This appeal to the remoteness of the past is a red herring. It is not as though we think the recent past is only somewhat fixed, and we can change it a bit, whereas as time goes on it “solidifies” until it is eventually unchangeable. Rather, the past — remote or not — cannot be changed by anything we can do now. The only reason to appeal to a “remote” past in defense of the Consequence Argument is to make sure that we do not appeal to a time at which people (not necessarily we) were going around performing choosy actions. If our proposition is made true before the first choosy action, though, we are in the clear. 3.2.4 The Argument for Trickier Cases In the above argument, I supposed there was a microphysical causal chain between the db’s and the r’s with indeterministic links only “in the middle,” as it were. But there may instead be a causal chain from the db’s up to but not including r in which r itself is the undetermined link. This means that there could be two possible worlds (with

51 the same laws of nature) in which the entire causal chain strictly between the db’s and the r’s occurred, but the r’s occur in only one of them. How would this go? It is implausible that r is a microphysical event. Microphysical events are (in general) too small to be actions. Since the occurrence or non-occurrence of r will have to supervene on something microphysical, and since r was undetermined by everything that went on before it, there must be some undetermined microphysical event — call it x — concurrent with r,12 in r’s supervenience base. In other words, x is the microphysical event that “makes the difference” between the occurrence and non-occurrence of r.

Suppose r4 is x, the undetermined event. (We shall call it x from here on, thus making r’s supervenience base r1, r2, r3, and x, as shown below.)

db1 e1 d1 r1

db2 r2 in d2 db3 c1 r3

db4 e3 c2 x Figure 3

“At-the-end” indeterminism

It appears that nobody has, or ever had, any choice about “X,” the proposition that expresses the occurrence of x. How could anyone exercise such control over the truly objective chance happenings of particle physics? What could I do, for instance, to ensure that an electron will have a certain property at a certain time, if it is objectively undetermined whether or not it will gain said property at said time? (Cf. Loewer, 1996; van Inwagen, 1983, pp. 142-143.) As far as I can see, there is nothing I (or anyone) could do that would determine the outcome of an undetermined microphysical event. What I would like to do is

52 agglomerate “N(DB & C & IN & L)” and “NX,” which would allow me to offer the following argument.

The Tricky Supervenience Argument

(1) □((DB & C & IN & L & X) → (L & R1 & R2 & R3 & X)) Premise (N2) (2) N(DB & C & IN & L & X) Premise

(3) N(L & R1 & R2 & R3 & X) β□: 1, 2

(4) □((L & R1 & R2 & R3 & X) → R) Premise (N1) (5) NR β□: 3, 4

The first premise is unproblematic: “DB & C & IN & L” entails “R1 & R2 & R3” since r1,

r2, and r3 are nomically necessitated by the causal chain leading up to them, and “L & X” trivially entails “L & X.” The premise in the fourth line of the argument is equally

unimpeachable, since it simply expresses the nomic supervenience of r on r1, r2, r3, and x. The problem is that I cannot use agglomeration to support the second premise, and X does not lie in the “broad past” of r. Nonetheless, I claim that “N(DB & C & IN & L & X)” is true. According to Finch and Warfield,

…it is important to be clear that the McKay and Johnson argument [against agglomeration] shows only that the inference from Nφ and Nψ to N(φ & ψ) is invalid. This does not, by itself, provide any reason at all for thinking that [in the case of NP and NL] NP and NL are true, while N(P & L) is not. An inspection of the difference [between the two cases] shows that the McKay/Johnson case seems to cast no doubt on the truth of N(P & L). In the McKay/Johnson case, one has no choice about either conjunct of a conjunction but does have control over the conjunction because although there is nothing one can do that would falsify either particular conjunct there is something one can do that might falsify either conjunct and would falsify the conjunction… [I]t is not at all plausible that though one

53 cannot, for example, do anything that would falsify… the laws of nature, one might somehow do so. (1996, pp. 523-524, emphasis added)

Similar remarks apply here. There does not seem to be anything one could do that even might ensure the non-occurrence of x, the truly undetermined event, or might falsify true propositons about the past,13 and by virtue of this would falsify “N(DB & C & IN & L & X).” Premise (2) is vindicated and the argument follows.

3.3. Implications The Supervenience Argument was presented above as involving a particular act with a particular causal history, but its generality should be clear. We of course have no idea what the actual causal history of the (alleged) first choosy act is like. If we subscribe to naturalism,14 though, we will be committed to the supervenience of the first choosy act on some set of microphysical events. Furthermore, there will be some microphysical causal chain running from the supervenience base of whatever caused the action to the supervenience base of the action, and some of the events in the chain will be indeterministic while others won’t be. If we let “IN” express the occurrence of all the indeterministic elements in the causal chain and “X” express the occurrence of all the indeterministic elements in the action’s supervenience base we can use (β□) to generate essentially the same argument for any alleged first choosy action. If the foregoing is correct, then it looks like anyone who finds the Consequence Argument a persuasive argument for incompatibilism should also hold that the existence of choosy actions — and therefore the existence of free will — requires the falsity of naturalism. In the following section, I defend the Supervenience Argument against various objections. In this section, I consider the ramifications if that defense is successful. The Supervenience Argument occupies a position in rhetorical space similar to that of the Mind argument. The Mind argument was supposed to show that, if (β) is valid, then free will is also incompatible with indeterminism (van Inwagen, 1983, pp. 142-150; note that van Inwagen calls this the “third strand” of the Mind argument). According to

54 the argument, if indeterminism is true, then if an action r has a particular set of indeterministic causes db, nobody could have done anything to ensure that db caused r (van Inwagen, 1983, pp. 142-143). Thus, N(DB → R). And, if r is the first choosy act,15 then N(DB). A single application of (β), however, yields the conclusion that NR. Thus r is not a choosy act after all, and so choosy acts and therefore free will are non-existent. Indeterminism precludes free will. Thus, libertarians have as much reason to reject (β) as compatibilists do. Before the advent of independent counterexamples to (β), the Mind argument placed a lot of pressure on libertarians. They could not use their favorite argument against compatibilists without opening themselves up to charges of self-contradiction. Once a version of the Consequence Argument that made use of (β□) was available, though, the Mind argument lost most of its rhetorical force. Libertarians could now happily reject (β) — and the Mind argument with it — without giving up the Consequence Argument (Finch & Warfield, 1998). Of course, if the Mind argument could also be reworked to use (β□), libertarians would once again be in a bind. But it cannot. The best attempt at such a reconstruction has been given by Dana Nelkin (2001). She argues that, if r is the first choosy action and is indeterministically caused by db, then both “DB” and “DB → R” lie in the “broad past” of r. She then reformulates the argument as follows:

Nelkin’s Revised Mind Argument (1) N(DB & (DB → R)) Premise (2) □((DB & (DB → R)) → R) Logical Truth (3) NR β□: 1, 2

Notice that “(DB & (DB → R))” is truth-functionally equivalent to “(DB & R),” so at first blush it appears too strong a premise for libertarians to accept. As O’Connor (2000, n. 12) notes, it would be “an unusually inept compatibilist” that would accept a Consequence argument requiring “N[(P & L) & ((P & L) → F)]” as a premise, since they would be in essence granting “N((P & L) & F).” Likewise, it would be an unusually inept

55 incompatibilist who would accept (1), at least without an argument, for it comes uncomfortably close to simply granting “NR.” Of course, Nelkin has an argument: both “DB” and “DB → R” lie in the “broad past” of r (2001, p. 113). The problem now is that this claim appears false. In particular, in an indeterministic world where “DB” is true, “DB → R” — equivalent to “~DB ∨ R” — appears to be made true no earlier than “R” is. It is thus difficult to see how it could then lie in r’s “broad past” — it seems more accurate to describe it as simultaneous with “R.” Unless a better argument is forthcoming for (1), the (β□)-revised Mind argument appears to have lost.16 In the absence of a working (β□)-Mind argument, the Supervenience Argument can put a similar sort of pressure on libertarians: give up (β□) or give up on free will in order to remain consistent. This pressure is limited, however, to libertarians who are also naturalists; non-naturalistic libertarians are free to simply reject the assumptions that generate certain premises of the Supervenience Argument. If the pressure is limited in this way, though, then is the Supervenience Argument even worth bothering with? Yes. First, it is not insignificant that naturalistic libertarians cannot use the Consequence Argument to support their position. The most popular libertarian position on the table — that of Robert Kane (1996, 1999) — is explicitly naturalistic, and in its wake, other attempts to secure libertarian free will in a naturalistic framework — like that of Laura Waddell Ekstrom (2000) — have been proposed. Kane, however, is committed to the claim that if agents have free will of the sort his theory calls for, then the Consequence Argument shows that determinism precludes a necessary condition for free will (1996, pp. 75-77), and Ekstrom relies primarily on the Consequence Argument to defend her incompatibilism (2000, pp. 26-42).17 If the success of the Consequence Argument as an argument for incompatibilism entails the incompatibility of free will and naturalism, then two well-received theories of libertarian free will fall to internal inconsistency. Second, there is a reason views like Kane’s and Ekstrom’s are so popular. In the philosophical arena, positions are evaluated on a number of merits, one of which is overall plausibility. Naturalism is considered by many to be an extremely plausible view,

56 and while its denial may not count decisively against a position, it is a cost to be avoided. Naturalistic libertarianism, by making use of the indeterminism ready-to-hand in the natural order, thus comes cheap. Non-naturalistic libertarianism, by contrast, comes with a high cost. For some, at least, that cost will seem high enough to warrant a rejection of either (β□) or free will rather than payment. This point can be put another way. Van Inwagen (1992, p. 58) has suggested that the debate between compatibilists and incompatibilists should be viewed as taking place in front of an audience as of yet undecided about whether or not free will is compatible with determinism. The goal of each debater is not the overly ambitious one of convincing her opponent, but of converting the agnostic audience to her position. Presumably, the debate between libertarians and their opponents (compatibilists and skeptics) should be viewed the same way: the libertarian is trying to convince the agnostic audience both that we have free will and that it is incompatible with determinism. As already noted, naturalism is a widespread view. Many in the agnostic audience are liable to resist the thought that free will has to be sought outside the scientific world. Many philosophers become skeptical when their colleagues begin to “look for [free will] in mysterious sources outside of the natural order or to postulate unusual forms of agency or causation” (Kane, 1996, p. 115). Naturalism, for better or for worse, is a widely held view, and people do not want to sacrifice it on the free will altar if an alternative can be found. This means that a defense of libertarianism in general is much more difficult. Before, a non-naturalistic libertarian might have been able to convince many in the agnostic audience to at least be libertarians even if they would not go all the way with her into non-naturalism. Now, however, convincing the audience that libertarianism is warranted consists in convincing them that the high non-naturalistic costs of libertarianism are both necessary for free will and worth paying. This is likely to be a tough sell.

57 3.4. Objections and Replies Naturalistic libertarians motivated by the Consequence Argument are in something of a bind, for they must find a way to reject the Supervenience Argument which does not in turn license a rejection of the Consequence Argument. In this section I will consider potential objections to the Supervenience Argument and, in each case, argue that either the objection fails or that, if it is successful, a parallel objection can be used against the Consequence Argument.

(Objection 1) Suppose that the universe is such that, for every time, there is an earlier time at which someone performed a choosy action. Then a set of all choosy actions would not have a first element, but it clearly would not be empty.

This is correct, and it is the only way a non-empty, linearly-ordered class of choosy actions could fail to have a first element. The first thing to note is that this universe is not, it seems, our universe, and so we can view the Supervenience Argument as an argument to the effect that, if agents have free will in our universe, then our universe is not one in which naturalism is true. One might not think that this reply is very compelling on the grounds that the argument is supposed to show the conceptual incompatibility of naturalism and free will. This is too strong, though, for if the possibility of the sort of universe in question undermines the Supervenience Argument, it also undermines the Consequence Argument in exactly the same way. In a universe with an infinite backwards cascade of choosy actions, there is no time t such that a proposition expressing the state of affairs of that universe at t is one about which no one has or ever had a choice — i.e., there is no “P” expressing the state of the universe at a time t such that “NP.”18 Thus, this objection faces a dilemma. Either the existence of (non-actual) naturalistic universes with infinite backwards cascades of choosy actions does not pose a problem for the Supervenience Argument, or if they do, similar deterministic universes pose a problem for the Consequence Argument.

58 (Objection 2) Your argument begins from the premise that some beliefs and desires (or their realizers) caused r and proceeds from there. Actions, though, aren’t caused by beliefs and desires (or their realizers), in which case they do not nomically supervene on beliefs and desires (even in part). So you are not entitled to your (N2)-supervenience premise.

I mention this objection mainly for completeness. If the objection is just about what causes the action, we can easily replace beliefs and desires (or their realizers) with whatever states and events one thinks did cause the action. The objection that actions are not caused by states and events at all — either because actions are uncaused or because they are caused by a substance, the agent — can still be dealt with. The contention that actions are agent-caused is clearly incompatible with naturalism, so we can set it aside. If actions instead are uncaused, the situation is trickier. How we respond to an anti-causalist will depend on what the charge that actions are uncaused amounts to. Naturalism will still entail that r supervenes on r1, …, r4, and anyone who holds that each of these has a cause but r itself does not has no resources for escaping the argument: none of the argument’s premises rely on r’s being caused by anything, but only on there being a causal chain leading up to r’s supervenience base.

The anti-causalist likewise receives no relief by insisting that some of r1, …, r4 have causes while others do not; the elements of r’s supervenience base that are uncaused can be treated just like those that are undetermined, and the argument goes through. It seems highly implausible that every element of r’s supervenience base is uncaused. Suppose, however, that this were the case. Then, in order to avoid “NR,” the

objector must hold that “N(R1 & R2 & R3 & R4)” is false. Given the invalidity of (Agg),

we cannot say that the objector is committed to there being a particular “Ri” such that

“~NRi” (1 ≤ i ≤ 4) is true. We can say, in light of Finch and Warfield’s comments, that there is something S could have done which, if he had done it, then it would have ensured

the falsity of at least one of the “Ri”s (although it may be undetermined which one). The only available action for this falsifying role appears to be r, which means this objection

59 relies on the claim that S could have avoided r-ing. This claim is dealt with in a later objection, so I will postpone discussion of it until then.

(Objection 3) Pick some action, a, which occurred before r. Then a is clearly something an agent could have done, since it is something an agent did. Since in and x are undetermined, either of IN or X might have been false. Thus a is an action that might have falsified X and that might have falsified IN. So, you are not entitled to the second premise of the “tricky” version of your argument, because there is an action, a, which the agent could have done and which might have falsified either of the two conjuncts of which that premise is a conjunction.

There is an ambiguity in the expression “event e falsified φ.” It may mean that φ is false somehow in virtue of e, that e somehow explains φ’s falsity. On this interpretation, “e might have falsified φ” would mean that φ might have been false in virtue of e’s occurrence. This is what I had in mind when I claimed that there is nothing the agent could have done that would have falsified either IN or X. Since, if either in or x had not occurred, it would not have been in any way because of the occurrence of a, this claim is correct. A weaker interpretation of “e falsified φ” understands it as nothing more than the counterfactual, “if e had occurred, φ would have been false.” On this interpretation, “e might have falsified “φ” would presumably mean “if e had occurred, φ might have been false.” This reading is less natural than one might initially suspect; on it, for instance, all necessary falsehoods turn out to be falsified by every event that ever occurs, (cf. Schnieder, 2004, pp. 416-417). On the weaker interpretation, though, “a might have falsified ‘IN’” and “a might have falsified ‘X’” both turn out true. Does it follow that my defense of the second premise of the “tricky” argument, “N(DB & C & IN & L & X),” is flawed? No. My defense was that there is no action which one could perform that might falsify “DB & C & IN & L” and which might falsify “X” and which, by virtue of this fact, would falsify “DB & C & IN & L & X.” The fact that one could act in a way that might

60 falsify φ and might falsify ψ is not sufficient to demonstrate that this act would falsify φ & ψ. Suppose Herbert rolls a die; his die-rolling might falsify “the die does not land one” and it might falsify “the die does not land six,” but it is simply wrong to say it would falsify “the die does not land one and the die does not land six.” The defense of the tricky supervenience premise is more like this die-rolling case than McKay and Johnson’s coin-tossing case. The easiest way to see this is by noting that our agent did a, but “DB & C & IN & L & X” wasn’t falsified. Thus it simply cannot be right to say that had a occurred, “DB & C & IN & L & X” would have been false. The objection has not located an action that undermines the premise.

(Objection 4) In all discussions of r’s supervenience base, you supposed that no event subvenient on r occurs after r. If a microphysical event s both subvened on and postdated r, though, then you would have no resources for blocking “NS” (where “S” expresses the occurrence of s) and thus could not use (β□) to conclude “NR.”

An objector could complain that backwards causation undermines the argument: perhaps someone has a choice about “IN” because they can do something after r which, if they did it, would cause in to not have occured. This complaint needs little attention, though: in addition to undermining the Consequence Argument (since backwards causation makes it much more plausible that “~NP”), backwards causation simply seems, to most minds, too farfetched to serve in the defense of free will. (Even if it is possible, nobody wants to say that free will requires backwards causation.) On the other hand, backwards supervenience does not suffer from this prima facie implausibility. Some actions do seem to supervene on future events: if Jan shoots and kills Ron, then Jan’s killing of Ron is arguably an action that is finished once her finger finishes pulling the trigger but that supervenes on the (future) event of Ron’s death. This response is only available to someone who holds the fine-grained theory of action-individuation. On a coarse-grained account, Jan’s single action can be described in a number of ways: as a moving of her finger, as a shooting of a gun, or as a killing of

61 Ron. If Ron had not died from the wound, then her action would not be accurately describable as a killing of Ron, but it still would have occurred. Thus, Jan’s action does not supervene on a future event, although the fact that it can be described in a certain way does. On a fine-grained account, Jan performed a number of distinct actions, including a moving of her finger, a shooting of a gun, and a killing of Ron. The last of these is plausibly thought of as supervening on Ron’s death — it is not possible that Jan shoot Ron, Ron die the death he in fact died (the one involving the bullet that came out of the gun Jan fired), and Jan not kill Ron. Call an event e immediate if and only if e does not supervene on any future events. We can use the Supervenience Argument to show that no events, immediate or non-immediate, are choosy. I claim that (a) some non-immediate events are choosy only if some immediate event is choosy and that (b) no immediate events are choosy. Since every event is either immediate or non-immediate, (a) and (b) together entail the non- existence of choosy events. Here is the argument for (a). Let f be some non-immediate choosy event that occurs at a time t1. Since f occurs at t1, it is reasonable to assume that it supervenes on at least some immediate event that occurs at t1. Call that event e1. Since f is non-immediate, it must also supervene on some future event e2, which we shall say occurs at t2. If e2 is in turn non-immediate, we can then consider synchronous and future events it supervenes on. Some of these future events may in turn be immediate or non-immediate, but if we are to avoid a regress, we will eventually come to some collection of immediate events in the future of a that form the post-t1 portion of a supervenience base for f. (We can’t have non-immediate events all the way down.) Since the number of future immediate events will not matter for the argument I am about to give, suppose that f supervenes on three immediate events: e1, which occurs at t1, e2, which occurs at t2, and e3, which occurs at t3.

If “E1,” “E2,” “E3,” and “F” express the occurrence of e1, e2, e3, and f, 19 respectively, then □((E1 & E2 & E3) → F). Since F is choosy, ~NF. Thus, by (β□),

~N(E1 & E2 & E3). Since “E1 & E2 & E3” is true, there must be something someone could have done that would have ensured its falsity. If one of e1, e2, or e3 is choosy, then we

62 have found an immediate choosy event and our conclusion is reached. So suppose that none of them are choosy. In this case, there is something that someone could have done

(but did not do) that might ensure the falsity of some conjunct of “E1 & E2 & E3” and would, as a result, ensure the falsity of “E1 & E2 & E3.”

By the broad past principle, nothing that anyone could have done after t2 even might have ensured the falsity of “E1” or “E2.” Thus, if e3 is not choosy, there must have been something someone could have done that would have ensured the falsity of “E1 &

E2 & E3” sometime before t2. Let t be the time at which this act could have been done. As I have already argued, if there is something someone could have done but did not do, then there is some (actual) choosy act that occurred no later than the merely possible act could have occurred. Call this act a, and suppose it occurred at t*. Since t* is no later than t, it is sometime earlier than t2. Since a is choosy, if it is immediate, then the claim we are trying to prove is true. If a is not immediate, we repeat the above argument replacing a and its supervenience base for f and its. If we never come to some immediate choosy act, then we face a regress, which can take either of two forms: a backwards cascade of choosy actions (see objection 1) or an infinite number of choosy actions which all occur after some particular time and before t2. Either option is, I take it, unacceptable.

The argument for (b) is slightly simpler. Consider the first element — ri — of the class of immediate choosy events. Note that, in the Supervenience Argument, in order to show that NR we only needed to show that true conjunctions of the laws of nature and propositions expressing the occurrence of microphysical events which causally contributed to r’s supervenience base (or, as with x, were in this superveniene base) were such that nobody could have ensured their falsity. But, while it may be plausible that some events are non-immediate, it is quite implausible (at least given naturalism) to hold that there is any event r that does not allow for some exhaustive list of immediate microphysical events causally contributing to its occurrence. Thus we shall be able to find the needed premises for an instance of the Supervenience Argument to show that ri is not choosy. In this case, there are no immediate choosy events, and thus (by claim (a)) there are also no non-immediate choosy events.

63

(Objection 5) Although a from objection 3 does not undermine the tricky supervenience premise, there is an action which does. If S had not r-ed, something in r’s supervenience base would not have occurred. Since the first premise states, in essence, that nobody had a choice about anything in r’s supervenience base, if S had not r-ed, that would have falsified “DB & C & IN & L & X.” Thus, S had a choice about “DB & C & IN & L & X.”

It is true that, had S not r-ed, “DB & C & IN & L & X” would have been false. In order for the objection to succeed, though, it must be true that S could have avoided r-ing. On the face of it, this sort of objection begs the question. I have offered an argument with the conclusion “NR”; the objection rejects one of the argument’s premises on the basis of the claim “~NR.” On the other hand, some may insist that I beg the question if I do not allow my opponents this claim, for I then (they may say) unfairly shield my premises from any objections that do not presuppose my argument’s conclusion. I do not wish to embroil myself in sticky issues about question-begging and burdens of proof, so I will search for a response elsewhere. Suppose first that “~NR” is true. Then, goes the objection, “N(DB & C & IN & L & X)” is false. Why is this? Because S is able to not-r, and if S had not r-ed, not-r would have falsified “DB & C & IN & L & X.” It is part of the defense of the first premise of the Consequence Argument that people cannot now do anything that would, or even might, falsify propositions entirely about the past (Finch & Warfield, 1996, p. 523). Thus, libertarians cannot say that if S had not-r-ed, that would have rendered false any propositions entirely in the past of r — including “DB & C & IN & L.” What is left? It must be the case that S’s not r-ing falsifies “X.” We now have two options. Either S is able to falsify “R” because S is able to falsify “X,” or S is able to falsify “X” because S is able to falsify “R.” I take it that “because” is anti-symmetrical: if φ because ψ, then it had better not be the case that ψ because φ. Thus, if S’s ability to falsify “X” depends upon S’s ability to falsify “R,” we

64 may reasonably then ask why we should believe that S is able to falsify “R,” and vice versa. An objector who claims that S is able to falsify “R” because she is able to falsify “X” has the tricky task of explaining how we are able to prevent the occurrence of microphysical undetermined events. The most natural response is to say that S does this by performing some action that precludes x, but no action is available other than r. Clearly r cannot be invoked without circularity. The objector is forced into the uncomfortable position of claiming agents can directly influence the outcomes of microphysical undetermined events — and, to make the objection relevant, that agents can do this without violating the naturalistic hypothesis. As O’Connor writes, “it is simply an illusion that I (a macrophysical object) am freely and directly controlling the course that my process of deliberation takes [if] the direct action all takes place ‘down below’” (2000, p. 109). This line begins to look more like a reductio than a response. My interlocutor may grasp the other horn of the dilemma and claim that ~NX because ~NR (and thus ~N(DB & C & IN & L & X)). Thus S can keep x from occurring by not r-ing. This objection looks problematic as well. It seems to imply that x depends for its occurrence on the occurrence of r, which gets things the wrong way around, at least by naturalism’s lights. The driving idea behind naturalism is that the microphysical (conjoined, perhaps, with the laws of nature) should provide an ontological basis for the rest of reality (or, at least, physical reality). On the proposed reply, however, the microphysical depends upon the macrophysical. Notice that the proposal we were considering at the end of objection 2 is in the same boat. The naturalistic libertarian needs to say that the occurrence or non-occurrence of some (or every) microphysical event in r’s supervenience base somehow depends on the occurrence or non-occurrence of r. If I understood supervenience to be a dependence relation it would be a contradiction to say that microphysical events depend on things that supervene on them. Since I understand supervenience to be a purely logical relation, the claim is not contradictory. It is, however, out of sync with the spirit of naturalism, if not the letter. It requires agents to have powers — powers to avoid r-ing and thus keep x from occurring — that they do not get by virtue of the way the microphysical world is. Thus,

65 even if this response can be made consistent with the official declaration of weak naturalism at the beginning of §3.2, it still leaves the libertarian with the dialectical worries outlined at the end of §3.3. Furthermore, the objector has yet to explain why we should think that ~NR in the first place. Although I do not accuse anyone who objects by appealing to “~NR” of begging the question, since I have offered an argument with the conclusion “NR” I think the least my objector can do is tell some sort of story about why we should entertain “~NR.” The burden is now on the shoulders of naturalistic libertarians to give an account of the ability to do otherwise in fully microphysical terms (or in terms that are uncontroversially supervenient on microphysical terms) without relying on an ability to control the undetermined microphysical elements of the relevant action’s supervenience base (or facts that supervene on this ability). Furthermore, the account had better not be one that would license these ability ascriptions in a deterministic universe, or it will be inconsistent with the Consequence Argument. In the absence of such a story, we have no reason to accept this objection and every reason to think that libertarian free will is incompatible with naturalism.

Notes to Chapter 3

1 “φ entails ψ” here means nothing more than “□(φ → ψ)”; see p. 1.

2 In this and following quotes, I have taken the liberty of replacing the propositional variables “p” and “q” with “φ” and “ψ” in order to avoid confusion later in the chapter.

3 Here is the proof:

(1) Nφ Premise (2) Nψ Premise (3) □(φ → (ψ → (φ & ψ)) Logical truth (4) N(φ → (ψ → (φ & ψ))) α: 3 (5) N(ψ → (φ & ψ)) β: 1, 4 (6) N(φ & ψ) β: 2, 5

(McKay & Johnson, 1996, p. 115).

66

4 In fact, a similar counterexample can be used to directly show the invalidity of (β). When an agent doesn’t flip a coin but could have, both of “N(the coin does not land heads)” and “N(the coin does not land heads → the coin is not flipped)” are true — the only way the agent could falsify the relevant conditional is by ensuring that the coin is flipped and lands tails. But “N(the coin is not flipped)” is clearly false. (Carlson, 2000, pp. 283-234; Crisp & Warfield, 2000, 178-179. See Widerker 1987 for independent counterexamples to (β), and O’Connor, 1993, p. 209 and McKay & Johnson 1996, pp. 117-118 for criticisms of Widerker’s example.)

5 Understanding supervenience in this weak way also makes the formulation of (N1) and (N2) much simpler. Depending on what dependence and constitution are, no event will depend upon or constitute itself, so (for instance) (N1) would be trivially false since microphysical events would not supervene on microphysical events.

6 I leave it open how large a given macro-event’s supervenience base must be, but I take it that any sensible naturalist will want its microphysical supervenience base to be smaller than the totality of micro-events in a world (Stalnaker, 1996, pp. 228-230).

7 One could posit a form of agent-causation in which there is an irreducible relation of agent-causation which supervenes on an event-causal microphysical base. (Any agent- caused event would in this case be, in some sense or another, overdetermined.) Although (N2) does not rule out this sort of agent-causal account, I do not intend the Supervenience Argument to be effective against it. In order to accommodate this potential position, (N2) should probably be augmented with the claim “and all causation is event-causation.” For stylistic purposes I relegate this minor modification to this footnote.

8 According to J. A. Cover and John O’Leary-Hawthorne (1996), agent-causal theories that subscribe to (something like) (N1) but not (N2) are coherent but implausible. I remain neutral about that claim here.

9 The set of choosy actions will be a subset of the set of choosy events. Depending on how certain questions in action theory are settled, there may be a first choosy event that precedes the first choosy action. For instance, “beginning to choose to r” may fail to be an action but still be choosy if r is the first choosy action. In this case, though — as in other cases where the first choosy event is not an action — the causes of r will be the same as the causes of the first choosy event, and it will be a trivial matter to modify the arguments below to show that the first (alleged) choosy event is not choosy after all.

10 They will, for instance, all be on the same “action tree.” Very roughly, an action tree contains all actions performed by a single agent that are related to each other via the “by”-relation in the right sort of way. For instance, if Joe pleases Mary at 10:45 by

67

purchasing her some flowers at 9:00, then the actions described by “Joe pleases Mary at 10:45,” “Joe purchases some flowers at 9:00,” and “Joe purchases some flowers at 9:00 while thinking about Mary” are all on the same action-tree (see Ginet, 1990, pp. 19, 46- 52).

11 Likewise, for those worried that general relativity will throw a spanner into the works, we can arbitrarily pick a frame of reference for our ordering to operate within without violating either of the needed conditions.

12 I do not say that, in general, every element in an event’s supervenience base must be synchronic with the supervenient event itself. However, if x occurs before r, then it will still be in the “broad past” of r and offer no more help to the naturalistic libertarian than in did. The suggestion that r supervenes on future events will be dealt with below.

13 This is different than the claim that there is nothing one could have done which might have falsified the past, which is in general false (since one might have falsified it back before it was the past) but in this case true by virtue of the fact that r is the first choosy act.

14 Since any stronger version of the naturalistic thesis entails weak naturalism, I purposely leave the claim ambiguous.

15 Framing the Mind arguments (both this one and Nelkin’s revised version below) with reference to the first choosy act is a liberty I have taken, but it is a liberty that strengthens the arguments.

16 Erik Carlson (2002) offers his own reformulation of the Mind argument, which does not use (β□). It argues from “NS,tDB” and “NS,t(DB → R),” where “NS,tφ” means “φ and agent S has at time t no choice about whether φ” (where “having a choice” is understood in the way suggested by Finch and Warfield (1998, p. 516)), to “NS,tR,” via the principle

** (β ) From (i) “NS, tφ,” (ii) “NS, t(φ → ψ),’” and (iii) the fact that, for any way in which S is able at t to act so that ψ would be false, either: if S were to act in that way, φ would be false, or if S were to act in that way, φ would be true, deduce “NS, tψ” (cf. Carlson, 2002, p. 397).

For the argument, t is “the last time at which it is possible for S to exercise her (alleged) ability to choose whether or not [r] shall occur” (Carlson, 2002, n. 11). It is reasonable to think that some of r’s causes are still occurring after others have run their course. Suppose that d and b are two of r’s causes, where d occurs at t and b occurs before t. (It is not implausible to think that some of r’s causes will occur at the last moment at which S could have exercised her (alleged) ability to choose whether or

68

not r would occur.) Suppose further that b causes r via a causal chain which includes e, which occurs at t. Suppose finally that both b and e are undetermined. Then there may be some worlds in which r does not occur because b doesn’t occur (and hence in which “DB” is false) and some other worlds in which r does not occur, b does occur, and e does not occur (in which “DB” is true). Condition (iii) of (β**) would thus not be met for the relevant propositions, so the argument fails.

17 Ekstrom prefers a version of the Consequence Argument that uses the operator “BS,t,” where “BS,tφ” means “φ and S is not able at t to prevent its being the case that φ” (2000, pp. 28-29). It is easy enough to deploy Ekstrom’s preferred formalism in the Supervenience Argument, though. Note first that she subscribes to the following principles:

(T) From “BS,tφ” and “BS,t(φ → ψ),” deduce “BS,tψ,” (A) From “□φ,” deduce “BS,tφ,” and (C) From “BS,tφ” and “BS,tψ,” deduce “BS,t(φ & ψ).”

(According to Ekstrom (2000, p. 41), (C) is not invalidated by McKay and Johnson’s counterexample. While I do not find her response convincing, I leave it aside for present purposes. If (C) — and therefore (T) — is rejected, (D) below may be a plausible fallback principle.) Suppose r occurs at t. Then clearly, for all S and t, BS,t(DB & C & IN & L). And, it seems, BS,tX for reasons canvassed above. (How could one prevent an undetermined microphysical event from occurring?) Also, note that (A) and (T) together license

(D) From “BS,tφ” and “□(φ → ψ),” deduce “BS,tψ.”

So, by her lights, the following argument is valid:

(1) □((DB & C & IN & L & X) → (L & R1 & R2 & R3 & X)) Premise (2) BS,t(DB & C & IN & L) Premise (3) BS,tX Premise (4) BS,t(DB & C & IN & L & X) C: 2, 3 (5) BS,t(L & R1 & R2 & R3 & X) D: 1, 4 (6) □((L & R1 & R2 & R3 & X) → R) Premise (7) BS,tR. D: 5, 6

18 I owe this point to Joe Campbell.

19 Perhaps we would be better off saying that □((L & E1 & E2 & E3) → F). For simplicity I suppress the reference to the laws of nature; the argument is not affected by the modifications required by adding “L.”

69

CHAPTER 4 PUTTING IT ALL TOGETHER: IN DEFENSE OF COMPATIBILISM

4.1. Limitations of an Offensive Defense In Chapter 1 I promised that I would defend compatibilism by the offensive strategy of arguing against rival positions. In Chapters 2 and 3 I made good on that promise. Chapter 2 undercut the Bottom-Up Argument for semantic eliminativism with the use-plus-eligibility view of language and argued that metaphysical eliminativism faced difficulties raised by that view of language plus the truth of intuitional anarchy. Chapter 3 used the Supervenience Argument to discredit naturalistic libertarianism (at least insofar as that view is committed to the Consequence Argument) and suggested that this had negative dialectical implications for libertarianism in general. This kind of defense, which proceeds by attacking its enemies, has its own limitations, though. Notice that, although both Chapters 2 and 3 made strong cases against some of compatibilist’s rivals, neither is much of a positive argument for compatibilism. Consider Chapter 2. Given everything said there, it may be that in some contexts “free” refers to a property incompatible with determinism and “free will is compatible with determinism” is false. This is hardly the thesis of compatibilism. A friend of compatibilism may reply on my behalf that it is also open that in some contexts “free” refers to a property compatible with determinism and “free will is compatible with determinism” is true. Thus, at least some instances of “S acted freely” will be true even if determinism is true. Since compatibilism can be defined as the view that, if determinism were true, not all of our free will ascriptions would be false, compatibilism is vindicated. This friendly reply can give some consolation to compatibilists — but not much. Recall from Chapter 1 that one of the main jobs of free will is the grounding of moral

70 responsibility. In §2.3, I discussed the possibility that the different properties variously meant by “free” may each ground different components of “full” moral responsibility, in which case the entire package would only be secured by actions that exemplified all of these properties. If the truth of determinism is incompatible with the exemplification of some of these properties, then one could have “full” moral responsibility only if determinism is false. Compatibilism is, very roughly, the view that all of the kinds of freedom “worth wanting” (in the language of Dennett, 1984; cf. Kane, 1996, p. 15) are compatible with determinism. It is not quite enough that not all of our ascriptions of freedom turn out false if determinism is true; compatibilism must be the view that not all of our ascriptions of a kind of freedom worth wanting turn out false if determinism is true. A kind of freedom that grounds “full” moral responsibility, however, seems worth wanting — and, for all that was said in Chapter 2, may be incompatible with determinism.1 Chapter 3 likewise appears to offer little direct support for compatibilism. Suppose, for the sake of argument, that incompatibilism is true if and only if the Consequence Argument is valid. I argued, in essence, that the Consequence Argument is valid if and only if the Supervenience Argument is. If this is right, then following triad is inconsistent:

(1) People have free will. (2) Naturalism is true. (3) Free will is incompatible with determinism.

This is hardly an argument for compatibilism, though. An inconsistent triad can be escaped by the rejection of any of its members. In order to have an argument for compatibilism, I would need to supplement the Supervenience Argument with a further argument that, given the inconsistency of the above trio, (3) is the claim that ought to be rejected. Both eliminativists (who reject (1)) and non-naturalistic libertarians (who reject (2)) will not find a compelling reason to become compatibilists when faced with the Supervenience Argument.

71 Some may think, perhaps partially on the basis of my remarks in §3.3, that I can turn the Supervenience Argument into an argument for compatibilism by pointing out that (3) is the most easily rejected claim on that list. If the rejection of incompatibilism is more plausible than the rejection of either the existence of free will or the rejection of naturalism, then this fact, coupled with the inconsistency of (1)-(3), comprises an argument for compatibilism. I find this line of reasoning compelling. (1)-(3) are listed, in my opinion, in order of descending plausibility: I would rather give up (3) than (2) and I would rather give up (2) than (1). Insofar as an audience agnostic about the compatibility of free will and determinism (see van Inwagen, 1992, p. 58; §3.3) agree with me in how they order the plausibility of (1)-(3), I have an argument for compatibilism. It is not clear, however, how far this agreement goes. Among a non-agnostic audience there is genuine disagreement. The libertarian Peter van Inwagen, for instance, essentially says that he would find the falsity of (3) “a very great mystery indeed, a mystery much greater than that which would attend [our having] properties that do not supervene upon the properties of the atoms that we consist of” (1983, p. 217). Trenton Merricks (2001, pp. 155-161) uses his version of the Supervenience Argument as an argument for the falsity of (2) (which implies that he finds the case for (1) and (3) more compelling), and Peter Unger concludes that Supervenience-Argument-like considerations demonstrate the need for “attempts to develop metaphysical alternatives to the Scientiphical [i.e. naturalistic] Metaphysic… more conducive to our having full choice” (2002, p. 25). I find naturalism more plausible than incompatibilism. For van Inwagen, this ranking is reversed. He suggests that, in our disagreement about whether (2) or (3) is more plausible, we have “reached bedrock. We have nothing more to say to each other; or, at any rate, though we may call each other names we have no more arguments” (1983, p. 217). I agree that we will not be able to settle, by argument, whether the falsity of naturalism or the truth of compatibilism is a greater mystery. There is, however, another argument, which draws upon considerations of both Chapters 2 and 3, that may be able to tip the scales in favor of the truth of compatibilism without settling the issue over

72 whether compatibilism or non-naturalism is more plausible. I call that argument the Use- Plus-Eligibility Argument for compatibilsm and sketch it in the following section.

4.2. Introducing the Use-Plus-Eligibility Argument for Compatibilism Here is the Use-Plus-Eligibility Argument:

(1) Semantic content is determined by use plus eligibility. (2) The semantic value of “free” has multiple candidate meanings all of which fit use (roughly) equally well. (3) These candidate meanings can be divided into those which are compatible with determinism (the C’s) and those which are not (the I’s). (4) The I’s are exemplified only if some acts are agent-caused. (5) Agent causation is impossible. (6) Therefore, the I’s are impossibly exemplified. (7) Hence, the I’s are less eligible to be meant than (at least some of) the C’s. (8) So, any properties that serve as the semantic value of “free” are compatible with determinism.

If the Use-Plus-Eligibility Argument is correct, then free will is compatible with determinism even in the sense that grounds “full” moral responsibility. There are no properties that ever serve as the semantic value of “free” that are incompatible with determinism, so there is no reason to suppose that an action cannot exemplify all these properties in a deterministic world and therefore reason to think that a “full” moral responsibility package can be grounded even if determinism is true. The first premise is a consequence of the use-plus-eligibility view of language. If we, along with David Lewis (1983, p. 371, 1984, p. 226), take Putnam’s (1980) model- theoretic paradox to be an argument for the use-plus-eligibility view of language (see §2.2.1), we have ample reason to accept (1). The second premise is roughly a consequence of intuitional anarchy. Of course, depending on what counts as use, it may take more than mere intuitional anarchy to ensure that multiple candidate meanings for

73 “free” fit it equally well. A brief consideration of other possible sources of use- determination does not appear to enhance the plausibility of there being a single candidate meaning that fits use best, for, inasmuch as intuitions seem to conflict, so, it seems, do folk theories and practices. I take it that (3) is relatively uncontroversial. Grant (4) and (5) for the moment; we will return to them in the next section. If they are true, then clearly so is (6). If impossible properties are extremely ineligible to be meant (§2.4) and if any of the C’s are possibly satisfied, then these C’s are more eligible to be meant by “free” than any of the I’s, in which case (7) is true. The supposition that some of the C’s are possibly satisfied appears quite reasonable. I am aware of many critiques of compatibilist theories of free will, but to my knowledge no one has ever complained that the sort of conditions compatibilists say are sufficient for free will are not possibly satisfied (the complaint is generally instead that those conditions really aren’t sufficient for free will). Finally, as discussed in §2.2.2, (8) follows from (7) and (1). If (4) and (5) can be adequately defended, the Use-Plus-Eligibility is a strong positive argument for compatibilism. Notice, however, that the argument is not an argument for any specific compatibilist theory. It is consistent with the argument that there is no fact of the matter as to what semantic value “free” has: if there are multiple candidate meanings, all compatible with determinism, that fit both use and eligibility equally well (and better than any competitors), then the content of “free” is indeterminate among them. In this eventuality, however, “free will is compatible with determinism” would be true in every context since there is no semantic value for “free” that is incompatible with determinism, although “an act is free if and only if it satisfies analysis A” may be true in some contexts and false in others. Thus, the Use-Plus-Eligibility Argument supports compatibilism without being wedded to any particular compatibilist analysis. Given the difficulty in producing such analyses, the argument’s independence of them may be a point in its favor (cf. Lycan, 2003, pp. 112-114).

74 4.3. Agent-Causation, (β□), and the Use-Plus-Eligibility Argument Now we must consider premises (4) and (5). At this point, what began as a full- out defense of the Use-Plus-Eligibility Argument becomes a mere sketch of a defense. Premise (4) is controversial, and a full defense of it is the work of an independent thesis; rather than provide what I take to be conclusive arguments for its truth, I offer instead in this section an outline of the way I think its proponents should argue in its support. I leave the filling-in of much needed detail as work for a future project. Premise (5) I do not defend at all: it would take the work of a book to show the impossibility of agent-causation. However, I do not take it that my lack of a defense of (5) renders the Use-Plus-Eligibility Argument uninteresting or dialectically useless. Given my failure to argue for (5), I may present the Use-Plus-Eligibility Argument as a conditional argument for the conclusion that, if agent causation is impossible, compatibilism is true. This would be a significant result on its own merits. First, even those incompatibilists who think that agent-causation is possible would not want to admit that its possibility is a necessary condition for the truth of incompatibilism — presumably, they would not change their minds about incompatibilism if they were convinced that agent-causation were impossible after all. Further, since many incompatibilists may be inclined to agree that agent-causation is impossible, they will want to look to undermine the Use-Plus-Eligibility Argument somewhere other than premise (5). Of course, if naturalism, as I understand it in this thesis, is a necessary truth, then agent-causation is impossible; naturalism as defined in §3.2 is meant to rule out causal processes other than the event-causal kinds postulated by science. I do not recommend supporting the Use-Plus-Eligibility Argument in this manner, though. First, even if naturalism is true, it may only be a contingent truth. Many philosophers who find it highly plausible that all events and causal relations supervene on the microphysical may balk at the suggestion that this is a matter of necessity. A lot of very strange things are possible, and there is no reason to think that non-naturalism is not among them. Second, insofar as a proponent of the Use-Plus-Eligibility Argument employs it to escape bedrock disagreement with those like van Inwagen who find the falsity of incompatibilism a

75 greater mystery than the falsity of naturalism, she will not want to run the risk of re- opening this dispute by employing naturalism a second time around. However, there are many philosophers — compatibilists (e.g. Bok, 1998, pp. 44- 46; Watson, 1982, p. 10), libertarians (e.g. Ekstrom, 2000, pp. 98-99; Kane, 1996, p. 121; Ginet, 1991, p. 13-14, 1997), and eliminativists (e.g. Broad, 1952; Smilansky, 2000, pp. 62-68; G. Strawson, 1986, pp. 52-55, 2001, pp. 443-444) alike — who find agent causation intrinsically problematic independent of its non-naturalistic claims. Roderick Chisholm’s (1964) description of agent causation as the thesis that “each of us, when we act, is a prime mover unmoved” (p. 134) provides, in the minds of those who find the notion of a “prime mover unmoved” unintelligible, sufficient reason to accept (5) (cf. G. Strawson, 2001, pp. 443-444). Even those who do not find agent-causation impossible express reservations. Van Inwagen writes,

I find the concept of immanent or agent causation puzzling, as I suspect most of my readers do (those who don’t find it downright incoherent). In fact, I find it more puzzling than the problem it is supposed to be a solution to (1983, p. 151).

The agent-causalist Richard Taylor concedes,

One cannot hardly affirm such a theory of agency without complete comfort, however, and wholly without embarrassment, for the conception of men and their powers which is involved in it is strange indeed, if not positively mysterious (1963, p. 52), and even Randolph Clarke, the author of a recent defense of agent-causation, admits that there are considerations “which incline the balance against the possibility of substance causation in general and agent causation in particular” (2003, p. 209). The puzzlement of these philosophers is not due to the mere non-naturalism entailed by agent causation. We have already seen that van Inwagen is comfortable with rejecting naturalism, but here he seems to suggest that the truth of agent causation may be even more puzzling by his

76 lights than the falsity of incompatibilism, and as agent-causation precludes naturalism, the dialectical worries of Clarke and Taylor likely stem from some other source. And one need not believe that all events and causal relations supervene on the microphysical in order to find the notion of a “prime mover unmoved” unintelligible. A common, and perhaps more precise, objection to agent-causation complains that the notion of a substance irreducibly causing an event is incoherent. Since agent- causation essentially involves the claim that a substance (the agent) causes an event (the action), where this is not reducible to any sort of event-causation, it is incoherent and thus impossible. An objection of this sort was first leveled by C. D. Broad (1952), who holds that only events can cause other events. Here is the basic objection. According to agent- causation, agents as causes determine their effects. Suppose that, in a world W, an agent S causes an event e at a time t. It is possible that the agent not cause e at t; there is thus a world W* which is identical with W up until t in which e does not occur.2 According to the agent-causalist, the agent determined the occurrence of e. Yet it was not any property of the agent that determined the occurrence of e, for S has the same properties in W and W*. It also was not the existence of the agent that determined the occurrence of e, for S existed in both W and W*. Worlds W and W* clearly show that e is undetermined; yet by the agent-causal hypothesis e was determined, but by the agent. Agent-causation is thus embroiled in contradictions. Broad’s objection comes in different forms. One version (which seems to be the version Broad himself preferred) complains that there is no reason why e occurred when it did rather than at some other time. Another version objects that there is no explanation of why S caused e for the reason S caused it rather than another reason. On any account, the objection is the same: agents, as substances and without reference to their properties (which reference would reduce agent-causation to event-causation), do not provide a rich enough subject matter to explain all of the things that a cause of an event should be able to explain. A full defense of the Use-Plus-Eligibility Argument would argue for the impossibility of agent-causation using something like Broad’s objection, shoring up the objection against recent attempts by agent-causalists to circumvent it (Clarke, 2003, pp.

77 197-210; O’Connor, 2000, pp. 74-76). If compatibilists can show that, independent of naturalistic assumptions, agent-causation is untenable, then if they can also show premise (4) — that properties like I are exemplified only if some acts are agent-caused — they will be able to conclude that, if a property is favored by incompatibilist intuitions as the semantic value of “free,” it is impossible. We can give an argument for (4) if we can establish the following two claims: (a) if M is a candidate meaning incompatible with determinism, then (i) the exemplification of M entails the existence of some choosy acts and (ii) the incompatibility of M and determinism can be established by the Consequence Argument; and (b) if M is a candidate meaning and M is incompatible with both determinism and naturalism, then M is a property that entails that some acts are agent-caused. By (a) and the Supervenience Argument, any I is incompatible with naturalism, and so by (b), it requires the truth of agent-causation. Thus if (a) and (b) are both correct, then the exemplification of any of the I’s entails that some act is agent-caused. I have no argument for (b). Note, though, that the incompatibilist theories of free will discussed in the literature, as a matter of course, divide into the naturalistic event- causal, the agent-causal, and the non-causal. The latter two are the only extant non- naturalistic, non-agent-causal theories of free will for the incompatibilist to choose from; other positions in logical space have remained unoccupied. I suspect that there is a good reason for their vacancy: only these three kinds of theories come close to capturing what libertarians want out of free will. While I do not argue for this claim, I do suggest that someone wishing to flesh out my sketch of a defense into a full-out argument for compatibilism would do well to pursue this line of thought. Once this claim is established, the proponent of (b) would need to argue further that non-causal accounts of free will fail to capture what libertarians want out of free will (in the manner of e.g. O’Connor, 2000, pp. 24-27).3 Thus, only naturalistic and agent-causal properties are left as viable candidate meanings for “free.” Support for (a) comes from the belief that the Consequence Argument captures what lies at the heart of all uses of the term “free” that are incompatible with determinism (cf. Horgan, 1985; Slote, 1982; van Inwagen, 1989). Van Inwagen writes,

78

in my view, one could have no reason for being an incompatibilist if one did not accept Beta. If one accepts Beta, one should be an incompatibilist, and if one is an incompatibilist, one should accept Beta (1989, p. 405).

Presumably, in the wake of (β)’s invalidity (see §3.1), van Inwagen would need to revise his claim to be about (β□). But linking incompatibilism’s fate to (β □)) is dangerous. There are at least some readings of the “N”-operator, for instance, on which (β□) is provably valid.4 On these readings of “N,” traditional compatibilists should accept (β□) but reject the claim that N(P & L).5 Thus one can accept (β□) without being an incompatibilist simply by rejecting N(P & L). Furthermore, it is not even true that one ought to be an incompatibilist if one accepts the soundness of a suitably conditionalized consequence argument. Semi- compatibilists (Fischer & Ravizza, 1998) hold, roughly, that free will does not require the existence of choosy actions. However, a proponent of (4) can make use of a revised version of van Inwagen’s claim according to which one has a reason to be an incompatibilist if and only if one thinks that the Consequence Argument, suitably conditionalized, is a sound argument for incompatibilism. Then she can argue for (a) as follows. Let M be a candidate meaning for “free” incompatible with determinism. Since M is incompatible with determinism, insofar as it is taken to be a semantic value for “free,” it provides reason for being an incompatibilist. Suppose for reductio that M is compatible with (say) the non-existence of choosy acts. Then acceptance of the soundness of a conditionalized Consequence Argument would not provide any reason for being an incompatibilist — at least, with respect to M — since all the argument shows is that there are no choosy acts in a deterministic world. This fact alone does not show that M is incompatible with determinism, since M is compatible with the non-existence of choosy acts. Thus one should be an incompatibilist even if one does not accept the Consequence Argument — which contradicts our revised version of van Inwagen’s claim. In order to remain consistent with this revised claim, M must be incompatible with the non-existence of choosy acts.

79 John Martin Fischer and Mark Ravizza (1992) have challenged van Inwagen’s original claim in a way that may also have bearing on the revised version of that claim we are considering here. They contend that incompatibilsm can be argued for in a way that makes no appeal to (β□).6 Specifically, they contend that the following two claims support an argument for incompatibilism that is independent of the validity or invalidity of (β□):

(FP) For any action Y, agent S, and time T, if it is true that if S were to do Y at T, some fact about the past relative to T would not have been a fact, then S cannot do Y at T. (FL) For any action Y, and agent S, if it is true that S were to do Y, some natural law which actually obtains would not obtain, then S cannot do Y.

(pp. 427-428). Given (FP) and (FL), they claim, one can argue for incompatibilism without recourse to a (β□). I take it that, if one can accept the (FP)-(FL) argument without accepting (β□), one can have a reason to be an incompatibilist even if one does not accept the Consequence Argument. The argument from (FP) and (FL) goes to the conclusion that, in deterministic worlds, one cannot do other than one does. If being able to do other than one does entails that one’s act is choosy, then the (FP)-(FL) argument supports incompatibilism only if choosy acts are a necessary condition for free will, so a defender of the revised version of van Inwagen’s claim needs now only to show that one should accept the (FP)-(FL) argument if and only if one should accept the Consequence Argument. There are two structurally similar strategies a proponent of (a) could take at this point. The first would be to note that (FP) and (FL) are, as Fischer and Ravizza concede, “controversial though not implausible” (p. 427). If they are controversial, we may ask why anyone ought to accept them. If the only reason one could have to accept (FP) and (FL) was a commitment to (β□) and the premises of the Consequence Argument (namely, “N(P & L)),” the revised version of van Inwagen’s claim is still correct and (a) is not threatened. The second strategy argues that (FP) and (FL) may be independently

80 plausible but that they in turn entail the soundness of the conditionalized Consequence Argument. Thus one need not appeal to the Consequence Argument to justify a belief in (FP) and (FL), but one still cannot argue for incompatibilism without being committed to it. It is not implausible that one of these two strategies would succeed in the case of Fischer and Ravizza’s (FP)-(FL) argument. It is not implausible, for example, that the intuition that one cannot do anything that would require facts about the past to be different than they are stems from the intuition that one cannot now do anything that would require the falsity of facts that one cannot now falsify. Suppose that “φ” and “□(φ → ψ)” both express facts and are true,7 that “ψ” expresses the occurrence of an action, and that nobody can do or could have done anything that would ensure the falsity of “φ.” (That is, assume Nφ.) Assume that no one’s ever being able to do anything that would ensure the falsity of “φ” entails that one cannot now falsify “φ.”8 Since “φ” and “φ → ψ” are both true, “ψ” is true also. However, the falsity of “ψ” clearly requires the falsity of “φ” — there are no possible worlds in which “φ” is true but “ψ” is not — so if someone could have done something that would have ensured the falsity of “ψ,” then they could have done something that required the falsity of a fact — the truth of “φ” — that they cannot (now or ever) falsify. Thus, Nψ. Hence, “Nφ” and “□(φ → ψ)” imply “Nψ”; the intuitions underwriting (FP) entail the validity of (β□). Furthermore, if (FP) and (FL) are taken to agglomerate — that is, if the intuitions that support (FP) and (FL) also support

(FP&L) For any action Y, agent S, and time T, if it is true that if S were to do Y at T, either some fact about the past relative to T would not have been a fact or some natural law which actually obtains would not obtain, then S cannot do Y at T,

and if P expresses the state of the world at a time before there were any human agents, then if one accepts the (FP)-(FL) argument one should also accept N(P & L). If this is correct, then one should accept the (FP)-(FL) argument if and only if one accepts the Consequence Argument.

81 Even if the strategy succeeds in the case of Fischer and Ravizza’s argument, it may not succeed against other arguments for incompatibilism. The Manipulation Argument (Kane, 1996, pp. 61-74; Pereboom, 2001, pp. 110-120), in essence, contends that the intuitions that cause us to judge covertly manipulated agents (agents controlled without their knowledge by e.g. brilliant neurosurgeons or aliens) as unfree should also cause us to judge determined agents as unfree. It may be that the Manipulation Argument is not underwritten by the intuitions that support the Consequence Argument, in which case the defender of (a) had better find a way to reject the claim that it provides some reason for being an incompatibilist. (See Mele, n.d., for one response to the Manipulation Argument.) If the intuition that manipulated agents are not free does not support incompatibilism, then the fact that some people believe it does gives no additional reason to be an incompatibilist. The intuition at play in the manipulation cases will track a property, P, which may be a candidate meaning for “free” with a high fit with use which does not require the truth of (β□), N(P & L), or the existence of choosy acts, but which also does not require the falsity of determinism and therefore does not threaten (a). Someone who defends premise (4), therefore, should argue as follows: for any consideration someone proposes as a reason for being an incompatibilist, either that consideration stands or falls with the Consequence Argument or it is not a reason for being an incompatibilist at all. If this is correct, then the Supervenience Argument shows that every reason for being an incompatibilist becomes a reason for being a non- naturalist. If, in turn, every reason for being an incompatibilist and non-naturalist is a reason for believing in agent-causation (claim (b)), then (4) will follow. If (a) and (b) can be shown, then premise (4) follows, and if premises (4) and (5) can be established, the rest of the Use-Plus-Eligibility Argument follows much more easily. Although there is much left to be said in defense of each of its premises, I here leave my sketch of the argument unfinished.

4.4. Conclusion When all is said and done, a partial defense of compatibilism is not a defense of compatibilism. I have not argued against certain of compatibilists’ opponents, including

82 contingent eliminativists (e.g. Pereboom, 2001) and agent-causalists (e.g. Clarke, 2003; O’Connor, 2000), and I have not considered certain arguments for various incompatibilist positions, including the Manipulation Argument (Kane, 1996, 61-74; Pereboom, 2001, pp. 110-120), the Top-Down Argument for semantic eliminativism (Double, 1996, ch. 8- 9), and Galen Strawson’s regress-style argument for metaphysical eliminativism (n.d.; 2001, pp. 445-449). Furthermore, premises in my positive argument cry out for a fuller defense than I have given them here. On the other hand, my conclusions in this thesis are not insignificant. The failure of the Bottom-Up Argument for semantic eliminativism (Chapter 2), the inconsistency of the commitments of the Consequence Argument with naturalistic libertarianism (Chapter 3), and the potential for a novel argument for compatibilism (Chapter 4) are all significant in their own right. Narrowing the playing field, while globally inconclusive, is useful philosophical work, and compatibilists may take some comfort in diminishing the number their of opponents even while bracing for the ensuing battle with those that remain. Additionally, each argument or counter-argument has weighty, if not decisive, implications for other positions — most prominently, metaphysical eliminativism and non-naturalistic libertarianism — and compatibilists may thereby gain the advantage of wrestling with opponents crippled though not defeated. Thus compatibilism is not established, and her opponents are not all vanquished; but her prospects look brighter and her remaining battles appear fewer. Insofar as this is the goal of a partial defense — not to secure the victory, but merely to move the battle forward — this thesis has succeeded. Insofar as compatibilism is to be defended, there is much left to be done.

Notes to Chapter 4

1 This last point may be slightly overstated. “Full” moral responsibility may include kinds of moral responsibility that (arguably) aren’t worth wanting — for instance, the kind of moral responsibility that would justify God sending someone to hell for an eternity (cf. Strawson, 2001, pp. 451-452). However, it is still open that some of the kinds of moral

83

responsibility worth wanting are incompatible with determinism, so that the full package of moral responsibility worth wanting is incompatible with determinism.

2 Excepting in trivial ways, of course. W differs from W* in being a world in which e is going to occur, but this fact should have no bearing on the present case since it is clearly parasitic on the phenomenon being explained and therefore cannot enter into the proffered explanation.

3 One could also argue that non-causal accounts of free will are impossible by noting that they entail non-causal accounts of action and arguing that these accounts of action are impossible (in the manner of e.g. Mele, 2003, ch. 2). We could then replace premises (4) and (5) with the following without affecting the rest of the argument:

(4*) The I’s are exemplified only if some acts are agent-caused or uncaused. (5*) Agent-causation and uncaused acts are impossible.

4 I owe this point to Tom Crisp. Here is one reading of “N” for which (β□) is provably valid: let “Nφ” mean “φ & ~∃x[C(x) & (P(x) □→ ~φ)],” where “C(x)” means “someone can perform x”, “P(x)” means “someone does perform x,” and “φ □→ ψ” is the counterfactual, “if φ had been the case, then ψ would have been the case.” We prove (β□) by reductio:

(1) Nφ Assumption (2) □(φ → ψ) Assumption (3) ~Nψ for reductio (4) φ & ~∃x[C(x) & (P(x) □→ ~φ)] Df. N: 1 (5) φ PC: 4 (6) ψ □MP: 2, 5 (7) ~(ψ & ~∃x[C(x) & (P(x) □→ ~ψ)]) Df. N: 3 (8) ∃x[C(x) & (P(x) □→ ~ψ)] PC: 6, 7 (9) ∃x[C(x) & (P(x) □→ ~φ)] Below: 2, 7 (10) ~∃x[C(x) & & (P(x) □→ ~φ)] PC: 4 (11) Nψ reductio: 9, 10

The move from (2) and (7) to (9) is valid on a possible-worlds semantics for counterfactuals (Lewis, 1973). “P(x) □→ ~ψ” says, roughly, that in the nearest worlds where x is performed, ~ψ. But, by (2), in every world where ~ψ, ~φ in that world also. Thus, in the nearest worlds where x is performed, ~φ — i.e., P(x) □→ ~φ.

5 For instance, on the reading of “N” used in n. 4, traditional compatibilists would claim that there is something I can do such that, if I had done it, the past or laws would have been different (i.e., “~N(P & L)” is true on the above reading of “N”) but deny that this is

84

the same as being able to make the past or laws different (cf. Lewis, 1981), and they would claim further that it is only on the latter, stronger reading of “Nφ” as “φ and nobody can or ever could have done anything that would have made φ false” that “N(P & L)” is plausible.

6 Fischer and Ravizza’s (1992) discussion predates the widespread acceptance of counterexamples to (β) and a Consequence Argument modified for (β□), but I presume they take their remarks to apply to any (β)-like principle one might use to revise the original Consequence Argument — (β□) included. The argument attributed to them here has been modified accordingly.

7 I am not exactly certain what “facts” are supposed to be, but I take it that facts can be put in a one-to-one correspondence with propositions that express them and that if a proposition that expresses a fact were to be false, then the fact would not be a fact.

8 Since I read “ensure the falsity of φ” as “falsify φ,” this claim appears trivially true. However, on a more robust reading of “ensure the falsity of φ,” it may be contestable.

85

REFERENCES

Barnett, David. (2000). Is water necessarily identical to H2O? Philosophical Studies 98, 99-112.

Bishop, John. (2003). Prospects for a naturalist libertarianism: O’Connor’s Persons and Causes. Philosophy and Phenomenological Research 66, 228-243.

Bok, Hilary. (1998). Freedom and Responsibility. Princeton: Princeton University Press.

Bratman, Michael. (1996). Identification, decision, and treating as a reason. Philosophical Topics 24: 1-18.

Broad, C. D. (1952). Determinism, indeterminism, and libertarianism. In his Ethics and the History of Philosophy (pp. 195-217). London: Routledge & Kegan Paul.

Carlson, Erik. (2000). Incompatibilism and the transfer of power necessity. NOÛS 34, 227-290.

Carlson, Erik. (2003). Counterexamples to principle beta: a response to Crisp and Warfield. Philosophy and Phenomenological Research 66, 730-737.

Chisholm, Roderick M. (1964). Human freedom and the self, The Lindley Lecture, University of Kansas, 3-15. Reprinted in L. W. Ekstrom (Ed.), Agency and Responsibility (pp. 126-137). Boulder, Co.: Westview Press, 2001.

Churchland, Paul. (1981). Eliminative materialism and propositional attitudes. The Journal of Philosophy 78: 67-90.

Clarke, Randolph. (1995). Indeterminism and control. American Philosophical Quarterly 32, 125-138.

Clarke, Randolph. (2003). Libertarian Accounts of Free Will. New York: Oxford University Press.

Cover, J. A. & O’Leary-Hawthorne, John. (1996). Free agency and materialism. In J. Jordan and D. Howard-Snyder (Eds.), Faith, Freedom, and Rationality (pp. 47-71). Lanham, Md.: Rowman & Littlefield.

86

Crisp, Thomas M. & Warfield, Ted A. (2000). The irrelevance of indeterministic counterexamples to principle beta. Philosophy and Phenomenological Research 61, 173- 84.

Daw, Russell & Alter, Torin. (2001). Free acts and robot cats. Philosophical Studies 102, 345-357.

Dennett, Daniel C. (1984). Elbow Room: The Varieties of Free Will Worth Wanting. Cambridge, Mass.: MIT Press.

DeRose, Keith. (2004). Single scoreboard semantics. Philosophical Studies 119, 1-21.

Double, Richard. (1991). The Non-reality of Free Will. New York: Oxford University Press.

Double, Richard. (1996). Metaphilosophy and Free Will. New York: Oxford University Press.

Double, Richard. (2001). Metaethics, metaphilosophy, and free will subjectivism. In R. Kane (Ed.), The Oxford Handbook of Free Will (pp. 506-528). Oxford: Oxford University Press.

Ekstrom, Laura Waddell. (2000). Free Will: A Philosophical Study. Boulder, Co.: Westview Press.

Finch, Alicia & Warfield, Ted A. (1998). The Mind argument and libertarianism. Mind, 107, 515-28.

Fischer, John Martin. (1994). The Metaphysics of Free Will. Cambridge, Mass.: Blackwell.

Fischer, John Martin & Ravizza, Mark. (1992). When the will is free. In J. E. Tomberlin (Ed.), Philosophical Perspectives 6: Ethics (pp. 423-451). Atascadero, Calif.: Ridgeview Publishing Co.

Fischer, John Martin, & Ravizza, Mark. (1998). Responsibility and Control: A Theory of Moral Responsibility. New York: Cambridge University Press.

Flew, Antony. (1955). Divine omnipotence and human freedom. In A. Flew and A. McIntyre (Eds.), New Essays in Philosophical Theology (pp. 144-169). London: SCM Press.

Frankfurt, Harry. (1971). Freedom of the will and the concept of a person. Journal of Philosophy 68: 5-20.

87

Ginet, Carl. (1990). On Action. New York: Cambridge University Press.

Ginet, Carl. (1997). Freedom, responsibility, and agency. The Journal of Ethics 1, 85-98.

Goetz, Stewart C. (1988). A noncausal theory of agency. Philosophy and Phenomenological Research 49, 303-316.

Graham, George, & Horgan, Terrence. (1994). Southern fundamentalism and the end of philosophy. In E. Villanueva (Ed.) Truth and Rationality: Philosophical Issues 5 (pp. 219-247). Atascadero, Ca.: Ridgeview.

Hawthorne, John. (2001). Freedom in context. Philosophical Studies 104, 63-79.

Heller, Mark. (1996). The mad scientists meet the robot cats: compatibilism, kinds, and counterexamples. Philosophy and Phenomenological Research 61, 333-337.

Horgan, Terrence. (1985). Compatibilism and the consequence argument. Philosophical Studies 47, 339-356.

Kane, Robert. (1989). Two kinds of incompatibilism. Philosophical Studies 50, 219-254. Reprinted in T. O’Connor (Ed.), Agents, Causes, and Events (pp. 115-149). New York: Oxford University Press (1995).

Kane, Robert. (1996). The Significance of Free Will. New York: Oxford University Press.

Kane, Robert. (1999). Responsibility, luck and chance. Journal of Philosophy 96, 217- 240. Kim, Jaegwon. (1984). Concepts of supervenience. Philosophy and Phenomenological Research 48, 315-326. Reprinted in his Supervenience and Mind (pp. 53-78). Cambridge: Cambridge University Press (1993).

Kripke, Saul. (1972). Naming and Necessity. Cambridge, Mass.: Harvard University Press.

Lewis, David K. (1970). How to define theoretical terms. Journal of Philosophy 67, 427- 446.

Lewis, David K. (1973). Counterfactuals. Malden, Mass.: Blackwell.

Lewis, David K. (1979). Scorekeeping in a language game. Journal of Philosophical Logic 8, 339-359.

Lewis, David K. (1982). Are we free to break the laws? Theoria 47, 113-121.

88 Lewis, David K. (1983). New work for a theory of universals. Australasian Journal of Philosophy 61, 343-377.

Lewis, David K. (1984). Putnam’s Paradox. Australasian Journal of Philosophy 62, 221- 236.

Loewer, Barry. (1997). Quantum mechanics and free will. Philosophical Topics 24, 91- 112.

Lycan, William. (2003). Free will and the burden of proof. In A. O’Hear (Ed.), Minds and Persons: Royal Institute of Philosophy Supplement 53 (pp. 107-122). Cambridge: Cambridge University Press.

Mackie, J. L. (1977). Ethics: Right and Wrong. London: Penguin Books.

McKay, Thomas & Johnson, David O. (1996). A reconsideration of an argument against compatibilism. Philosophical Topics 24, 113-22.

Mele, Alfred R. (1995). Autonomous Agents. New York: Oxford University Press.

Mele, Alfred R. (2003). Motivation and Agency. New York: Oxford University Press.

Mele, Alfred R. (n.d.). A critique of Pereboom’s “four case argument” for incompatibilism. Analysis, forthcoming.

Merricks, Trenton. (2001). Objects and Persons. Oxford: Clarendon Press.

Nelkin, Dana K. (2001). The consequence argument and the Mind argument. Analysis 61, 107-115.

Nichols, Shaun. (n.d). The folk psychology of free will: fits and starts. Mind & Language, forthcoming.

Nichols, Shaun, & Stitch, Stephen. (2003). Mindreading. Oxford: Oxford University Press.

O’Connor, Timothy. (1993). On the transfer of necessity. NOÛS 27, 204-218.

O’Connor, Timothy. (2000). Persons and Causes: The Metaphysics of Free Will. New York: Oxford University Press.

Pereboom, Derk. (2001). Living Without Free Will. Cambridge: Cambridge University Press.

89 Pereboom, Derk. (2002). Living without free will: the case for hard incompatibilism. In R. Kane (Ed.), The Oxford Handbook of Free Will (pp.477-488). Oxford: Oxford University Press.

Putnam, Hilary. (1975). The meaning of meaning. In H. Putnam (Ed.), Mind, Language, and Reality: Philosophical Papers (Vol. 2, pp. 215-271). Cambridge: Cambridge University Press.

Putnam, Hilary. (1980). Models and reality. The Journal of Symbolic Logic 45, 464-482.

Ramsey, William. (1998). Prototypes and conceptual analysis. In M. R. DePaul & W. Ramsey (Eds.), Rethinking Intuition (pp. 161-177). Lanham, Md.: Rowman & Littlefield.

Schnieder, Benjamin Sebastian. (2004). Compatibilism and the notion of rendering something false. Philosophical Studies 117, 409-428.

Smilansky, Saul. (2000). Free Will and Illusion. Oxford: Oxford University Press.

Smilansky, Saul. (2001). Free will, fundamental dualism, and the centrality of illusion. In R. Kane (Ed.), The Oxford Handbook of Free Will (pp. 489-505). Oxford: Oxford University Press.

Sider, Theodore. (2001a). Criteria of personal identity and the limits of conceptual analysis. In J. E. Tomberlin (Ed.), Philosophical Perspectives 15: Metaphysics, (pp. 189- 209). Boston: Blackwell.

Sider, Theodore. (2001b). Four-Dimensionalism. Oxford: Clarendon Press.

Slote, Michael. (1982). Selective necessity and the free-will problem. Journal of Philosophy 79, 5-24.

Strawson, Galen. (1986). Freedom and Belief. Oxford: Clarendon Press.

Strawson, Galen. (2001). The Bounds of Freedom. In R. Kane (Ed.), The Oxford Handbook of Free Will (pp. 441-460). Oxford: Oxford University Press.

Strawson, Galen. (n.d.). Free will. The Determinism and Freedom Philosophy Website, T. Honderich (Ed.). http://www.ucl.ac.uk/~uctytho/dfwVariousStrawsonG.html.

Strawson, Peter. (1962). Freedom and resentment. Proceedings of the British Academy 48. Reprinted in L. W. Ekstrom (Ed.), Agency and Responsibility (pp. 183-204). Boulder, Co.: Westview Press, 2001.

Taylor, Richard. (1963). Metaphysics. Englewood Cliffs, N.J.: Prentice-Hall.

90 Unger, Peter. (2002). Free will and scientiphicalism. Philosophy and Phenomenological Research 65, 1-25.

Van Inwagen, Peter. (1983). An Essay on Free Will. Oxford: Oxford University Press.

Van Inwagen, Peter. (1989). When is the will free? In J. E. Tomberlin (Ed.), Philosophical Perspectives 3: Philosophy of Mind and Action Theory (pp. 399-422). Atascadero, Calif.: Ridgeview Publishing Co.

Van Inwagen, Peter. (1992). Reply to Christopher Hill. Analysis 52, 56-61.

Watson, Gary. (1982). Free Will. Oxford: Oxford University Press.

Widerker, David. (1987). On an argument for incompatibilism. Analysis 47, 37-41.

91

BIOGRAPHICAL SKETCH

Jason Turner was born in Chester, Montana in 1978. In 1996 he graduated from Ponoka Composite High School of Ponoka, Alberta, Canada, and in August 2002 he received his Bachelor of Arts in Philosophy from Washington State University in Pullman, Washington. He is the author of “Strong and Weak Possibility,” Philosophical Studies (forthcoming), and co-author (with Eddy Nahmias, Stephen G. Morris, and Thomas Nadelhoffer) of “The Phenomenology of Free Will,” Journal of Consciousness Studies, (forthcoming). In the fall of 2004 Jason will move with his wife Starr and daughter Cami to New Jersey, where he will become a candidate for a Doctorate of Philosophy in the Department of Philosophy at Rutgers, the State University of New Jersey.

92