Social Theories of Reasoning
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of
Philosophy in the Graduate School of The Ohio State University
By
Paul David Robinson, M.Sc.
Graduate Program in Philosophy
The Ohio State University
2020
Dissertation Committee
Richard Samuels, Advisor
Neil Tennant
Declan Smithies
Copyrighted by
Paul David Robinson
2020
2 Abstract
This dissertation consists of four independent papers that address topics concerning the nature and function of reasoning. Together, they draw on a wide range of paleontological, archaeological, and ethnographic evidence, to constrain evolutionary theorizing about reasoning. Further, they draw on empirical results from literatures on persuasion and reasoning, to evaluate hypotheses about the characteristic features of reasoning. On these foundations, I argue against the increasingly influential view that the capacity for reasoning evolved in a social context, for public practices of persuasion and self- justification. The principal thesis of this dissertation is that the capacity for reasoning evolved for the private practice of generating true beliefs from rule-like cultural knowledge transmitted through stories, to solve novel problems in the natural environment.
ii Dedication
To Alan Weir
iii Acknowledgments
I wish to thank the following people. My advisor, Richard Samuels, who provided invaluable feedback. He is the second author of the fourth chapter in this dissertation, which was published in the Routledge Series in Philosophy of Mathematics and Physics, and I am grateful to him and to Routledge for permission to include it here. Dan Sperber and Tania Lombrozo, both of whose research I draw on, provided encouragement during the critical early stages of the project. Neil Tennant, Declan Smithies, and Kevin Scharp, helped structure the project, and ensured I didn’t lose sight of the big picture. I am fortunate that Neil Tennant is also my academic grandfather, having mentored my first philosophy teacher, Alan Weir, to whom this dissertation is dedicated. Alan has supported me throughout my studies, which have taken me all over the globe (though, thankfully, not yet to an Independent Scotland). I am grateful to all my teachers, who helped me draw from multiple disciplines: the archaeologists, Aren Maeir and Amit Dagan; the cognitive scientists, Nick Chater, Dave Lagnado, and Tom Lawson; the social psychologists, Duane
Wegener, Kentaro Fujita, and Russ Fazio. I have been shaped by many philosophical discussions with Gonga, Billy, Geezer, Jackie, Rab, Marty, Big Gilbert, and Wee Jim. I have had constant support from my parents, David and Brenda, and my American family,
Akram and Monda. Finally, I am most grateful to Lydia, who “deserves a medal”.
iv Vita
2001 ……………………………….. A-level Mathematics, Physics, Biology Portora Royal School
2005 ……………………………….. B.A. Philosophy Department of Philosophy Queen’s University Belfast
2008 ……………………………….. M.Sc. History and Philosophy of Science Department of Logic and Scientific Method London School of Economics
2009 ……………………………….. M.Sc. Cognitive and Decision Sciences Department of Psychology and Language Sciences University College London
2012 to present …………...... Graduate Teaching Associate Department of Philosophy The Ohio State University
Publications
Robinson, P. D. & Samuels, R. (2018). Reasoning, rules, and representation. In S. Bangu (ed.) Naturalizing Logico-Mathematical Knowledge: Approaches from Psychology and Cognitive Science (Routledge Series in Philosophy of Mathematics and Physics), pp. 30-51.
Fields of Study
Major Field: Philosophy
Interdisciplinary Specialization: Cognitive and Brain Sciences
v
Table of Contents
Abstract ...... ii
Dedication ...... iii
Acknowledgments ...... iv
Vita...... v
List of Tables ...... ix
List of Figures ...... x
Introduction ...... 1
1. Theoretical Background ...... 2
2. Overview ...... 7
Chapter 1 ...... 8
Chapter 2 ...... 9
Chapter 3 ...... 10
Chapter 4 ...... 11
3. Methodology ...... 12
Chapter 1. The Evidence is Not on My Side: Reason, Evolution, and Bias ...... 13
1. Introduction ...... 13
vi 2. The Rationalization-First View ...... 18
2.1 The Theory ...... 19
2.2 The Empirical Case...... 21
3. Does the Rationalization-First View Explain Myside Bias?...... 26
3.1 Persuasion ...... 26
3.2 Information-Gain ...... 34
3.3 Deception...... 37
4. From Persuasion to Justification ...... 39
5. The Inquiry-First View ...... 43
6. Concluding Remarks ...... 49
Chapter 2. Cracking the Enigma: Cultural Knowledge, Social Learning, and Private Reason ..... 51
1. Introduction ...... 51
2. The Rationalization-First View of Reason ...... 55
2.1 Exposition of the Theory ...... 55
2.2 Criticism 1: The Paleo-Anthropological Context ...... 59
2.3 Criticism 2: The Scope of Reason ...... 66
3. A New Inquiry-First View of Reason ...... 69
4. Empirical Characteristics of Reason ...... 80
5. Concluding Remarks ...... 84
Chapter 3. Wason Confirmed: Why Confirmation Bias is Not Myside Bias in Disguise ...... 86
1. Introduction ...... 86
2. Confirmation Bias ...... 90
2.1 The Four Card Selection Task ...... 90
2.2 The 2-4-6 task ...... 105
vii 3. Myside Bias...... 113
4. Concluding Remarks ...... 121
Chapter 4. Reasoning, Rules, and Representation ...... 123
1. Introduction ...... 123
2. The Virtues of Intentional Rule-Following Accounts of Inference ...... 128
3. The Regress ...... 132
3.1 The General Strategy ...... 133
3.2 Establishing Condition 2 ...... 134
3.3 The Regress Within Reach ...... 136
3.4 The Regress Regimented ...... 138
4. Sub-Personal Processes, Revenge, and the Strengthened Regress ...... 140
4.1 Getting Sub-Personal...... 140
4.2 The Revenge Regress ...... 141
4.3 Regress Strengthened ...... 143
5. Rejecting the Strengthened Regress ...... 145
5.1 The “Primitivist” Strategy ...... 146
5.2 Primitive Processes and Reflexes ...... 147
5.3 Primitive Processes and Stored Program Computers ...... 148
5.4 Primitive Processes and Cognitive Architecture...... 149
6. Counter-Arguments ...... 150
6. Conclusion...... 156
References ...... 157
viii List of Tables
Table 1: Rationalization-first views of reason ...... 5
Table 2: Inquiry-first views of reason ...... 5
Table 3. Card selections predicted by strategies in the negations paradigm ...... 101
ix List of Figures
Figure 1: Card faces in the selection task ...... 90
Figure 2: Card-turning scenarios in the selection task ...... 91
Figure 3: Canonical card selections made in the selection task ...... 93
Figure 4. Card faces in the selection task explicit negations paradigm...... 104
Figure 5. Performance of selection task strategies with respect to falsification...... 110
Figure 6. Performance of selection task strategies, restricted to relevant cases ...... 112
Figure 7. Arguments cited for and against one’s view (Perkins) ...... 114
Figure 8. Arguments cited for and against one’s view (Toplak & Stanovich) ...... 117
x Introduction
Of all the faculties of the human mind, it will, I presume, be admitted that reason stands at the summit. Only a few persons now dispute that animals possess some power of reasoning.
— Charles Darwin, The Descent of Man (1871/1874)
Neither can I easily bring myself to the idea that man’s reasoning faculties & above all his moral sense, cd. ever have been obtained from irrational progenitors, by mere natural selection … This seems to be doing away altogether with the Divine Image which forms the insurmountable distinction between man & brutes.
— Leonard Jenyns, letter to Charles Darwin (4 Jan, 1860)
The topics addressed in this dissertation fall squarely within a longstanding debate regarding the nature of reasoning (what reasoning is) and the function of reasoning (what reasoning is for). Although the debate has been especially prominent within the evolutionary and cognitive sciences, it is also one to which empirically oriented philosophers have contributed. It is in this spirit that I approach the topics in this dissertation.
1 The psychological phenomenon at issue is what some philosophers have called
“active” reasoning (e.g. Broome, 2013). Minimally characterized, active reasoning is a person-level process, involving reflection on explicitly represented reasons to draw a conclusion. Henceforth, all uses of the term “reasoning” should be taken to refer to active reasoning. Thus, this dissertation is not concerned with unconscious, sub-personal, inference. Nor is it restricted to any particular kind of reasoning, such as logical reasoning, probabilistic reasoning, analogical reasoning, or causal reasoning. Rather, my focus is on reasoning tout court.
In contemporary discussions of reasoning, “function” is taken to be synonymous with primary adaptive function, unless otherwise qualified. To specify the primary adaptive function of a trait is to identify fitness-enhancing biological effects of the trait that led it to persist in the population. The principal thesis of this dissertation is that the function of reasoning is the private generation of true beliefs to solve novel problems in the natural environment.
In the remainder of this introduction, I provide a theoretical background for this dissertation (section 1), an overview of its four chapters (section 2), and some brief comments on methodology (section 3).
1. Theoretical Background
The primary goal of this dissertation is to critically assess the recent “social turn” in the study of reasoning. That reasoning evolved for a social function was first proposed in detail by the linguist, philosopher, and cognitive anthropologist, Dan Sperber, in his research on
2 the evolution of human capacities for metarepresentation and verbal communication
(Sperber, 2000, 2001a). According to his “argumentative theory of reasoning”, reasoning evolved not for the “individual Cartesian thinker” to generate knowledge (Sperber, 2001a, p. 408) but rather to produce and evaluate arguments that individuals use to manipulate others through persuasion.
Around the same time, the social psychologist Jonathan Haidt proposed his highly influential social intuitionist model of moral judgment, according to which reasoning tends to justify the reasoner’s intuitions because it evolved to protect and enhance the reasoner’s reputation, rather than to search for the truth (Haidt, 2001, 2012a).
Recent work by Mercier and Sperber incorporates both persuasion and self- justification into an overarching theory of reason:
Reason, we argue, has two main functions: that of producing reasons for justifying oneself, and that of producing arguments to convince others. These two functions rely on the same kinds of reasons and are closely related. (Mercier & Sperber, 2017, p. 8)
Since both persuasion and self-justification are social practices, they refer to this new theory as the “interactionist” account of reason. The interactionist account is increasingly popular, with many researchers in philosophy, psychology, and the cognitive sciences finding the core claims compelling (see, e.g., De Neys, 2018a; Dutilh Novaes, 2018;
Evans, 2019; Sterelny, 2018).
The interactionist account is typically contrasted with a supposed “intellectualist” tradition which holds that reason evolved to “improve individual cognition and arrive on
3 one’s own at better beliefs and decisions” (Mercier & Sperber, 2017, p. 330). However, it is important to distinguish the function of reasoning and the contexts reasoning evolved to operate in. Given the nature of the functions Mercier and Sperber posit, the relevant contexts will always be social. Nevertheless, the two can come apart. For example, the capacity to reason may have evolved for individuals to gain knowledge through group argumentation. Here the context is social (interactionist), but the function is epistemic
(intellectualist). Thus, I introduce the terms “rationalization-first” and “inquiry-first” to partition theories of reason. According to rationalization-first views, reason evolved to support the reasoner’s pre-existing beliefs and decisions for the purposes of persuasion and/or self-justification (Table 1).1
According to inquiry-first views, reason evolved to test the reasoner’s pre-existing beliefs and decisions, and/or to discover new truths (Table 2). There are prima facie plausible asocial versions of the inquiry-first view, according to which reason evolved for
1 “Rationalization” was adopted as a technical term in Freudian psychology, where it originally referred to the practice of offering socially acceptable motives for one’s actions that hide the repressed true motives (Lundholm, 1933; Taylor, 1923). However, it was also commonly used to cover implicit justifications of one’s beliefs and attitudes (Hollitscher, 1939; Jones, 1908; Piddington, 1928; Robinson, 1921; Walker, 1928). Some communication scholars further expanded the concept to include a type of persuasion that falls between genuine deliberation and emotional appeal (Oliver, 1936). I use “rationalization” to refer to the post hoc practice of attempting to cite normative reasons in favor of an actual or pretended judgment or decision (where these result in beliefs, attitudes, intentions, etc.), typically under the guise of those reasons also explaining why the relevant conclusion was reached.
4 lone individuals to attain epistemic goals through private deliberation. This dissertation develops and defends one such account.
SOCIAL ASOCIAL
PERSUASION SPERBER (2001) NA
SELF-JUSTIFICATION HAIDT (2001) NA
BOTH MERCIER & SPERBER (2017) NA
Table 1: Rationalization-first views of reason.
SOCIAL ASOCIAL
TESTING
DISCOVERY ROBINSON (2020)
BOTH INTELLECTUALISM
Table 2: Inquiry-first views of reason.
Inquiry-first views face genuine challenges. First, there is the challenge of explaining why the relevant epistemic goals were fitness-enhancing, and why they were best achieved through reasoning. Answers to these questions are often gestured towards rather than explicitly spelled out (e.g. Berwick & Chomsky, 2016, p. 166). Although it is clear that reasoning – used in conjunction with the more recent ability to read and write – underpins modern achievements such as science and engineering and perhaps even the occasional
5 doctoral dissertation, it is not a trivial task to show that reasoning was highly beneficial in ancestral environments.
Second, there is the challenge of explaining what appear to be irrational tendencies of reasoning. For instance, within psychology there is a growing consensus that reasoning tends to justify rather than correct mistaken intuitions (De Neys, 2018b). This is often taken to be one manifestation of a wider “confirmation bias” – or, as Mercier and Sperber prefer to call it, “myside bias” – that reasoning seems to display. Such a bias is especially problematic for asocial versions of the inquiry-first view, because they hold that reasoning evolved to operate without critical feedback from interlocutors. For instance, in an interview for a popular audience that has been viewed online over a quarter of a million times, Haidt states:
Something we need to talk about here is what's called the confirmation bias. That is, you might think that our reasoning is designed to find the truth. And if you want to find the truth, you should look on both sides of a proposition. But in fact what happens is, when someone gives you a proposition, our minds, we send them out, we sent them out to do research for us. But it’s research, like, as a lawyer does, or as a press secretary would do, it’s like, “Find me one piece of evidence that will support this claim that I want to make”. And if I can find one piece of evidence, I'm done. I can stop thinking. Well, that's the way we've been for … hundreds of thousands of years. (Haidt, 2012b)
Since the notion of confirmation bias is well-known even outside of academia, and is often cited to explain pressing societal issues such as increased levels of polarization, the claim that confirmation bias conflicts with inquiry-first views is rhetorically powerful.
6 Moreover, it is difficult to see how the kinds of epistemic functions that characterize inquiry-first views can be reconciled with “motivated reasoning” (Kunda, 1990) and
“motivated skepticism” (Ditto & Lopez, 1992; Taber & Lodge, 2006), i.e., the tendency to judge preference-consistent information as more valid or accurate than preference- inconsistent information, due to subjecting preference-inconsistent information to a greater degree of critical reasoning. For although considerations of truth play some role, the process is “best characterized as a compromise between the wish to reach a particular conclusion and the plausibility of that conclusion given the available data” (Ditto & Lopez
1992, 569).
The secondary goal of this dissertation is to meet these challenges by developing an inquiry-first view of reason that explains the empirical features of reason, while remaining faithful to the paleo-anthropological evidence concerning human evolution.
2. Overview
Each chapter in this dissertation is written as an independent paper. However, the topics addressed form a natural progression: from critiquing the empirical case for the rationalization-first view, to developing my own version of the inquiry-first view, to reconsidering the received wisdom about confirmation bias and myside bias, to defending a kind of rule-following account of the nature of reasoning that dovetails with (though is not presupposed by) my account of the function of reasoning.
7 Chapter 1
This chapter argues that the existence of myside bias does not support the rationalization- first view over the inquiry-first view.
As glossed by Mercier and Sperber, myside bias results in people being good at formulating arguments which support their beliefs, but bad at formulating arguments which counter their beliefs. Mercier and Sperber claim that the rationalization-first view explains myside bias, since producing arguments that support one’s beliefs facilitates persuasion and self-justification.
My response is as follows. Empirical research on communication suggests that if the function of argument production is persuasion, then people should be good at anticipating objections to their beliefs. What allows the rationalization-first view to explain myside bias is not the function of reason it posits, but rather the auxiliary hypothesis that on average it was cost-effective to offload to interlocutors the task of formulating counterarguments. Ethnographic studies of hunter-gatherer societies suggest that this auxiliary hypothesis is implausible.
However, even if the auxiliary hypothesis is granted, inquiry-first views can mimic the rationalization-first strategy for explaining myside bias. Social versions of the inquiry- first view can claim that myside bias evolved to enable a division of argumentative labor that enables groups to pursue epistemic goals efficiently. Asocial versions of the inquiry- first view can claim that myside bias evolved as a cost-effective means of checking whether there are good grounds for one’s beliefs.
8 Chapter 2
This chapter argues that: (i) the rationalization-first view is based on an inaccurate reconstruction of ancestral environments, and fails to capture inquisitive reasoning; whereas (ii) my version of the inquiry-first view is based on an accurate reconstruction of ancestral environments, and captures argumentative reasoning.
Mercier and Sperber propose that the capacity to reason evolved to overcome an information bottleneck in verbal communication that mainly resulted from the threat that senders of messages were engaging in deception. My response is as follows. Given the small-scale, egalitarian nature of the societies that existed when verbal communication evolved, the risk of deception was negligible. Moreover, a reasoning mechanism adapted for backwards inference, from pre-existing beliefs to arguments in favor of those beliefs, is fundamentally ill-suited for the kind of forwards inference from premises to previously unconsidered conclusions that enables science and engineering, for example, as well as a host of other distinctly human achievements.
I propose that the capacity to reason instead evolved to generate true beliefs for the purpose of solving novel problems in the natural environment, for which the reasoner has no intuitions. Drawing on ethnographic studies of hunter-gatherer societies, I argue that the fitness of individuals in the relevant ancestral environments was enhanced by reasoning taking as input rule-like (generalizable) cultural knowledge transmitted through stories. Since under this account reason is a problem-solver, it is a highly flexible system that can serve a wide variety of motives and goals. In particular, reasoning may
9 be used to persuade others, justify oneself, or find fault with views one doesn’t want to believe. Thus, my account captures argumentative reasoning. Further, since I do not claim that reasoning evolved to test and correct the reasoner’s pre-existing beliefs, my account is compatible with the existence of myside bias and motivated reasoning, even without adding an auxiliary hypothesis of the sort discussed in chapter 1.
Chapter 3
This chapter argues that there is good evidence for confirmation bias with respect to hypothesis-testing, but no good evidence for myside bias with respect to one’s beliefs.
Mercier claims that confirmation bias holds relative to the beliefs of the reasoner but not relative to mere hypotheses the reasoner is testing, and is thus better described as
“myside bias”. My response is as follows. When people test a hypothesis that takes the form of a conditional statement, rather than search for falsifying cases, they tend to unintentionally search for instances of the conditional which, if discovered, confirm the hypothesis. It is in this sense that a bias towards confirming evidence exists.
In contrast, there is no good evidence that people tend to unintentionally search for evidence that favors their beliefs. The classic myside bias effect is that when people are tasked with coming to a conclusion on a controversial topic, and are then asked to explain the reasoning for their conclusion, they tend to provide more arguments for their view than arguments against it. I argue that these effects can be explained away as reflecting why the (unbiased) reasoner came to the conclusion they did.
10 Chapter 4
This chapter, which is co-authored with Richard Samuels, argues that no vicious regress results from supposing that reasoning essentially involves following rules that are explicitly represented.
What we call “intentional rule-following” accounts of the nature of reasoning make the following two commitments. First, reasoning essentially involves following rules concerning the premises from which one reasons. Second, all rule-following involves
(conscious or unconscious) intentional states which represent the rules being followed. A number of philosophers – e.g. Boghossian, Broome, and Wright – argue that such theories fall prey to a vicious regress. The problem is generated by supposing that all rule-following involves grasping what the rule states, forming a belief that its antecedent is satisfied, and drawing the conclusion the act required by its consequent is required. In other words, all rule-following involves reasoning.
Our response is as follows. We posit a level of primitive processing mechanisms that take rule-representations as inputs. The primitive processes they subserve will be rule-guided in the thin sense that a rule-representation is causally implicated in the process. However, those primitive processes are not rule-guided in the sense that they involve further rule-guided or inferential sub-processes. Thus, if non-primitive processes—such as those involved in reasoning—ultimately decompose into primitive ones, no regress is generated.
11 3. Methodology
Research on the nature and function of reasoning has tended to proceed in disciplinary isolation. Although previous work has cited empirical data from the “psychology of reasoning” literature, I also draw on empirical data from the “attitudes and persuasion” and “communication studies” literatures. Further, although empirical claims are routinely made about ancestral environments, evidence is rarely cited.2 This dissertation aims to make a unique contribution to the literature by using paleontological, archaeological, and ethnographic data to constrain evolutionary theorizing about the function of reasoning.
The notion of empirical confirmation at play in this dissertation is the comparative confirmation of one theory relative to another, in the light of all the available evidence, given the assumption that human evolution was a wholly naturalistic affair.
Due to the interdisciplinary ethos of this dissertation, a number of terminological clarifications are in order. I use “attitude” to refer to an evaluative summary of an object, rather than a mental state towards a proposition. With respect to reports of meta- analyses, “r” refers to the correlation co-efficient and “k” refers to the number of studies surveyed. Finally, “my” and “ky” abbreviate “millions of years ago” and “thousands of years ago”, respectively.
2 A similar observation is made by Sterelny (2018, p. 4).
12 Chapter 1 The Evidence is Not on My Side: Reason, Evolution, and Bias
1. Introduction
In recent years a number of influential researchers in philosophy, psychology, and cognitive science have converged on the idea that the human capacity to reason evolved for the attainment of socially strategic goals through persuasion and self-justification (e.g.
Haidt, 2012a; Mercier & Sperber, 2017). Call this the “rationalization-first” view of reason.
Rationalization-first because typically these public practices of persuasion and self- justification are attempts to cite normative reasons in favor of a belief or decision that the reasoner has already made, often under the guise of those reasons explaining how the relevant conclusion was reached. Rationalization-first because the capacity to reason is held to have emerged and persisted within the Homo sapiens lineage mainly due to the fitness-enhancing effects of persuasion and self-justification. Given that these are the primary adaptive functions of reason, ceteris paribus they should be the tasks to which our reasoning system naturally defaults.
13 The rationalization-first view is often contrasted with the supposedly dominant idea that the primary adaptive function and modus operandi of reason is to attain epistemic goals, such as avoiding false beliefs and gaining true beliefs (e.g. Haidt, 2012a, p. 74; Mercier & Sperber, 2017, p. 218). Call this the “inquiry-first” view of reason. Inquiry- first because attaining these goals involves attempting to ascertain the facts about some issue. Applied to a preexisting belief or decision, inquiry is a form of testing through the weighing of normative reasons concerning the relevant conclusion. Alternatively, in the absence of a preexisting belief or decision, inquiry is a form of discovery that searches for relevant evidence and seeks to abduce to a new conclusion based on a consideration of the normative implications of the gathered evidence.
Why suppose that the rationalization-first view is preferable to the inquiry-first view? Here is one oft-cited line of argument. Empirical research shows that reasoning displays a characteristic known as “confirmation bias”. Though the term “reasoning” gets used in a variety of ways, the phenomenon of relevance here is not sub-personal inference but rather a person-level activity which delivers a verbalizable awareness of something to the effect that “P is a reason for Q ”. To a first approximation, confirmation bias is the tendency to unwittingly search for evidence in ways which favor information that supports rather than undermines one’s pre-existing belief or working hypothesis (Klayman, 1995;
Nickerson, 1998). The existence of such a bias is puzzling if reason evolved to facilitate the attainment of epistemic goals, since it can lead people to be overconfident that their beliefs are correct. However, confirmation bias appears to facilitate persuasion and self-
14 justification, since it helps the reasoner build a case. Thus, in his bestseller The Righteous
Mind, the social psychologist Jonathan Haidt writes:
confirmation bias is a built-in feature (of an argumentative mind), not a bug that can be removed … each individual reasoner is really good at one thing: finding evidence to support the position he or she already holds. (2012a, p. 105f.)
Sometimes the claim is framed in terms of “myside bias”, which, in this context, is taken to be a restricted form of confirmation bias that holds relative to the reasoner’s beliefs or decisions rather than mere hypotheses they happen to be testing.3 Thus conceived, myside bias will result in people being better at formulating arguments which support their beliefs and decisions, than formulating arguments which undermine their beliefs and decisions. Hence, in The Enigma of Reason the cognitive scientists Hugo Mercier and Dan
Sperber state:
A lot of evidence shows that reasoning has a myside bias… This is pretty much the exact opposite of what you should expect of a mechanism that aims at improving one’s beliefs through solitary ratiocination … [However] If the function of reasoning, when it produces reasons, is to justify one’s actions or to convince others, then it should have a myside bias. (2017, p. 218)
Arguably, the rationalization-first view and the associated claim that it explains myside bias are becoming the mainstream position.4 For instance, they have been used as a
3 Henceforth, I refer only to “myside bias”. 4 For pushback see critical reviews of The Enigma of Reason by Dutilh Novaes (2018) and Sterelny (2018).
15 framework for understanding System 2 in dual-process theories of reasoning, and the tendency to defend rather than correct mistaken initial responses in experimental tasks
(De Neys, 2018a; Evans, 2019). It has also been suggested that they help elucidate cases of confabulation (Ganapini, 2019), group-level scientific inquiry (Peters, 2019), the propagation of pseudoscientific beliefs (Blancke, Boudry, & Braeckman, 2019), and the reliability of our intellectual faculties (Bland, 2018).
Suppose that myside bias exists in the sense outlined above. What does this empirical phenomenon tell us about the function of reasoning? My goal in this paper is to show that the existence of myside bias does not support the rationalization-first view over the inquiry-first view. To explain myside bias, the rationalization-first view needs to be supplemented by an auxiliary hypothesis that is not warranted by the empirical evidence.
But even if this auxiliary hypothesis is granted, serious theoretical problems remain. For as I will show, the resulting explanation of myside bias undermines core components of the rationalization-first view concerning the evolutionary history and adaptive functions of reason. When this is appreciated, it becomes clear that if the inquiry-first view is supplemented with similar auxiliary hypotheses, it provides an explanation of myside bias which is at least as plausible as the explanation provided by the rationalization-first view.
Here’s how I proceed. I begin by reconstructing the standard version of the rationalization-first view, according to which the ability to produce and evaluate arguments evolved to overcome an information bottleneck which resulted largely from the threat of being misinformed through deception. Proponents of the rationalization-first
16 view claim that since the capacity to produce arguments evolved for senders of messages to persuade the receivers of those messages, the view predicts that people will be good at producing arguments in support of their beliefs but relatively bad at producing arguments that undermine their beliefs.5
The main burden of this paper is to develop my objections to this claim (section 3).
I show that if the adaptive function of argument production is persuasion, then ceteris paribus people should be good at anticipating objections to their beliefs. Mercier and
Sperber appear to suggest that it was advantageous for senders to offload this task to receivers, because anticipating objections is cognitively demanding. Based on paleontological, archaeological, and ethnographic evidence, I argue that this auxiliary hypothesis depends on a cost-benefit analysis that is unlikely to hold for the environment of evolutionary adaptedness. However, even if the auxiliary hypothesis is granted, the resulting explanation of myside bias undermines two core components of the rationalization-first view. The first is that argument evaluation evolved to help receivers accept genuine information and reject misinformation. The second is that argument production evolved in part to help senders deceive receivers with arguments in favor of views the senders don’t hold. Although these two problems do not suffice to show that the rationalization-first explanation of myside bias is incorrect, they do substantially weaken its plausibility. In section 4, I examine whether the problems can be overcome by
5 Of course, over time any given individual will act as a sender and as a receiver.
17 hypothesizing that justification rather than persuasion is the primary adaptive function of argument production. Although this modification has some independent merit, I show that the relevant problems resurface for the rationalization-first view.
Finally, I sketch alternative explanations of myside bias under the inquiry-first view. On the one hand, we can imagine social versions of the inquiry-first view, according to which the capacity to reason evolved for the attainment of epistemic goals through the public trading of arguments. Such views can explain myside bias as resulting from a cost- effective division of argumentative labor that is efficient at reaching the truth, without the need to accommodate an evolved capacity for deception. On the other hand, we can imagine asocial versions of the inquiry-first view, according to which the capacity to reason evolved for the attainment of epistemic goals through private deliberation.6 Such views can explain myside bias as resulting from a cost-effective heuristic that tests the reasoner’s beliefs by checking whether there are good grounds to hold them. I thus conclude that the existence of myside bias fails to adjudicate between the rationalization-first view and the inquiry-first view.
2. The Rationalization-First View
The adaptive function of any trait is wedded to its evolutionary etiology. The rationalization-first view represents a thoroughgoing departure from traditional ways of
6 That there is no plausible asocial version of the rationalization-first view follows from the nature of the functions it posits.
18 thinking about reason, based on considerations relating to the social aspects of human evolution. In this section I clarify what the proposal is.
2.1 The Theory
The rationalization-first view was first proposed in detail by Dan Sperber in his work on the evolution of human capacities for verbal communication (Sperber, 2000, 2001a). For communication to be evolutionarily stable in a population the benefits of both sending and receiving messages must, on average, outweigh their costs. In particular, if sending messages had on average been more costly than beneficial then the practice would have ceased, resulting in the complementary practice of receiving messages also ceasing (and vice versa). The verbal exchange of messages allows senders to manipulate how receivers behave, and receivers to gain useful information from senders. This practice is often highly beneficial for both sender and receiver, as when it is used to coordinate cooperative ventures. However, it can also be highly detrimental for a receiver to be misinformed by a sender.
Sperber suggested that in the early stages of the history of verbal communication the threat of being misinformed generated selective pressure for mechanisms of “epistemic vigilance”. One set of mechanisms evaluate the source of the information based on cues relating to the competence and honesty of the sender. Another set evaluate the content of the information based on how it coheres internally and with respect to some relevant subset of the receiver’s beliefs. The major factor driving the evolution of these mechanisms
19 was not the risk that senders were mistaken, but rather the risk that senders were engaging in deception:
the major problem posed by communicated information has to do not with the competence of others, but with their interests and their honesty. While the interests of others often overlap with our own, they rarely coincide with ours exactly. In a variety of situations, their interests are best served by misleading or deceiving us. It is because of the risk of deception that epistemic vigilance may be not merely advantageous but indispensable if communication itself is to remain advantageous. (Sperber et al., 2010, p. 360)
At this point a vigilance-persuasion “arms race” developed. To the extent that epistemic vigilance delivered “true positives” that blocked misinformation, it deprived senders of the benefits of deception. On the other hand, when epistemic vigilance delivered “false positives” that blocked genuine information, it deprived both senders and receivers of the benefits that would have resulted from cooperation based on the shared information. This
“information bottleneck” generated selective pressure for senders to take advantage of epistemic vigilance through “honest displays” in which they present considerations in favor of the target message that the receiver either already accepts or will accept more readily than the target message in isolation:
The argumentative use of reasons helps genuine information cross the bottleneck that epistemic vigilance creates in the social flow of information. It is beneficial to addressees by allowing them to better evaluate possibly valuable information that they would not accept on trust. It is beneficial to communicators by allowing them to convince a cautious audience. (Mercier & Sperber, 2017, p. 194)
20 The evolution of the ability to produce arguments in support of a target message stimulated the flow of genuine information.7 However, it also provided an additional means by which senders could deceive receivers, through specious arguments:
sophistry is a way to use the “honest display” strategy of argumentation in a dishonest way and thereby make it more advantageous for the communicator. In other words, sophistry contributes to making argumentation adaptive. (Sperber, 2001a, p. 411)
Senders engaging in argumentative deception generated further selective pressure for the ability to critically evaluate arguments. In turn, this led to a sharpening of the ability to produce persuasive arguments. And so on. In this way, a specialized cognitive system
(“reason module”) dedicated to the production and evaluation of arguments evolved for persuasion and information-gain.8
2.2 The Empirical Case
In evolutionary approaches to cognition, research is often described as being either theory- driven or observation-driven (Lewis, Al-Shawaf, Conroy-Beam, Asao, & Buss, 2017; Reeve
& Sherman, 2007). A theory-driven “top-down” approach specifies conditions in ancestral environments that would have impacted biological fitness (i.e., hypothetical selection pressures), and then postulates the evolution of a psychological trait which increased
7 Mercier (2020) prefers the term “open vigilance” to “epistemic vigilance” because the mechanisms are as much about accepting genuine information as rejecting misinformation. 8 This was dubbed the “argumentative theory of reasoning” (Sperber et al., 2010).
21 fitness in that environment (i.e., a hypothetical adaptation). In contrast, an observation- driven “bottom-up” approach specifies a currently existing psychological trait, and then postulates selection pressures in past ancestral environments which would render the trait an adaptation. Although the latter approach is often criticized as “just-so storytelling”, qua science both the top-down and bottom-up approaches must be used to generate novel predictions, which may concern either the nature of the trait (its causes and effects) or the nature of ancestral environments (as revealed by archaeology, paleontology, and the like).
The early development of the rationalization-first view represents a top-down approach. Indeed, when the theory was first proposed by Sperber around the turn of the millennium, no empirical reasoning research was cited. Instead, Sperber used the theory to predict that reasoning will be facilitated in argumentative contexts:
logically equivalent tasks will yield significantly better performance when they are presented in a context where subjects are scrutinizing arguments plausibly aimed at persuading them than when they are evaluating these arguments in the abstract (as happens in most experimental tasks). (Sperber, 2000, p. 136)
By the turn of the next decade, Sperber – with his new collaborator Hugo Mercier – began to link the argumentative theory to results in the psychology of reasoning literature. The finding that groups outperform individuals on average in experimental tasks on reasoning
(e.g. Moshman & Geil, 1998) was taken to confirm the prediction that argumentative contexts should facilitate performance (Mercier & Sperber, 2009, p. 261). The rationale for this prediction is that the ability to evaluate arguments evolved to facilitate the acceptance
22 of genuine information and the rejection of misinformation. Thus, people will be
“demanding so as not to be deceived by poor or fallacious arguments into accepting false ideas” and “objective so as to be ready to revise our ideas when presented with good reasons” (Mercier & Sperber, 2017, p. 332):
DEMANDING: People are persuaded by an argument only when it is good.
OBJECTIVE: People are persuaded by an argument when it is good.
Compared to a single individual, a group of individuals will tend to generate more potential solutions, and more arguments for and against any given solution. Since everyone in the group tends to accurately evaluate arguments, the correct solution is more likely to win out.9 But as Mercier and Sperber seem to recognize (2017, pp. 235; 332),
DEMANDING and OBJECTIVE are also likely to hold if reasoning evolved to facilitate epistemic goals. Hence, the evidence in favor of DEMANDING and OBJECTIVE does not support the rationalization-first view over the inquiry-first view.
However, around this time a new empirical claim began to emerge. This was the claim that the rationalization-first view explains a generalized form of confirmation bias:
when we try to persuade someone that something is true (or false), a confirmation (or disconfirmation) bias may help us achieve our goal. (Mercier & Sperber, 2009, p. 163)
9 There are caveats, of course. For example, if everyone in the group starts out with the same opinion, then “group polarization” may result.
23 The term “confirmation bias” has been used to refer to a wide variety of phenomena. Early research using the 2-4-6 rule-discovery task (Wason, 1960) and four-card selection task
(Wason, 1966) suggested that people tend to test hypotheses in ways that unintentionally privilege supporting evidence at the expense of falsifying evidence, perhaps due to cognitive limitations such as the difficulty of holding more than one hypothesis in mind or forming mental representations of hypothesis-inconsistent information (Evans, 1989).
Mercier and Sperber claim to explain a different kind of phenomenon that holds when what is at stake is not an arbitrary hypothesis but rather the views of the reasoner.
In the classic “myside bias” paradigm, participants are asked to think about a controversial issue and come to a conclusion if possible. When explaining the reasoning for their conclusion, participants tend to provide more (myside) arguments for their view than (otherside) arguments against it (Perkins, 1989). This effect holds even when participants are instructed at the outset to produce arguments on both sides of the issue.
Furthermore, people find it easier to provide counterarguments to opposing beliefs than to their own (Kuhn, 1991, pp. 171; 276). The explanation given by Sperber and Mercier is that since reason evolved to allow senders to persuade receivers, it will search for and output arguments that tend to favor the sender’s beliefs:
MYSIDE BIAS: People produce arguments that tend to favor their beliefs.
24 These arguments can favor the sender’s beliefs directly by concluding to the truth of the sender’s beliefs, or indirectly by concluding to the falsity of an opposing belief (2017, pp. 9,
218, 235):
What they find difficult is not looking for counterevidence or counterarguments in general, but only when what is being challenged is their own opinion. Reasoning does not blindly confirm any belief it bears on. Instead, reasoning systematically works to find reasons for our ideas and against ideas we oppose. (Mercier & Sperber, 2017, p. 218)
Although this may appear to represent a bottom-up approach, as we have seen, the rationalization-first view had already been proposed on independent grounds. However, neither was the existence of myside bias a novel prediction of the theory. Rather, myside bias effects were known and subsequently explained by the theory.10 Crucially, Mercier and Sperber add that MYSIDE BIAS should not hold if reasoning evolved to facilitate epistemic goals (2017, pp. 235; 332). They conclude that the data “unambiguously supports” their theory over what they take to be its main rival (Mercier & Sperber, 2017, pp. 203; 275). Thus, the existence of myside bias is held to provide comparative confirmation of the rationalization-first view relative to the inquiry-first view.
10 Sometimes it is claimed that the existence of myside bias is a “retrodiction” of the theory. However, it is more common to use “retrodiction” to refer to the derivation of unknown data gathered in the past.
25 3. Does the Rationalization-First View Explain Myside Bias?
In this section I show that myside bias does not follow from the adaptive functions posited by the rationalization-first view. Rather, the explanatory work is shouldered by auxiliary assumptions that are empirically unwarranted.
3.1 Persuasion
Suppose the primary adaptive function of argument-production is persuasion. Intuitively, experience shows that a failure to anticipate counterarguments is often fatal to a sender’s attempt to persuade a receiver. For example, my attempt to persuade an open-minded acquaintance to support the introduction of single-payer healthcare to the United States floundered due to my inability in the spur-of-the-moment to deal with their objection that this would increase wait times. Moreover, as I will show, beyond these kinds of anecdotal observations the social science literature on persuasion and argumentation provides robust evidence that anticipating counterarguments aides persuasion. Thus, prima facie, we should expect people to be relatively good at thinking of counterarguments to their views.
Call messages that present at least one myside argument “one-sided” when they do not include any counterargument, and “two-sided” when they do. The communication of a two-sided message may be interpreted as a sign that the sender is uncommitted, uncertain, or confused (Baron, 1995). As a result, asking people to author a message with the explicit goal of persuading an audience sometimes suppresses the communication of counterarguments (Nussbaum & Kardash, 2005; Felton, Crowell, & Liu, 2015). However,
26 the empirical research to-date is clear that two-sided messages tend to be more persuasive than one-sided messages when the relevant counterarguments are rebutted by the sender
(refutational two-sided messages), rather than merely dismissed or deemed to be outweighed by myside arguments (non-refutational two-sided messages). The two landmark meta-analyses investigating this issue, Allen (1998) and O’Keefe (1999), both found that refutational two-sided messages were more persuasive than one-sided messages (r = .074, k = 43; r = .077, k = 42), whereas non-refutational two-sided messages were less persuasive than one-sided messages (r = – .087, k = 26; r = – .049, k = 65).11
Subsequent studies corroborate these meta-analytic results. For example, Wolfe et al. (2009) found that written arguments were more persuasive when the author raised and adequately addressed objections to them. Relatedly, studies of mock jury trials find that the impact of negative information about defendants is reduced when it is preemptively addressed by the defense attorney (e.g. Howard, Brewer, & Williams, 2006).
Expert communicators appear to be sensitive to these two-sidedness effects. A content analysis of published essays, editorials, and opinion pieces in which professional authors
11 The mean effect sizes are small even by newly lowered benchmarks (e.g. Bosco, Aguinis, Singh, Field, & Pierce, 2015). However, they are typical in relation to the persuasion literature specifically. O’Keefe (2009) notes that for message variations which permit meta-analytic conclusions the largest mean effect size is .14. In the context of an evolutionary arms race we should expect effects that are not easily perceptible. Small effects can have a substantial impact when they are repeatedly manifested (Abelson, 1985), as they would be in argumentative exchanges over the course of an individual’s lifetime.
27 were aiming to persuade the general public revealed that they tended to include counterarguments for rebuttal (Wolfe & Britt, 2008, p. 2f.). In contrast, studies which show that the goal of persuasion tended to suppress the use of counterarguments typically used student participants who were merely aiming to fulfil academic assignments. Moreover, practicing attorneys and legal reasoning textbooks almost uniformly recommend that defense teams utilize the technique of “stealing thunder” by pro-actively addressing negative information (Williams et al., 1993).
So far we have considered the merits of the sender pro-actively raising and rebutting counterarguments. In argumentation theory this strategy is known as
“prolepsis” (Walton, 2009). In any particular case, whether the sender should voluntarily communicate counterarguments that come to mind may depend on situational variables such as how compelling the counterarguments are, how likely the receiver is to independently think of the counterarguments, and whether the receiver has a strong pre- existing opinion about the issue. But even when prolepsis is not used, there are two further ways in which anticipating counterarguments might be highly beneficial for the sender.
First, if those counterarguments are subsequently raised by the receiver, the sender will have had more time to think of a suitable framing and rebuttal. And, second, if the sender is aware of multiple myside arguments that establish the target conclusion, the sender can adopt a risk-averse strategy of choosing to communicate what they deem to be the least objectionable line of reasoning. Thus, for example, if a cleric is seeking to persuade a biologist that God exists it may be prudent for the cleric to argue from the
28 premise that the laws of physics are fine-tuned for the existence of life rather than from the premise that certain biological systems are irreducibly complex, even if in the cleric’s experience the latter “intelligent design argument” strikes most people as more intuitively compelling. Although the persuasive effects of these two kinds of strategies have not been studied directly, research on face-to-face conversations by Waldron and colleagues suggests that communication goals such as persuasion are facilitated by the formulation of concrete message plans that take into account the receiver’s perspective, since they allow senders to guide the conversation and better deal with contingencies as they arise
(Waldron, Caughlin, & Jackson, 1995).
Research findings about what is persuasive to well-educated Homo sapiens in the
20th and 21st centuries should not be carelessly generalized to our evolutionary past.
However, the factors proposed to underlie two-sidedness effects would also have been highly relevant in ancestral environments:
• Framing: Two-sided messages allow the sender to frame counterarguments in
favorable ways before they occur to the receiver.
• Reactance: Two-sided messages put less psychological pressure on the receiver
to accept the conclusion, thus reducing psychological reactance to the message.
• Effort: Refutational two-sided messages ease the argumentative burden of the
receiver, thus reducing processing effort and negative feelings.
29 • Reputation: Refutational two-sided messages avoid the receiver publicly raising
the relevant counterargument(s), thus reducing defensiveness due to
reputational concerns.
• Credibility: Refutational two-sided messages increase the perceived credibility
of the sender because they signal that the sender is both competent enough to
consider counterarguments and honest enough to communicate both sides of
the issue.
Indeed, if reasoning evolved to solve the adaptive problem of facilitating communication in cases where the message cannot initially be taken on the basis of trust, then enhancing sender credibility would have been especially relevant.12
In summary, the ability to anticipate counterarguments enables four basic kinds of argumentative strategies. The counterarguments may be pro-actively communicated in two-sided messages that allow the sender to either (i) rebut the counterarguments or (ii) minimize their importance. Alternatively, the sender can keep the counterarguments private at the outset and either (iii) spend additional time thinking of a good response in case the receiver raises them or (iv) avoid opening gambits that are likely to lead to the receiver raising them. Thus, if the function of argument-production is persuasion, prima
12 Interestingly, O’Keefe (1999, p. 230) found increased judgments of sender credibility for both refutational (r = .11) and non-refutational (r = .08) two-sided messages.
30 facie we should expect that in argumentative contexts people will be good at anticipating counterarguments to their views.
Instead, Mercier and Sperber claim their theory implies that people should “have trouble finding counterarguments to their favorite theories” (2017, p. 218). Although they never explicitly explain why this is, their rationale seems to be as follows. For senders, on average it was beneficial to offload the task of formulating counterarguments to the receiver and instead use the deliverances of the reason module, which specializes in producing myside arguments, to respond to those counterarguments. This offloading allowed senders to conserve resources that were used more productively for other tasks.
It also cost the sender little because “failing to convince one’s interlocutor right away carries little cost” (2017, p. 322 f.).
However, the cost-benefit analysis provided by Sperber and Mercier rests on an implausible picture of the relevant ancestral environments. Estimates of when syntactical spoken language emerged vary widely, from (very) roughly 500ky with archaic humans such as Homo antecessor or Homo heidelbergensis (de Boer, 2017; Fitch, 2017) to ca. 80ky with anatomically modern Homo sapiens (Berwick & Chomsky, 2016). For present purposes, it suffices to note that during this time span, and up until the Neolithic revolution ca. 12ky, humans were organized into nomadic band societies. A wide range of paleontological, archaeological, and ethnographic evidence suggests that bands had a fluid membership ranging from a handful to a maximum of around 50 people who camped together, divided labor, and shared food (Hill et al., 2011; Layton, O’Hara, & Bilsborough,
31 2012). In turn, each band belonged to a widely dispersed clan within which individuals could move freely, join a different band, or form a new band. Simulation models suggest that for a clan to be self-sustaining over multiple generations it needs to number at least
150, while for it to be socially cohesive in the absence of any hierarchical political authority it needs to number at most 500 (Dunbar, 2016; White, 2017).
As Mercier and Sperber (2017, p. 227) recognize, failing to initially state a strong case often carries a significant cost for the sender in terms of reputation. Senders who blurted out weak or easily objectionable arguments would have developed a reputation for incompetence. Given the small size of bands, in which every individual knew every other individual, having such a reputation may have seriously hampered future opportunities to persuade others.
Furthermore, in a wide variety of situations the iterated nature of the argumentative exchange would have been highly limited. Ethnographic studies consistently show that hunting groups keep verbal communication to a bare minimum so as not to alert prey (Hindley, 2015). More generally, any group that is away from the security of their base camp – whether they are hunting, scavenging, foraging for water or plants, or journeying to another camp – need to avoid making noises that will alert predators. At pivotal moments decisive leadership may be called for, as when the group is attacking or being attacked. Moreover, even in scenarios that offer both time and safety for a group, such as when a large number of people congregate around a camp fire, the iterative nature of the exchange will be limited by the participant’s energy and willingness
32 to engage. Tempers may flare, especially when there are disputes within a camp. In band societies, such disputes are often resolved not by argumentation but rather by people joining a different band for a period of time, until tempers cool (Lee, 1979, p. 372;
Turnbull, 1965, p. 106; Woodburn, 1982, p. 435).
Finally, even if there is time and energy for an iterative exchange between interlocutors, the responses from the receiver may be evasive. In competitive scenarios involving bargaining or arbitration – forms of communication which Sperber claims contributed most to the selective pressure for epistemic vigilance – the receiver may have an incentive not to “show their hand” and provide feedback that will aid the sender (cf.
Mercier & Sperber, 2017, p. 322). Furthermore, persuasion can be achieved through means other than argumentation. A sender’s argument may be met not with a counterargument but rather with ridicule, ad hominem attacks, browbeating, and so forth.
This is not to deny that there were ever iterated exchanges in which counterarguments were offered and responded to, leading to rational persuasion.
However, it appears to be a rare occurrence in contemporary hunter-gatherer societies, and there are no independent grounds to hold that in ancestral environments this kind of argumentation would have occurred sufficiently often to render myside bias adaptive.
Although it is tempting to imagine that life in small-scale band societies is inherently more communal than life in agrarian or industrialized societies, it should be borne in mind that contemporary hunter-gatherers spend much of their time alone. For example, to make foraging more efficient, band members typically hunt alone for hours or days, and form a
33 hunting party only if they spot suitable prey and require help tracking or killing it
(Marlowe, 2010, p. 227). Our evolutionary ancestors would thus likely have had ample time to think alone and, in particular, to anticipate objections to what they planned to say at camp.
For these reasons, I submit that the cost-benefit analysis provided by Sperber and
Mercier is empirically unwarranted. However, even if it is granted that it was adaptive for senders to offload the task of producing counterarguments to receivers, I will now show that the resulting explanation of myside bias faces two theoretical problems relating to information-gain and deception.
3.2 Information-Gain
Thus far, we have been considering the issue of myside bias from the perspective of a sender. But we can also look at it from the vantage point of a receiver. Suppose the primary adaptive function of argument evaluation is to decrease acceptance of costly mis- information (DEMANDING) and increase acceptance of beneficial information (OBJECTIVE).
This function of reasoning requires that argument evaluation be accurate. However, if receivers display myside bias in responding to senders then they will tend to pile up what they take to be confirming evidence for myside arguments and disconfirming evidence for otherside arguments. This problem is especially pressing because the psychology of reasoning literature on “disconfirmation bias” suggests that this kind of phenomenon – producing more reasons to favor than to disfavor myside arguments, and more reasons to disfavor than to favor otherside arguments – gives rise to a tendency to overrate the
34 strength of myside arguments and underrate the strength of otherside arguments
(Edwards & Smith, 1996).
Nevertheless, Mercier and Sperber claim that under their theory we should expect that when presented with an otherside argument, people tend to be good at formulating objections to it and bad at formulating defenses of it (Mercier & Sperber, 2017, p. 218).
They suggest that people intuitively evaluate arguments accurately, but subsequently search for further reasons to reject otherside arguments. If such reasons are found, this leads to a final all-things-considered rating of argument strength that is lower than their initial intuitive evaluation (Mercier, 2017, p. 109f.). The hypothesis that there is a two- stage process of evaluation is an empirical one, and ought to be empirically tested.
However, note that postulating a two-stage process is irrelevant with respect to
DEMANDINGNESS and OBJECTIVITY, since it the final all-things-considered judgment that determines whether the argument is actually accepted and hence whether the information bottleneck is eased.
Mercier and Sperber might respond that, for receivers, on average it was beneficial to offload the task of providing support for otherside arguments to the sender; and since the sender’s reason module was disposed to provide rebuttals to objections raised by the receiver, over time the receiver’s final evaluation moved from underrating the strength of
35 the otherside argument to accurately evaluating it.13 It might be wondered why receivers would outsource argumentative labor to senders who they believed may be misinformed or trying to deceive them. However, setting this issue aside, the main problem with the proposal is that it would lead to what I will call an adversarial spiral. Suppose I receive an otherside argument and am disposed to produce objections to it. In turn, my interlocutor will be disposed to either produce objections to my objections, or to produce new arguments for the original conclusion. In the former case, I will be disposed to object to their objections; in the latter case, I will be disposed to object to the new arguments.
And so on.
Sometimes argumentation does involve this kind of dynamic. Think of debates concerning identity politics, or, indeed, the kind of morally saturated and emotionally charged issues that participants are typically asked to think about in myside bias experiments. Notoriously, such debates tend to lead not to agreement but rather to polarization (Lord, Ross, & Lepper, 1979) and in extreme cases even to conflict (Kennedy
& Pronin, 2008). If this is how reasoned persuasion works in general, then it is misleading to claim that people are “ready to revise our ideas when presented with good reasons” and
“accept even challenging reasons” (Mercier & Sperber, 2017, pp. 332; 235). If argument-
13 This cost-benefit rationale is similar to the one discussed in section 3.1 and is subject to the same kinds of problems concerning the iterative nature of argumentation. However, these problems are perhaps less pressing for receivers.
36 evaluation evolved so that receivers are persuaded when and only when the sender simply runs out of things to say, then reasoning is spectacularly inefficient at overcoming the informational bottleneck.
Typically a receiver is in a better position than the sender to know whether the argument being offered coheres with their beliefs. Hence there may be supporting arguments that the sender is not in a position to make. The accurate evaluation of arguments is better served by receivers who are able to think of both the pros and cons of otherside arguments, rather than by receivers who operate with a disposition to defend their view alongside a blindspot resulting from myside bias. As with cases in which senders are aiming to persuade, it may not always be wise for vigilant receivers to voluntarily communicate every argument they think of. For example, perhaps they should refrain from explaining why they think the sender’s argument is good and instead have the sender focus on responding to their misgivings. Nevertheless, the ability to think of both myside and otherside arguments can break the adversarial spiral and make argumentation more efficient.
3.3 Deception
According to the rationalization-first view, argument-production evolved to facilitate not just honest communication but also deceptive communication. Indeed, this is a central assumption that underwrites Mercier and Sperber’s account of how reasoning evolved.
The benefit of persuasion for the sender is producing some effect in the receiver that serves the sender’s interests. Hence, to benefit from communication senders should (ceteris
37 paribus) “communicate the information most likely to produce the intended effect in the addressee, regardless of whether it is true or false” (Sperber et al., 2010, p. 360). An important aspect of the persuasion-vigilance “arms-race” was that the benefits of deception generated selective pressure for an ability to efficiently produce arguments in favor of views senders publicly endorsed but privately held to be unjustified or false. Yet,
Mercier and Sperber claim that under their theory we should expect people to be bad at finding arguments for views we don’t hold:
Finding arguments for a position we do not support, or even one we disagree with, is difficult. It takes skills and training. We may be lawyers, but only when it comes to defending beliefs and decisions we actually endorse. (Mercier & Sperber, 2017, p. 219f.)
Mercier and Sperber thus seem to face a dilemma: either it is not difficult to produce arguments with which to deceive people, which seems to run counter to their explanation of myside bias, or it is difficult to produce arguments with which to deceive people, which seems to run counter to a core component of their account of how reasoning evolved.
On Mercier and Sperber’s behalf, it might be suggested that although the reason module evolved to take genuine beliefs as inputs it can facilitate deception through a kind of pretense in which it is “tricked” into treating states of make-belief as genuine beliefs. In this vein Mercier and Sperber do allow that the tendency to rationalize one’s views can sometimes be overcome:
individuals may develop some limited ability to distance themselves from their own opinion, to consider alternatives and thereby become more objective … But
38 this is an acquired skill and involves exercising some imperfect control over a natural disposition that spontaneously pulls in a different direction. (Mercier & Sperber, 2011, p. 72)
However, regardless of how plausible this kind of model may be, invoking it only pushes the problem back a level: if it were correct then experimental participants who are tasked with thinking on both sides of an issue, or with providing arguments in favor of alternative views, should be able to use “make-believe” to spark the reason module into producing otherside arguments. The importance of deception in the evolution of reasoning, then, provides further grounds to expect that under the rationalization-first view people should be good at producing otherside arguments.
4. From Persuasion to Justification
According to the rationalization-first view, reasoning evolved under pressure from multiple factors including not just persuasion but also self-justification. Traditionally, persuasion has been treated as the main factor. Thus, it is natural to ask whether it is plausible that self-justification was the main factor and, if so, whether revising the rationalization-first view in this way can help it better explain myside bias.
The idea that self-justification often drives reasoning was proposed by Haidt (2001,
2012a) in his highly influential “social intuitionist” model of moral judgment. According to this model, reasoning is typically a post hoc practice of producing arguments that justify the sender’s initial intuitive judgments, so as to achieve social goals such as making a good impression and having smooth social interactions. Likewise, Mercier and Sperber (2017,
39 p. 8) now hold that the reason module evolved not just to provide myside arguments but also to provide what might be called “myside explanations” that favor the sender:
our goal is not to give an objective sociological or psychological account of our actions and interactions; it is to achieve beneficial coordination by protecting and enhancing our reputation and influencing the reputation of others … The explanatory use of reasons, we suggest, is in the service of its justificatory use… (Mercier & Sperber, 2017, p. 186)
Just as the adaptive function of the production of reasons qua arguments is persuasion rather than gaining the truth, under the rationalization-first view of reasoning the adaptive function of the production of reasons qua explanations is self-justification rather than gaining an accurate understanding.
No account of how this supposed trait developed has been provided in the literature.
Perhaps the relevant evolutionary history might be fleshed out as follows. The ability to construct and keep track of reputations facilitates mutual understanding and hence cooperation. Clearly, it is often detrimental for a person to have a bad reputation or an insufficiently established reputation. Moreover, reputations can be inaccurate. It can be especially detrimental for an observer to think that a “bad actor” is in fact dependable and benevolent towards them. In particular, the bad actor – through words or deeds – may be actively deceiving the observer about why they behaved as they did. With the rise of verbal communication, people developed an ability to overcome mistrust by providing explanations of their behavior. In turn, it was in the interests of observers to develop an
40 ability to accurately evaluate these verbally communicated explanations. This allowed observers to trust people they previously had a negative view of.
Assume for the sake of argument that such a module did develop. Recall that, according to Mercier and Sperber, if the function of reasoning is “to justify one’s actions or to convince others” then it “should display myside bias”. Presumably, the idea is that the person sending the explanations should be good at finding explanations they think justify their views and actions, but bad at finding explanations that they think undermine their views and actions. More generally, they should be good at finding explanations that undermine their rivals, and bad at finding explanations that justify their rivals.
Prima facie, the trigger conditions of a module dedicated to this kind of self- justification better correspond to the context of experimental reasoning tasks than the trigger conditions of a module dedicated to persuasion. In particular, myside bias paradigms typically ask participants to explain why they reached the conclusion they did, and participants are given no indication that their conclusion disagrees with the views of the researcher. It is more plausible that participants interpret the situation as one in which they should provide an explanation of their reasoning that maintains their reputation as rational and knowledgeable, than as one in which they should persuade the researcher to share their view.
Nevertheless, for an explanation to function as an effective self-justification the sender must persuade the receiver that the explanation is correct. Persuading a receiver need not involve sending an argument that the explanation is correct. Sometimes, the
41 mere act of communicating an explanation that coheres with the beliefs of the receiver will be enough to persuade them. Thus, a hunter who has spent a week with a different band might explain to their partner that they were tracking a herd of giraffes in the area. If this is a common and widely accepted practice, then their partner may simply accept the explanation.
However, this cannot be the kind of case which led the ability to reason to emerge and persist in the population. Under the rationalization-first view, the ability to produce explanations evolved not for accuracy, to help a receiver achieve understanding, but rather for self-justification, to overcome a trust barrier with the receiver. Moreover, the sender may in fact be deceiving the receiver. If these are the adaptive problems underpinning the evolution of the ability to reason, then in general senders should expect the receiver to have thought of alternative explanations. For example, the hunter’s partner might suspect that they visited the camp to keep in touch with a potential mate. In that case, the objections I developed in the section 3 will arise again, with “explanations” being substituted in for “arguments”. If the ability to reason evolved for senders to provide explanations that justify their views and actions, then it would be surprising if people are relatively bad at anticipating alternative explanations that undermine their reputation.
Furthermore, if an alternative explanation is cited by the receiver, then we should expect the exchange to lead to argumentation. In our previous example, the hunter might attempt to persuade their partner by arguing that the explanation is correct because no one had caught an animal for weeks and they need meat, or the hunter may argue that
42 the alternative explanation is incorrect because the potential mate is in a relationship with someone in the other camp. Thus, the objections in section 3 will return in their original form. The partner will be disposed to pile up objections to the hunter’s argument; in turn, the hunter will be disposed to pile up considerations in favor of their view; and so on. The functions of reason are better served if both parties do not suffer from myside bias.
5. The Inquiry-First View
In the evolutionary sciences, the standard way to make a case that a trait is an adaptation is to show that the trait exhibits hallmarks of “special design”. Although there is debate about the evidentiary standards this requires, there is widespread agreement that the trait should perform the relevant function with sufficient precision, efficiency, and economy to reasonably rule out chance and rival theories of how the trait evolved
(Andrews et al., 2002; Williams, 1966). However, due to genetic and environmental constraints, it is likely that no adaptation conforms to an optimal design for its function
(Dawkins, 2006). The human eye, for instance, has a blindspot because by historical accident the nerves of photocells protrude outwards towards the light and thus have to connect to the brain by looping back through a gap in the retina. Natural selection is a tinkerer, not an engineer.
It might be objected that all I have shown so far is that myside bias is not optimal for the kinds of functions posited by the rationalization-first view; but perhaps the ability to reason was initially based on a genetic mutation that linked, say, a language system
43 and a belief system, in such a way that only myside arguments were produced, and this sufficed for persuasion and self-justification. In itself, this is a fair point. However, recall that the empirical case for the rationalization-first view rests on the claim that it can explain myside bias whereas the inquiry-first view cannot. In the previous two sections my aim was to show that the rationalization-first explanation relies on empirically implausible auxiliary hypotheses, and that myside bias undermines other roles that reason is supposed to have under the view. In this section, my aim is to further argue that if the inquiry-first view is granted the same kinds of auxiliary hypotheses and the same leeway concerning non-optimality, then it can explain myside bias at least as well as the rationalization-first view.
In examining what a plausible inquiry-first view might look like, we should clearly distinguish between the adaptive function of reasoning and the context reasoning evolved to operate in. A theory according to which reasoning evolved for individuals (not groups) to attain epistemic goals may postulate either that this was mainly achieved through group deliberation, or that it was mainly achieved through solitary ratiocination.
Consider first a social version of the inquiry-first view. According to the rationalization-first view, the threat of being misinformed – mainly through deception – generated selective pressure for epistemic vigilance. However, perhaps epistemic vigilance was driven by the threat of being misinformed mainly through incompetence. Whereas deception is typically beneficial to the sender but costly to the receiver, undetected incompetence is typically costly to the receiver and the sender. Under the rationalization-
44 first view, the fitness of senders was increased because senders used arguments to manipulate receivers, for the attainment of socially-strategic goals. To achieve these goals the relevant argument merely had to be persuasive, regardless of whether it had a true conclusion. However, if reason evolved for the attainment of epistemic goals such as gaining true beliefs and avoiding false beliefs, then the production of arguments increased the fitness of senders not through persuading receivers tout court, but rather through persuading receivers when and only when the conclusion of the relevant argument was true.
Imagine a hunter claiming to their band that giraffes will be found to the north.
The band members are skeptical, so the hunter argues that there are plentiful trees to the north. In response, the band members point out that there are also trees to the east. The hunter then argues that there is also plentiful water to the north, but not to the east.
However, the band members object that the water is likely to attract lions, which the giraffes will want to avoid. That the mechanisms which underpin reasoning evolved a myside bias under these kind of conditions can be explained by exactly the same cost- benefit assumptions made by proponents of the rationalization-first view. Group deliberation enables a division of cognitive labor, in which senders offload the work of producing otherside arguments to the receivers. This outsourcing of labor to receivers makes more sense if there is a negligible risk that they are trying to deceive the sender.
Moreover, if reasoning evolved to attain epistemic goals, then there is no need to suppose that reasoners will be good at thinking up otherside arguments for the purposes of
45 deception; for example, that the hunter-gatherer should be able to readily produce arguments which persuade the band members that the giraffes are north, while privately believing that they are east.
However, the problem remains that accurate argument evaluation does not seem to be well served by myside bias. Moreover, the cost-benefit assumptions are subject to the same objections I outlined in the previous section. In band societies, the frequency of iterated argumentative exchanges would likely have been highly limited, and blurting out ill-considered claims or arguments would likely have had significant reputational consequences. If the risk of incompetence drove the evolution of reasoning, it is more plausible that people would have developed the ability to think well for themselves without relying on others.
Consider, then, an asocial version of the inquiry-first view in which the adaptive function of reasoning is the attainment of epistemic goals through solitary ratiocination.
Clearly, such a theory will avoid objections relating to the need to persuade others, whether honestly or deceptively. It may also avoid objections relating to the need to be good at evaluating arguments that receivers present verbally to senders. For why suppose that typically people would consciously “evaluate” their own arguments at all, let alone in a way that would transfer to the arguments that others communicate to them? It is factors such as the fluency of the private reasoning process which determine the degree of confidence the reasoner experiences with respect to any conclusion that is drawn (cf.
Ackerman & Thompson, 2017).
46 An asocial inquiry-first view might hold that reason evolved for the individual to gain true beliefs, but not to test for and correct false beliefs. For example, reason might be a mechanism that exploits cultural knowledge that has been transmitted by trusted members of the community, to come to novel conclusions when faced with situations for which the reasoner has no intuitions. In that case, there are no grounds to suppose that the reasoner should be good at formulating arguments that run counter to their beliefs.
However, even if we build into the asocial inquiry-first view the hypothesis that reason evolved for the individual to test for and correct false beliefs, myside bias can still be plausible explained. Clearly, an asocial theory of reason view cannot introduce the kinds of cost-benefit assumptions that accompany social versions of the rationalization- first and inquiry-first views. However, it can introduce similar cost-benefit assumptions.
First, myside bias might partially reflect an efficient stopping rule for deliberation. As with all psychological processes, reasoning must stop at some point, and when it does should be a function of the perceived importance of the issue and how satisfactory the conclusion arrived at is deemed to be. In his pioneering studies of myside bias, Perkins suggested that participants use a “makes sense epistemology” such that they judge the conclusion true in virtue of it making (often superficial) sense in the light of their beliefs:
Such a thinker’s reasoning is dominated by a strategy of cognitive load minimization … the first model we deliberately generate that makes sense often serves perfectly well. When it does not, and we are dealing with a situation in practical terms, we quickly discover that failing through experience. (Perkins, Allen, & Hafner, 1983, p. 187)
47 This makes sense epistemology concerns inquiry in the sense of discovery. It is a cost- effective means of initial belief fixation.
Myside bias might also serve to efficiently test the truth of preexisting beliefs. Given that the reasoner has such a belief, it will be cognitively demanding to think in terms of opposing hypotheses. A less demanding method for testing the truth of the belief is simply to try to find supporting arguments. Indeed, Koriat et al. (1980) found that when participants were asked to consider two alternatives by writing down reasons for and against each alternative, there was in both cases a bias towards arguments for (the researchers termed this a bias towards “positive evidence”). If such a search does not easily return compelling arguments, then the belief should be doubted (cf. Nestler, 2010;
Schwarz, 2004). At that point, inquiry should revert from testing to discovery, with the aim of coming to a new belief.
Note that this is not merely the thesis that people evolved to be cognitive misers
(Stanovich, West, & Toplak, 2016), including when they reason. The cognitive miser, when faced with a task, tends to save energy by defaulting to a cognitive mechanism with low-computational expense. For example, rather than engage their reasoning system, the cognitive miser might passively accept information when listening to their favorite talk show host, or rely on their intuitions when faced with a problem. However, the cognitive miser has the ability to engage cognitive mechanisms of high-computational expense when they are sufficiently motivated to do so. Mercier and Sperber point out that people display myside bias even when they are motivated to come to a balanced conclusion. Thus,
48 they suggest that the reasoning mechanism itself evolved a myside bias because on average the resources needed to produce otherside arguments was better used for other tasks. My suggestion is that a proponent of the asocial version of the inquiry-first view can make the same move, since it is more efficient to test a belief by checking whether there are good grounds to hold it, than to try to formulate arguments against it. Just as myside bias is not optimal for persuasion or information-gain, neither is it optimal for testing one’s beliefs. However, myside bias is satisficing for this proposed function of reason.
6. Concluding Remarks
I have argued that, contra Sperber and Mercier, the existence of myside bias fails to adjudicate between the rationalization-first view and the inquiry-first view. To explain the existence of myside bias, the rationalization-first view must invoke auxiliary hypotheses that are based on a cost-benefit analysis which is rendered implausible by paleontological, archaeological, and ethnographic evidence concerning human evolution. When granted the same or very similar cost-benefit assumptions, both social and asocial versions of the inquiry-first view can generate explanations of myside bias that are at least as plausible.
However, the rationalization-first and inquiry-first views are not the only options.
For example, it may be that the sort of reasoning at issue is an evolutionary by-product of adaptations for comprehension, learning, planning, problem solving, and the like. While these kinds of views may make predictions based on how the relevant exaptation is spelled out, they make no predictions based on the adaptive function of reasoning since they deny
49 that reasoning has an adaptive function. Furthermore, an inquiry-first view has yet to be worked out in the kind of detail that Mercier and Sperber provide for the rationalization- first view. I have some thoughts on how such an inquiry-first view of reasoning might go; but that is work for another day.
50 Chapter 2 Cracking the Enigma: Cultural Knowledge, Social Learning, and Private Reason
1. Introduction
In recent years, Dan Sperber and Hugo Mercier have formulated an important challenge for theories of human reason. On the one hand, reason seems to be a domain-general and multi-purpose “superpower” that sets our species intellectually and technologically far apart from the rest of the animal kingdom. Based on this idea, they claim, there is a long- standing tradition which takes the pursuit of truth and knowledge by lone individuals to be the modus operandi of reason:
Most of the philosophers and psychologists we talked to … see reason as a means to improve individual cognition and arrive on one’s own at better beliefs and decisions. (Mercier & Sperber, 2017, p. 330)
Call this an “inquiry-first” view of reason, since the attainment of epistemic goals involves attempting to ascertain the facts about the issue at hand.
On the other hand, scientific studies suggest that people display seemingly irrational biases when reasoning. For example, research on “myside bias” shows that
51 people are better at producing arguments for their views than against their views
(Perkins, 1989). As a result, they may become overconfident and reluctant to give up unwarranted beliefs (Koriat et al., 1980). This is problematic for inquiry-first views because if reason evolved for the attainment of epistemic goals such as gaining true beliefs and avoiding false beliefs, it appears to be ill-adapted to its primary function. Hence we are faced with the “enigma of reason” – reason is both a superpower and badly flawed.
Sperber and Mercier aim to crack the enigma by offering a new account of the nature of reason (what reason is) and the function of reason (what reason is for). First, they propose that the ability to reason is based on a garden-variety cognitive module – i.e. a domain-specific mechanism individuated by its evolutionary etiology – dedicated to processing representations of the form < P is a reason for Q >.14 Second, they propose that this “reason module” is triggered by the detection of disagreements or situations in which reputations are at stake, and evolved for the attainment of socially-strategic goals through the restricted purposes of public acts of persuasion and self-justification. That reason displays a myside bias can then be explained as a feature of the module that allows it to rapidly build a case in the reasoner’s favor. Hence, “rather than being a towering superpower”, reason “fits quite naturally among other human cognitive capacities and, despite apparent evidence to the contrary, is well adapted to its true function” (Mercier &
14 I use angled brackets to denote representations. Mercier and Sperber (2017, p. 81) characterize representations by their information-bearing functional role rather than their format or location.
52 Sperber, 2017, p. 5). Call this a “rationalization-first” view of reason, since acts of persuasion and self-justification are attempts to cite reasons in favor of a pre-existing belief or decision.
My aim in this chapter is to develop a new version of the inquiry-first view which cracks the enigma of reason while capturing the traditional idea of reason as a domain- general and general-purpose superpower that pursues truth. According to my account, the primary adaptive function of reason is to generate true beliefs through private deliberation. More precisely, I claim that reason evolved to take rule-like cultural knowledge gained through social learning, and use it to generate solutions to novel problems for which the reasoner has no intuitions. I argue that in comparison to rationalization-first views, my account better fits how human societies functioned in ancestral environments and better captures the empirical characteristics of reasoning.
Here’s how I proceed. I begin (section 2) by reconstructing Mercier and Sperber’s theory, according to which the capacity to reason evolved to overcome an information bottleneck in verbal communication that mainly resulted from the threat that senders of messages were trying to deceive receivers. In response, I appeal to an important kind of evidence that is strangely absent in Mercier and Sperber’s work, viz., paleontological, archaeological, and ethnographic evidence concerning the conditions that existed in ancestral environments. I argue that given the kinds of small-scale, egalitarian societies that existed during the era in which verbal communication evolved, selective pressure for a module dedicated to persuasion and self-justification didn’t exist because the risk of
53 deception was negligible. I further argue that a module adapted for the kind of public argumentative reasoning required for persuasion and self-justification is fundamentally ill-suited to give rise to the kind of private inquisitive reasoning required for reaching a previously unconsidered conclusion. This is significant, I claim, because what underwrites reason’s status as a domain-general and multi-purpose superpower is the ability to apply cultural knowledge to novel problems concerning which we have no intuitions. It is this ability that enables science and engineering, for example, as well as a host of other distinctly human achievements.
Next, I develop a new inquiry-first account of the nature and function of reason
(section 3). Based on ethnologies of hunter-gatherer societies, I provide evidence that the biological fitness of individuals was enhanced by the capacity to use cultural knowledge attained through social learning mechanisms, to understand the environment and guide decision making. I argue that the reasoning system is triggered by the detection of novel problems for which the individual has no intuitions, and takes as input cultural knowledge in the form of rules. The function of the reasoning system is then to generate true beliefs that allow the reasoner to better understand and solve the problems they are faced with.
Since the relevant rules can be about any subject matter, reason is domain-general. And since the relevant problems can concern any issue, reason is multi-purpose. Thus, I provide a naturalistically acceptable grounding for the traditional idea of reason as a superpower. Furthermore, my account recaptures argumentative reasoning, in the sense
54 that persuading an audience and defending one’s reputation are problems that may trigger inquisitive reasoning.
In section 4, I sketch an empirical case for my account. I mimic the strategy used by Mercier and Sperber, which appeals to the evolutionary etiology and adaptive functions of reason to explain the kinds of biases that reason displays. Under my account, although reason evolved to generate true beliefs about novel problems, it did not evolve for the testing and correction of mistaken beliefs. Moreover, since under my account reason is responsive to anything deemed a problem, it should be a highly flexible system that may be put to work in the service of a wide variety of motives, including motives to rationalize one’s beliefs. Thus, I claim that my account is neutral with respect to the existence of myside bias. However, I argue that my account better explains the continued impact of discredited beliefs (belief perseverance). For, contra Sperber and Mercier, belief perseverance is not driven by the generation of arguments that support the belief (myside bias), but rather by the generation of explanations based on the belief. This is what we should expect if reason evolved not for persuasion, but rather to use cultural knowledge to generate true beliefs which help the reasoner solve problems in the natural environment.
2. The Rationalization-First View of Reason
2.1 Exposition of the Theory
Sperber and Mercier’s account of reason is based on a theoretical reconstruction of the evolutionary history of verbal communication. Communication allows animals to share
55 information and manipulate the behaviour of others. For a communication system to be evolutionarily stable in a species, on average the benefits of both sending and receiving messages must outweigh their costs (Dawkins & Krebs, 1978). By sharing information, individuals can facilitate cooperation in the pursuit of complementary goals. However, communication is vulnerable to the spread of misinformation. This occurs when a sender is mistaken or purposely misleads the receiver.
Sperber and colleagues propose that during the early stages of the evolution of verbal communication, the risk of receivers being misinformed by senders generated selective pressure for cognitive mechanisms dedicated to “epistemic vigilance” (Sperber,
2001a; Sperber et al., 2010). Epistemic vigilance has two components. First, the receiver may evaluate the trustworthiness of the sender, based on cues relating to competence and honesty. Second, the receiver may evaluate the plausibility of the communicated claim, based on how the content coheres internally and with respect to a relevant subset of the receiver’s pre-existing beliefs.
Here’s how reason enters the picture (Mercier & Sperber, 2011; Sperber, 2001a).
To the extent that epistemic vigilance blocked the transmission of misinformation, it was costly to senders who aimed to deceive receivers. Moreover, to the extent that epistemic vigilance blocked the transmission of accurate information, it was costly to both senders and receivers. Due to this information bottleneck, senders were subject to selective pressure for the ability to manipulate epistemic vigilance by pre-emptively presenting
56 arguments for their claims, using premises the receiver either already accepted or accepted more readily than the claim in isolation.
The introduction of arguments to verbal communication stimulated the flow of information. However, it also enabled senders to misinform receivers in a more sophisticated way, through specious arguments. As a result, receivers were subject to selective pressure for the ability to accurately evaluate arguments. In turn, this led to a sharpening of the sender’s ability to produce persuasive arguments. And so on. Thus, the two abilities – to produce and evaluate arguments – are mutually adapted.
These argumentative practices are held to be implemented by an evolved cognitive module (Mercier & Sperber, 2009). Modules are autonomous computational devices that automatically perform specific tasks on a restricted range of domain-specific inputs
(Sperber, 1994, 2001b). Although people are not conscious of how modules process information, they are conscious of the outputs of some modules. Furthermore, a conscious output may also be intuitive in the sense that it is accompanied by a feeling of confidence that it is correct, despite the person being unaware of why they take it to be correct.
However, due to the human capacity for meta-representation, people also have intuitions about intuitions, including intuitions about the considerations that justify their intuitions. For Sperber and Mercier, reasoning is the meta-representational process of drawing a conclusion by consciously attending to (purported) reasons in favor of the conclusion. Suppose a sender has an intuitive belief that Q, and wishes to persuade a skeptical receiver that Q is true. This situation triggers the reason module to search for
57 evidence that supports the message the sender wants to communicate (2017, 289). If successful, it will output an intuition that P is a reason for Q , which can then be used as an argument for Q. If either the sender or the receiver accepts Q on the basis of the argument, they will have a “reflective” or “reasoned” belief that Q rather than an intuitive belief that Q, since they will be aware of why they take Q to be true.
In recent work, Mercier and Sperber (2017, 130) propose that the reason module also underwrites the ability to justify ourselves. The rationale is that any argument for the correctness of a belief or action can be used as a justification for holding the belief or performing the action, and vice versa. Having a good reputation increases biological fitness because it facilitates cooperation, especially with non-kin in the pursuit of long-term goals.
Just as the adaptive function of producing arguments is persuasion rather than inquiry, the adaptive function of producing explanations is not accuracy but rather self- justification.
In summary, here’s what Mercier and Sperber tell us about how the hypothesized reason module operates. We have a passing sense of the inputs, outputs, and trigger conditions. According to Mercier and Sperber (2017, 280), the proper domain of reasoning is “disagreements between oneself and others”. Thus, the reason module is triggered by the “detection of a clash of ideas”. Furthermore, the module is also triggered by the detection of situations in which reputations are at stake (2017, 123). Representations of the form < P is a reason for Q > can be verbally communicated as arguments to persuade or as explanations to justify oneself. Once activated, the module either takes < Q > as an
58 input and outputs < P is a reason for Q > as an intuition, or takes < P is a reason for Q > as an input and outputs an evaluation of it as an intuition (2017, 148). The capacity to reason, then, derives from a module that outputs intuitions in the same manner as other modules that make up our animal minds, except that the reason module outputs intuitions about representations. Moreover, since this module is “a cognitive mechanism aimed at justifying oneself and convincing others” (2017, 10), it will display a myside bias. Thus, the enigma is solved. Or so it would appear.
2.2 Criticism 1: The Paleo-Anthropological Context
As we have seen, Sperber and Mercier’s reconstruction of the environment of evolutionary adaptiveness for reason is based on an informal game-theoretical model of how verbal communication developed. According to the model, the major factor that drove the evolution of reason was not the risk that senders were mistaken, but rather the risk that senders were engaging in deception:
the major problem posed by communicated information has to do not with the competence of others, but with their interests and their honesty. While the interests of others often overlap with our own, they rarely coincide with ours exactly. In a variety of situations, their interests are best served by misleading or deceiving us. It is because of the risk of deception that epistemic vigilance may be not merely advantageous but indispensable if communication itself is to remain advantageous. (Sperber et al., 2010, p. 360)
The model does not presuppose any particular account of how the capacity for language originally emerged and persisted in the hominid line. For example, it is compatible with
59 the theory that language initially evolved to enable complex inferential transitions in inner thought (Berwick & Chomsky, 2016; Fitch, 2011) or community bonding through stories, songs, jokes, and the like (Dunbar, 2017). However, it does presuppose that at some point in the evolution of language the verbal exchange of information about the world was sufficiently fitness-enhancing that epistemic vigilance and argumentation became adaptive.
Sperber and Mercier do not provide an analysis of the kinds of interactions which supposedly gave rise to the relevant selection pressures, beyond suggesting that bargaining and arbitration are prime examples:
Communication among people, in particular close kin, on matters where their interests are aligned does not require epistemic vigilance towards the risk of deception in order to be advantageous to all participants. On the other hand, several typical forms of human communication such as bargaining or arbitration could hardly be practiced in a manner advantageous to all in the long run without vigilance. (Sperber, 2013, p. 63)
However, as I will show, this proposal faces two major problems in the light of our best theories about how human societies functioned in ancestral environments. First, practices such as bargaining and arbitration are unlikely to have given rise to an informational bottleneck that became an adaptive problem. And, second, even if such an adaptive problem did arise, it is more plausible that it was solved by the development of moral norms and the threat of punishment.
60 Paleontological and genetic evidence suggests that our species (Homo sapiens) emerged from Homo erectus, or descendants of Homo erectus such as Homo antecessor and Homo heidelbergensis, in Africa, by at least 300–200ky (Brown et al., 2012; Richter et al., 2017). Populations of Homo sapiens gradually expanded beyond Africa, with the main dispersal occurring at 70–60ky and Australia being reached by boat around 50–40ky at the latest (Nielsen et al., 2017). Estimates of when syntactical spoken language emerged vary widely, from 500ky with Homo antecessor or Homo heidelbergensis (de
Boer, 2017; Fitch, 2017) to 80ky with anatomically modern Homo sapiens (Berwick &
Chomsky, 2016). During this period of the Paleolithic era, before the 12ky Neolithic revolution which introduced agriculture and permanent settlements, humans were organized into nomadic band societies (Grove, Pearce, & Dunbar, 2012).
A wide range of paleontological and ethnographic evidence suggests that bands had a fluid membership ranging from a handful to a maximum of around 50 people who camped together, divided labor, and shared food (Hill et al., 2011). In turn, each band belonged to a widely dispersed clan within which individuals could move freely, join a different band, or form a new band. Indeed, bands may have evolved within pre-existing clans as crystallized “fission-fusion parties” of the kind seen within chimpanzee societies, with the entire range of the community accessible within a day’s walk (Layton et al., 2012).
The fluidity of band membership within the clan allowed individuals to buffer local resource scarcity and find spouses. Simulation models suggest that for a clan to be self-
61 sustaining over multiple generations it needs to number at least 150, while for it to be socially cohesive it needs to number at most 500 (Dunbar, 2016; White, 2017).
Ethnographic studies of 21st century hunter-gatherer populations report that they are egalitarian (Woodburn, 1982). Since resources are not stored in nomadic societies, opportunities for greed and theft rarely arise. Individual hunts are usually unsuccessful, so whenever meat is procured it is shared within the band. Gift exchanges are also used to establish and maintain relationships which may be critical for survival in the future
(Winterhalder, 1986). Movement between bands, which is commonplace and encouraged, serves as a mechanism for avoiding and resolving conflicts. For example, often individuals will simply announce where they are going to hunt or set up camp, with others then being free to make their own decision about whether to join or not (Woodburn, 1982, p. 444).
This freedom of movement helps to even out access to resources within the clan’s territory.
When disputes do break out within a band, typically one of the disputants will simply move to a different band until tempers calm or the issue loses importance (Lee, 1979, p.
372; Turnbull, 1965, p. 106; Woodburn, 1982, p. 435).
If deception is harmful to receivers within the band, then the sender runs the risk of being ostracized from the group. In Paleolithic hunter-gatherer societies, ostracism is likely to have been fatal (Marshall, 1976, pp. 196; 287). This is not to say that deception never occurred. However, many commonplace forms of deception benefit the receiver. For example, a sender might produce arguments to reassure a receiver that they are in the
62 right during a dispute, that they are a good hunter, and so forth. In these kinds of cases involving white lies, it harms the receiver to exercise vigilance.
Might long-distance trade have given rise to epistemic vigilance? Paleolithic trade networks are typically inferred from lithic site-to-source distances. For example, although obsidian has been used for flaked stone-tool manufacture for at least 1.7 million years, evidence for long-distance transfer of obsidian and other raw materials does not become widespread until 50ky (Ambrose, 2012). Assuming that the human ability to reason was in place before the main dispersal out of Africa 70–60ky, this counts against the hypothesis that the ability to reason was driven by long-distance trade. However, recently a 200ky site in Kenya was found to contain obsidian from three different sources located 25 km,
140 km and 166 km away (Blegen, 2017). A modern band typically ranges over a 20km radius annually, and foraging zones are estimated to have a 40km radius at maximum
(Brooks et al., 2018; Gamble, 1993). This tentatively suggests that during the early history of anatomically modern Homos sapiens there were long-distance trade networks.
However, economic transactions are likely to have been infrequent, and materials such as obsidian, pigments, shells, beads, clothes, and so forth, carry their value on their face. Whereas a trader might know that an animal comes from a sickly stock, or that a complicated machine has a flaw, there is nothing to conceal about simple material objects.
Of course, in any exchange promises of future payment may be subject to doubt. But it is unclear why the exchange of material objects – prior to the rise of more sophisticated
Neolithic economies based on herding livestock, plant agriculture, and land ownership –
63 would often have involved deferred payments. Thus, there is little reason to suppose that
Paleolithic trade gave rise to doubts concerning the accuracy of communicated messages, to an extent that significantly impacted fitness.
At first glance, interactions that provided opportunities for self-justification are easier to imagine. For example, foragers might creatively explain away unsuccessful hunts, or boastfully recount the decisions that led to successful hunts. Having a good reputation is likely to have been fitness enhancing in ancestral environments.
Ethnographic studies of band societies, in which people have few personal possessions, have found that character traits such as honesty and generosity are less important than traits such as being a hard worker and a good forager (Smith & Apicella, 2019). Men and women with better reputations as foragers have more friends and are more preferred as camp co-residents than people with poorer reputations (Marlowe, 2010, p. 251). Moreover, the trait that women most value in a potential mate is that of being a good forager, and men who have better hunting reputations have more friends, younger wives, and more children, than men with poorer reputations (Marlowe, 2010, p. 215). Since food is shared in band societies, a good hunting reputation may primarily function as a sign of health and intelligence, and hence overall mate quality.
Nevertheless, within the clan everyone knows each other and bands can communicate relatively easily. Hence, reputations would likely have been well grounded in personal experience and difficult to manipulate. For example, in the ethnographic studies cited above, men with better hunting reputations had higher overall food returns
64 and higher hourly return rates than men with poorer reputations (Marlowe, 2010, p.
215).15 Experiments have shown significant positive correlations between hunting reputation and skills required for hunting such as aim, upper-body strength, and the ability to recognize animal vocalizations (Stibbard-Hawkes, Attenborough, & Marlowe,
2018). Moreover, often a man who is interested in a woman as a potential mate is required to live with her band for a period, and if they do not prove themselves as a hunter they will not be accepted (Marshall, 1959). Within small band societies, experience speaks for itself. Thus, despite the likely importance of reputation in ancestral environments, there is no evidence that self-justification would have had a significant impact on fitness.
In summary, given our best theories of the kinds of conditions that held in ancestral environments during the evolution of verbal communication, there was no selection pressure for the evolution of a reason module dedicated to persuasion and self- justification. The risk of deception may have been more pressing in the Neolithic world of permanent settlements and agriculture, than in the Paleolithic world of nomadic bands and foraging. However, the Neolithic period began around 12ky, which is long after the main exodus out of Africa, and is in any case too recent to allow sufficient time for the development of a sophisticated pan-cultural cognitive capacity such as reason.
15 However, hunting success is not easy to estimate. For example, Hill and Kintigh (2009) calculate that if hunters acquire meat on < 27% of hunting days, a sample of > 600 hunting days is required to estimate a hunter’s mean daily acquisition rate with ± 20% accuracy.
65 2.3 Criticism 2: The Scope of Reason
I aim to have shown that there is little empirical warrant for supposing that a module evolved for public practices of persuasion and self-justification. But even granting that such a module exists, it is unclear how it alone can account for the wide range of private reasoning practices that humans seem to engage in.
Mercier and Sperber (2017, p. 168) suggest that the reason module came to be triggered by the presence of conflicting intuitions within an individual, as well as mentally simulated disagreements with an imagined audience. As they point out, it is commonly allowed that the trigger conditions of modules only imperfectly capture their original target, and sometimes mismatches will be advantageous. For example, a module that evolved for the detection of snakes may become biased towards false positives, such as being triggered by the sight of an exposed tree root. In a similar fashion, the benefits of solitary argumentative reasoning may have biased the reason module towards being triggered by doxastic disagreements even in the absence of an actual interlocutor.
More problematic is whether the reason module can allow for inquisitive reasoning, where there are no initial intuitions to rationalize. Mercier and Sperber appear to treat such cases as a kind of problem-solving which involve the application of step-by-step methods, rather than reasoning as such:
This ability to understand and apply step-by-step methods is a hugely important aspect of human psychology. It plays a major role in the development and transmission of cultural skills, including specialized problem-solving skills. The psychologist Keith Stanovich has argued that the ease with which people acquire
66 and apply such methods (or “mindware”) correlates with individual differences in rationality. Still, reasoning doesn’t consist in applying such methods, and in general, it doesn’t need them. (Mercier & Sperber, 2017, p. 171)
According to their definition, reasoning tout court consists in “attending to reasons for adopting new conclusions” (Mercier & Sperber, 2017, p. 52). But note that “new conclusion” is ambiguous. The reason module is a mechanism for backwards (abductive) inference from an intuition that Q to an intuitive argument for Q. If the argument is judged to be good, then Q may be accepted on those grounds, in which case it will be held as a newly formed reflective belief (Mercier & Sperber, 2009). However, no account has been provided of how this mechanism can be exapted for forwards inference, in the absence of any intuition Q, to a previously unconsidered conclusion.16
This is important because, prima facie, it is inquisitive reasoning, rather than argumentative reasoning, which is responsible for the “superpower” that gives rise to the enigma of reason. There is a growing recognition that what sets the human species so far apart from other species is our heightened capacity to learn not just from our observations and interactions with the environment, but also from each other (Heyes, 2018; Tomasello,
2001). This “social learning” enables the creation and transmission of culture, and often gives rise to a “ratchet effect” such that successive generations improve on the work of previous generations, resulting in artifacts that are too sophisticated to be invented whole-
16 A similar point is made by Evans (2019); see also Sterelny (2018, p. 5f.).
67 cloth by any single community (Tennie, Call, & Tomasello, 2009). Arguably, we are now totally reliant on cultural knowledge for our survival:
During eons of relying on large bodies of cumulative cultural knowledge, our species became addicted to this cultural input … Neither our intelligence nor domain-specific psychological abilities fire up to distinguish edible from toxic plants or to construct watercraft, bone awls, snow houses, canoes, fishhooks, or wooden sledges. (Henrich, 2015, p. 33)
Social learning allows the individual to go beyond associative inferences based on personal experience, by reasoning to novel conclusions based on a consideration of the normative implications of cultural knowledge that has been gleaned from others. This process was supercharged by the development of writing systems around 6-5ky, which allowed humans to transmit vast amounts of information and led to religion, business, politics, and science as we know them today (Bottero, Herrenschmidt, Vernant, & Zabbal, 2000).
This role of reasoning dovetails well with conceptions of reasoning that are influential in recent philosophical discussion – what might be called “intentional rule- following” theories of reasoning (Boghossian, 2014). Such theories make a pair of commitments. First, they suppose that reasoning essentially involves following rules concerning the premises from which one reasons. In addition, they suppose that all rule- following involves (conscious or unconscious) intentional states which represent the rules being followed. The generality of rules explains why inferences often exhibit similar patterns across cases. Moreover, since the rules are represented, we can combine them and even use the rules themselves as premises. Finally, that rules are followed helps
68 explain the sense in which reasoning is a person-level activity that is subject to social and rational norms, since the reasoner can err by either failing to correctly follow a rule or by correctly following an inappropriate rule.
Although this is not the place to defend intentional rule-following theories of reasoning, it is worth noting that they can naturally model the application of cultural knowledge to problems for which we have no intuitions: the premises refer to the particular situation, and the rules are supplied by cultural knowledge gained through social learning. Thus, the conception of inquisitive reasoning outlined above has independent merit. In the following section, I provide an account of how the capacity for this kind of reasoning might have evolved.
3. A New Inquiry-First View of Reason
My suggestion is that reasoning evolved to put cultural knowledge to use. Since the kind of reasoning we are interested in involves attending to explicitly represented, verbalizable, reasons to draw a conclusion, prima facie we should expect the reasoning system to primarily draw on rule-like cultural knowledge that is linguistically transmitted. If my account of the function of reason is correct, then we should expect to see this kind of knowledge transmission in the educational practices of contemporary hunter-gatherers.
However, perhaps surprisingly, ethnologies of modern hunter-gatherer societies tend to conclude that there is scant evidence for overt teaching (Boyette & Hewlett, 2018;
Garfield, Garfield, & Hewlett, 2016; Lew-Levy, Reckin, Lavi, Cristóbal-Azkarate, & Ellis-
69 Davies, 2017; MacDonald, 2007). Instead, these studies find that children and adolescents mainly learn through self-guided observation and play. Technical skills such as digging roots, food preparation, tool manufacture, and weapon use, are gained through imitation.
These skills are honed through role-playing games, in which animal or adult human behavior is mimicked. For example, a study of the !Kung (Ju/’hoan) hunter-gatherers of
East Africa reports that ecological knowledge and tracking skills are gained with “little formal instruction” from adults:
Play hunting can begin as early as age 3 … Little girls participate in these games too, and frequently a play group of four to six kids pretends to cut up, cook, and serve an imaginary animal in the same way that North American children have an imaginary tea party … tracking skills, however, are acquired through the on- going study of nature as the young boys and girls learn to identify the hundreds of plants and animals in their environment. Studying animal tracks is a major pastime of the older boys. (Lee, 1979, p. 236)
Older children are brought on foraging expeditions by their parents, and while verbal instructions are sometimes given during these trips, they tend to be limited and brief
(Hewlett, 2016). Typically, parents will simply use gestures to direct their children. For example, a hunter may point to a track, and perhaps state something about the animal that made it, but will not explain how that conclusion was arrived at (MacDonald, 2007).
Moreover, although theoreticians commonly suppose that cooperative hunting requires verbal planning and commands, actual strategies used on the ground – such as hunters spreading out into a large circle and then progressively tightening it, while making loud
70 noises to direct prey towards the center – tend to be intuitive and can be coordinated through gesture (Coon, 1971; Endicott, 1988, p. 123).
Clearly, hunter-gatherer technologies and practices have likely changed significantly since language began to evolve. For instance, although there is evidence for hunting spears at 400ky (Dennell, 1997; Thieme, 1997), the earliest known evidence for a bow and arrow is at just 71ky (Brown et al., 2012). Until quite recently in our evolutionary history, Homo sapiens have relied on stone tools, and these may have required a greater degree of coordination and cooperation to manufacture and wield. Indeed, it has been suggested that the reliance of humans on lithic technologies created selection pressure for the evolution of language as a means to transmit the relevant manufacturing skills and knowledge (Laland, 2018, p. 200). Might manufacturing stone tools have required overt teaching?
The earliest known lithic technologies associated with the Homo genus – Oldowan cores and flakes – date to 2.6my and were used as axes, hammers, knives, and scrapers
(Leakey, Tobias, & Napier, 1964; Semaw et al., 2003). These tools were often produced using the freehand percussion method, in which a stone core is held in one hand and struck with a stone hammer. A major advance in sophistication came with the introduction of
Acheulian bifacial hand axes, cleavers, and picks, the earliest known examples of which date to 1.7my (Diez-Martín et al., 2015). These were much larger than Oldowan tools and featured neatly worked symmetrical sides, which resulted in sharp edges and points.
Nevertheless, Oldowan and Acheulian techniques are alike in treating flakes as a useful
71 by-product of shaping the core. The next technological advance came with the introduction of Levallois prepared core techniques, in which the core itself is treated as a means to an end, being carefully worked in multiple stages so that a thin and sharp flake of predetermined size and shape can eventually be detached from it. These prepared core techniques led to a decline in handaxes and their replacement by points and blades which could be used to make composite tools such as spears. Importantly, the emergence of
Levallois tools coincides with the emergence of archaic Homo sapiens, with the earliest
Homo sapiens fossils being found alongside both Acheulian and Levallois tools (Richter et al., 2017; Tryon, McBrearty, & Texier, 2005).
A number of experimental studies have focused on whether language, or at least some form of proto-language, is necessary to transmit the skills and knowledge required for the manufacturing of stone tools. Ohnuma et al. (1997) and Putt et al. (2014) investigated the production of Levallois flakes and Acheulean flakes respectively, and found no statistically significant differences between gestural instruction and verbal instruction. Morgan et al. (2015) and Lombao et al. (2017) investigated the production of
Oldowan and Acheulean bifaces, and found that both gestural instruction and verbal instruction were superior to mere imitation and emulation. However, although they also reported that verbal instruction was superior to gestural instruction, these differences did not reach statistical significance. Finally, noting that previous studies included gesture in their verbal instruction conditions, Cataldo et al. (2018) investigated the production of
Oldowan and found that speech alone is inferior to gesture alone and to gesture-plus-
72 speech. In summary, all five studies found that stone tools can be manufactured without verbal instruction, and any facilitation of performance through verbal instruction is dependent on gesture.
Although lithic technologies do not appear to require language, they do require a host of other cognitive capacities that are highly relevant to inquisitive reasoning. For example, knapping requires the flexible use of strategies held in long-term memory, which guide action routines and subroutines (Alperson-Afil, Goren-Inbar, Herzlinger, & Wynn,
2020; Putt, Wijeakumar, Franciscus, & Spencer, 2017). Prepared core techniques in particular require the knapper to form a plan, carefully monitor the core while keeping in mind the intended form of the flake, and make knapping decisions that will have implications several steps ahead in the process. Thus, it has been argued that Levallois tool manufacture resulted in selective pressure for an increase in working memory capacity and the development of a long-term working memory system, both of which underlie the capacity for modern day expert cognition and performance (Wynn & Coolidge,
2011).
Like a knapper, a reasoner may work towards a goal by retrieving from long-term memory strategies which guide step-by-step procedures for achieving sub-goals (Lee &
Johnson-Laird, 2013; Newton & Roberts, 2005; Schaeken et al., 1999). To take a simple example, suppose your goal is to find the sum of the digits from 1 to 100. Here we have no intuitions, and must rely on the arithmetical knowledge we were taught in school. One strategy is to add 1 and 2 (a sub-goal), store the answer in working memory, then add 3 to
73 the previous answer (another sub-goal), store that answer in working memory, and so on.
An alternative strategy is to break the numbers up into 50 pairs which each add to 101 – i.e. (1 + 100), (2 + 99), (3 + 98), (4 + 97), and so on – and then multiply 50 by 101, perhaps by multiplying 50 by 100 and then adding 50. Or to take a more practical example, consider the choice method known as “elimination by aspects” (Tversky, 1972). Suppose you are choosing which car to purchase. You select the characteristic of cars that is most important to you, such as automatic transmission, and eliminate all cars which do not possess this characteristic. Next you select a further characteristic, such as a maximum price of $30k, and eliminate all cars which do not possess that characteristic. And so on, until one car is left. Both everyday and expert reasoning make use of a myriad of such consciously deployed strategies, which can be regarded as involving sequences of mental actions performed to solve practical and theoretical problems.
However, this leaves unresolved the issue of how Homo sapiens came to possess the ability to arrive at conclusions by attending to the normative implications of linguistically transmitted cultural knowledge. Thankfully, all is not lost for my account of reason. For although there is no evidence for overt teaching in hunter-gatherer societies, either past or present, previous studies have overlooked the pedagogical significance of stories (Scalise Sugiyama, 2017). For example, a study of the Chabu forager-farmers of
Ethiopia found that adolescents learned ecological and hunting knowledge by listening to stories told to them by their fathers (Dira & Hewlett, 2016). Likewise, Lee reports that
!Kung adolescents learn about hunting from stories:
74 Before they actually go on a hunt, !Kung boys have listened to dozens of hunts described in the minutest detail in the storytelling around the campfire. This is a major component of their socialization as hunters. This vast body of knowledge is a treasure house of lore and information about animals and how to kill them. And the boys listen intently. (Lee, 1979, p. 236)
This is corroborated by Marshall (1976, p. 130), who notes that the !Kung “recount over and over memorable episodes of past hunts”.
Wiessner (2014) collected conversations among the !Kung and found that the overwhelming majority of stories were told at night around a campfire, with 81% of all night talk being dedicated to stories. The earliest evidence for regular controlled use of fire dates to 400–300ky (Sandgathe & Berna, 2017), which is just prior to the earliest evidence for Homo sapiens. The use of fire for cooking reduced the need for chewing and digestion, resulting in the evolution of smaller teeth, guts, and intestines (Wrangham & Conklin-
Brittain, 2003). Eating cooked food also reduced the risk of bacterial infection and increased net energy gain, which facilitated brain expansion (Carmody & Wrangham,
2009). Fire is a source of heat, which facilitated the movement of humans into colder regions outside of Africa. Fire is also a source of light, which extends the number of active hours of the day. However, since fire does not provide enough light for work, most of this extra time is taken up with social activities, such as singing and acting. Although gesture- based communication across a flickering fire may have led to the development of artistic performances, it would have been inadequate for sustained sharing of information. Thus,
75 it has been suggested that verbal communication evolved to enable group conversation around the campfire (Dunbar, 2014).
Since stories are often a source of fun and entertainment, it is easy to overlook their impact on biological fitness. First, as we have seen, stories transmit ecological and technological knowledge. Of course, the relevant knowledge need not be restricted to hunting. Consider the following example, taken from the !Kung stories transcribed by
Wiessner (2014).17 In this tale about a trip to a neighboring camp, a recipe is described:
the people fetched water, fetched water and said: “Why don’t you people look for look for the eland chyme, why don’t you look for it so you can cook it into porridge?” Early the next morning when the sun had not yet risen, they fetched water, fetched water, brought water. Some put the pots on the fire and while others prepared the chyme. They put eland fat into small tortoise shells, put it in there, and poured hot water into the shell to melt the fat. (Wiessner, 2014, supplementary material)
Stories also transmit knowledge about social norms and practices. Wiessner (2014) found that most !Kung stories focused on marriage and kinship. Similarly, Smith et al (2017) collected stories across seven different hunter-gatherer societies and found that 70% pertained to prescribing social norms and coordinating behavioral expectations (39% dealt with natural phenomena, while 34% dealt with resource use). A familiar theme of such
17 The examples cited in this section are selected from the very small sample of stories that have been translated and published. Although they may not be the best examples of knowledge transmission, it is important to remain wedded to actual practices when possible.
76 stories is that people who are not cooperative and peaceful suffer negative consequences, including being ejected from the community. For example, in the following story a woman leaves her poorly behaved husband for another man:
After she kicked him out she went away to the west. She went and married a man called Ku 'oan who was short and nicely built and who could kill an eland now and then. He would kill an eland now and then, and he would get a bag of ostrich eggshells and bring them to his wife. She went and choose that man herself. She, /Xoan, choose him all by herself and married him. When she married him her parents immediately agreed: “Yes”. People were living on the fat eland meat and Ku... did not beat or mistreat his wife. (Smith et al., 2017, supplementary material)
Knowledge of social norms and practices promotes interpersonal cohesion within the band.
Moreover, since stories often feature people in other bands, they may strengthen social bonds within the extended clan by creating a “virtual community” that exists in the imagination.
This transmission of knowledge through the medium of language supplies the materials for reasoning. For example, consider the following report from a !Kung camp:
In discussing fighting of kudu, one man described the sound one hears of their horns clashing, and how, if one hears the sound, one can approach to shoot them. The same man, a very enthusiastic and busy hunter also described how one tracks infant kudu, showing that it sleeps away from its mother, and that one can follow it to where it is hidden, and kill it by hitting it. (Jones & Konner, 1976, p. 340)
77 This is not information about a particular kudu, but rather rule-like information about all kudu. Such rule-like information can be consciously deployed as a heuristic to guide action.
Suppose you discover the tracks of an infant kudu while foraging, and wonder whether they are worth following. You may recall the elder telling you that infant kudu sleep away from their mothers. This knowledge, which takes the form of a generic, functions as a default rule, and you conclude that this kudu will sleep away from its mother. Hence, you decide that you should follow the tracks.
I agree with Mercier and Sperber that reasoning involves attending to reasons for drawing a conclusion. Thus it is worth noting that the ability to create and understand stories, which interpret the world in a reflective way, facilitates an awareness of something being a reason. For example, in a tale of moving to another camp, one man shares that he and his host acted as they did because they were scared:
He put our things down, because we were afraid of each other. He was afraid of me and I was afraid of him. (Wiessner, 2014, supplementary material)
Stories explain why people act as they do. Understanding that a mental state can be a reason for bodily action provides some basis for eventually understanding that a belief can be a normative reason to perform the mental action of accepting a conclusion.
Stories transmit knowledge indirectly, through narrative explanations that entertain the audience. Smith et al (2017) found that skilled storytellers were almost twice as likely as unskilled storytellers to be chosen as campmates, and had an additional 0.53 living offspring compared to non-skilled storytellers. However, in discussions with the
78 !Kung, Jones and Konner (1976) found that storytellers did not take license with the facts and readily admitted their ignorance. Since stories are shared with campmates, including close kin, in general it is not in the storyteller’s interests to mislead the audience. Thus, there is no need for the audience to be especially skeptical about the stories.
I propose that the ability to critically evaluate evidence may instead have arisen to test one’s own hypotheses about the environment. For example, tracks are sometimes ambiguous or unclear, and the hunter must formulate a hypothesis about their meaning and continually weigh the evidence as the tracking progresses (Liebenberg, 2012, pp. 71;
102). This process is illustrated by the following !Kung hunt:
They began following a gemsbok spoor which, the man said, was made the same morning. After about twenty minutes the man stopped and said, “No, it was made last night,” and abandoned the spoor. Asked what made him change his mind, he indicated a single gemsbok hoofprint with a mouse track inside it, that is, super-imposed on it. Since mice are nocturnal, the gemsbok print must have been left during the night. (Jones & Konner, 1976, p. 342)
However, non-human animals are also responsive to cues in the environment. In general, we should expect a wide range of perceptual and cognitive processes to involve the assessment of hypotheses. The weighing of evidence is mainly intuitive rather than reflective, with the exceptions being the rare cases which happen to bring to mind linguistically transmitted knowledge.
79 4. Empirical Characteristics of Reason
In the previous section, I have illustrated how the biological fitness of individuals in ancestral environments is likely to have been enhanced by cultural knowledge transmitted through stories. Reasoning, I claim, has a key role in this process, by taking rule-like inputs and delivering true beliefs about problems in the natural environment.
Thus, my account is a version of what I call the “inquiry-first view” of reasoning.
Mercier and Sperber rule out inquiry-first views on the grounds that people tend to be poor at correcting mistaken intuitions:
Reason rarely questions reasoners’ intuitions, making it very unlikely that it would correct any misguided intuitions they might have. This is pretty much the exact opposite of what you should expect of a mechanism that aims at improving one’s beliefs through solitary ratiocination. (Mercier & Sperber, 2017, p. 218)
For example, De Neys et al. (2017) found that, across 12 reasoning experiments in which participants gave an initial intuitive response followed by a final considered response, further deliberation typically did not help the participant find the correct answer:
• In 15% to 42% of trials participants gave the correct response initially and finally. • In 48% to 76% of trials participants gave an incorrect response initially and finally. • In 7% to 10% of trials participants initially gave an incorrect response and then a correct final response.
De Neys and colleagues thus conclude that the main role of reasoning in these experiments is to construct post-hoc justifications of intuitive responses.
80 Mercier and Sperber interpret this as one manifestation of a wider phenomenon of
“myside bias”. In the classic myside bias paradigm, participants are asked to think about a controversial issue and come to a conclusion if possible. When explaining the reasoning for their conclusion, participants tend to provide more arguments for their view than arguments against it (Perkins, 1989). This effect holds even when participants are instructed at the outset to produce arguments on both sides of the issue (Stanovich &
West, 2008). Thus, it appears that people find it easier or more natural to search for evidence that supports rather than undermines their view. According to Mercier and
Sperber, myside bias is predicted by rationalization-first views since it helps the reasoner build a case to persuade others or justify themselves, but is in tension with the epistemic goals posited by inquiry-first views.
However, note that inquiry can improve beliefs in two ways: it can test information to avoid or weed out false beliefs, or it can use information to generate or bolster true beliefs. Experimental reasoning tasks are often set up so that participants are cued to have a mistaken intuitive response. However, my view does not predict that people will be naturally inclined to test their intuitions. Under my version of the inquiry-first view, the function of reasoning is not to test information but rather to generate true beliefs from cultural knowledge when faced with novel situations for which we have no intuitions. My view instead predicts that people will tend to recognize that evidence undermines their beliefs, when presented with such evidence. Studies on reasoning universally agree that
81 when participants are presented with the correct answer (e.g. during debriefing) they tend to admit that it is the correct answer.18
Of course, if participants do wish to check their intuitive beliefs, they can use task- relevant knowledge to come a reasoned conclusion, which can then be compared with their intuitive belief. However, participants often do not possess task-relevant knowledge.
Indeed, participants who have expertise – training in logic, for example – are typically excluded from studies. Given that experimental tasks require a response, it is unsurprising that such participants would tend to rationalize their intuitions as best they can.
Under my view, the trigger for reasoning is the detection of a novel problem for which we have no intuitions. Thus, my account predicts that reason should be a flexible system that can be exapted to serve a wide variety of motives and goals. We might say that although the adaptive function of reason is to apply cultural knowledge to generate true beliefs, reasoning as such is neither a scientist nor a lawyer but rather a handy-man for hire.19 For example, the task of defending one’s reputation, and the task of persuading others, are problems that might initiate reasoning. In this way, my view of reason is more
18 An example: in Wason’s famous four-card selection task, very few participants correctly select the two cards that can falsify the given rule. However, when participants are instead asked to turn over the cards and state which of them falsify the rule, nearly everyone gets the answer correct (Wason, 1966, p. 146). 19 I owe this metaphor to Richard Samuels.
82 general than that of rationalization-first views. Moreover, reasoning is often motivated not by a desire to ascertain the truth about an issue, or to justify oneself, or to persuade others, but rather by a desire to defend cherished beliefs or attack unsettling propositions.
Thus, my account captures what is commonly referred to as “motivated reasoning”
(Kunda, 1990). Some of these motivations may have even been adaptive in ancestral environments. However, there is a crucial distinction to be made between reason evolving for such goals, and evolved motives using reason – and perhaps other cognitive systems as well – to attain such goals.
Finally, Mercier and Sperber claim that their account predicts the tendency for beliefs to continue to influence thought and behavior after the reasoner learns they are unwarranted. This “belief perseverance” effect, they suggest, results from reason operating outside of its natural social environment, so that it piles up supporting arguments which favor the relevant belief, without any pushback from an interlocutor:
Participants were asked to distinguish between real and fake suicide notes, and were told how well they’d done. They were then left to think about their performance for a little while. During this time, they thought of many reasons why the feedback made sense: they had always been very sensitive, they knew someone who was depressed, and so on. Then the participants were told that in fact the feedback had been completely bogus, bearing no relationship whatsoever with their actual performance. But it was too late. Participants had found many reasons to buttress their beliefs, so that even when the initial support was removed, the beliefs stood on their own. (Mercier & Sperber, 2017, p. 242)
83 Hence, even though participants are told that the belief is unwarranted, they will have found their own supporting arguments in the meantime.
As we have seen, my view can allow that we are sometimes motivated to build a one- sided case. However, this is not the commonly accepted cause of the perseverance effect.
Rather, initially the belief is used to generate a causal explanation, so that when the belief is subsequently discredited, the belief will rightly be given up but the causal explanation remains and can subsequently be drawn on (Anderson, Lepper, & Ross, 1980). For example, if participants are initially told that risky people are less/more successful as firefighters, they tend to generate plausible explanations of why this is so, e.g., risky firefighters save more people because they fight fires that others will not. When they are subsequently debriefed that the initial belief has no basis in evidence, they will drop the belief but unwittingly continue to be influenced by the explanation they generated. Thus, the perseverance effect does not result from participants thinking up new supporting arguments, but rather from participants thinking up causal explanations. This fits well with my account of reason, since socially transmitted knowledge is used to understand the world, with reasoning serving to apply it to novel situations rather than search for further evidence.
5. Concluding Remarks
In this paper I have outlined a new naturalistic account of human reason, according to which the function of reason is to generate true beliefs. Private reason is a superpower,
84 but only when it is fueled by cultural knowledge, gained through social learning. I have attempted to sketch the ways in which this account makes sense of biases that reason displays. However, many questions remain. Do some kinds of problems trigger reasoning more readily than others? Are there multiple reasoning systems that might be revealed through dissociations? Much work remains to be done as part of a wider research project to flesh out and empirically evaluate the theory.
85 Chapter 3 Wason Confirmed: Why Confirmation Bias is Not Myside Bias in Disguise
1. Introduction
In the early 1960s the psycholinguist Peter Wason reinvigorated the study of the psychology of reasoning by introducing a novel research project focused on whether people naturally attempt to falsify their hypotheses and, if so, how. To this end, he devised the
“2-4-6 task”, in which participants try to discover a rule by formulating and testing predictions, and the “four-card selection task”, in which participants judge the evidential relevance of potential tests for a given rule. Based on his studies, Wason famously concluded that people have a tendency to search for evidence in ways which unwittingly favor information that supports rather than falsifies their hypotheses (Wason, 1966). This was later termed “confirmation bias” (Mynatt, Doherty, & Tweney, 1977).
86 Issues relating to confirmation bias have been extensively studied in the six decades since Wason’s pioneering work.20 However, the literature has increasingly shifted away from his original interpretation of performance on the 2-4-6 and four-card selection tasks. Building on this trend, recently it has been argued that the notion of confirmation bias should be replaced by the notion of “myside bias”. For instance, in a review of the literature, the cognitive scientist Hugo Mercier writes:
most of the conventional wisdom about the confirmation bias is wrong – starting with its name … there is no such thing as a general tendency to confirm whatever one thinks about, only a tendency to find arguments that support one’s own views – a myside bias. (Mercier, 2017, p. 99f.)
This represents a sea change in how confirmation bias is thought about. The core of the new conception is that rather than searching for evidence in ways which unwittingly favor information that supports whatever proposition they happen to be testing, people unwittingly seek out evidence that favors propositions they believe.
Getting clear about what the bias amounts to may have practical value for attempts to debias reasoning. But the new conception also bears on weighty theoretical issues concerning the nature and evolutionary origins of the human capacity to reason. For
20 The notion of confirmation bias has sometimes been expanded to include phenomena such as “pseudo- diagnosticity” (Doherty, Mynatt, Tweney, & Schiavo, 1979), “selective exposure” (Jonas, Schulz-Hardt, Frey, & Thelen, 2001), and “motivated reasoning” (Hahn & Harris, 2014). In this paper, I restrict attention to the original phenomenon associated with Wason’s studies.
87 example, a number of prominent researchers who take myside bias to be a characteristic feature of reasoning have proposed that the best explanation of this purported fact is that reason evolved for attaining socially strategic goals through persuasion and self- justification (Haidt, 2012a; Mercier & Sperber, 2017).
My goal is to show that, when we sharpen the theoretical issues and the right distinctions are made, the new conception is seen to rely on a serious misunderstanding of the evidence. Wason was largely right back in the 1960s, and some missteps have occurred in the progression of the literature. Notably, there is a persistent failure to distinguish between effects and processes. In this paper I show that there is evidence for both confirmatory effects and myside effects. However, although there is also evidence that a process which searches for confirming evidence tends to operate when people test hypotheses, there is scant evidence that a process which searches for arguments that support one’s view tends to operate when people deliberate. This is significant because proponents of the theory that reasoning evolved for persuasion and self-justification hold that there is a cognitive module which has the adaptive function of producing myside arguments.
Here’s how I proceed. I begin by examining Wason’s tasks. The confirmation bias interpretation of the selection task faces two main objections. The first is that potential falsification is not avoided by participants. The second alleges that participants are not searching for confirming cases, but rather are guided by unconscious linguistic comprehension heuristics which under certain conditions inadvertently direct attention
88 towards confirming cases, with subsequent reasoning serving to rationalize this selectivity
(ala myside bias). In response, I argue that people tend to search for instances of their hypothesis – which, if discovered, confirm the hypothesis – and show one way in which this “positive” search strategy is biased towards confirmation despite the possibility of falsification.
I then turn to the 2-4-6 task. Here the main objection has been that in typical real- world scenarios, the statistical structure of the environment ensures that falsification is more likely when searching for cases that are compatible rather than incompatible with the hypothesis. In response, I show that if attention is restricted to test outcomes that participants deem relevant to the hypothesis, there does exist a confirmation bias effect that is robust across all environments.
Finally, I turn to reason generation tasks, which are often cited as demonstrating the existence of myside bias. In these studies, participants consider a controversial claim
φ and are invited to conclude either φ or not-φ. The classic finding is that when asked to cite their reasons, participants tend to provide more arguments for their view than against their view. I argue that even if this is taken to be a myside bias effect, it can be plausibly explained away as a side-effect of an unbiased reasoning process that led the participant to adopt the view.
89 2. Confirmation Bias
In this section I provide a detailed analysis of the 2-4-6 task (Wason, 1960) and the four- card selection task (Wason, 1966). Although expositions typically present the tasks in chronological order, I use my analysis of the selection task to shed light on the 2-4-6 task.
2.1 The Four Card Selection Task
Suppose you are shown four cards with a revealed face and a hidden face, and are told that each card has a letter on one side and a number on the other. For example, the revealed faces may display A, D, 4, and 7 (Figure 1).
A D 4 7
Figure 1: Example card faces in the four card selection task.
Which cards would you need to turn over to find out whether someone who utters the following statement about these four cards is lying?
“If a card has a vowel on one side, then it has an even number on the other side.”
90 Take a moment to answer before reading on. If you are like most participants you will have selected either card A alone or cards A and 4. However, the correct answer is A and
7 since either of these cards may falsify the statement (Figure 2).21
Wason noted that people tend not to accept that a conditional with a false antecedent is true (Johnson-Laird & Tagart, 1969). Thus, participants are likely to think that if the conditional is true, then there exists at least one positive instance of it – i.e. a card with a vowel and an even number. Since only the A and 4 cards may provide positive instances of the conditional when turned, Wason originally concluded that there is an
“apparent bias towards verification” of the statement (1966, p. 147).
A D 4 7 VOWEL CONSONANT EVEN ODD
ODD EVEN ODD EVEN VOWEL CONSONANT VOWEL CONSONANT
modal3selection falsifying3counter?instance verifying3positive?instance
Figure 2: Possible scenarios in the four card selection task.
21 A and 7 may not be the correct answer relative to how the participant understands the instructions. This feeds into a wider debate about rationality and the value of normative models (Elqayam & Over, 2016). Nevertheless, A and 7 is the correct answer assuming that the cards are to be turned at the same time and that the rule is restricted to the four cards, rather than a pack to which they belong.
91 The “four card selection task” has been used in over 300 published studies, perhaps making it the most researched task in the history of the psychology of reasoning (Evans,
2017). The now standard “abstract” version uses an indicative conditional with abstract content that cites instances rather than categories (e.g. “If a card has A on one side, then it has 4 on the other side”), along with instructions to determine whether the “rule” is “true or false”. By convention, across studies the four cards are denoted True Antecedent (TA),
False Antecedent (FA), True Consequent (TC), and False Consequent (FC).
A recent meta-analysis of the abstract selection task found that 11% of participants made the correct selection, which corresponds to TA and FC (Figure 3). Improved results have sometimes been found in studies that use realistic or thematic content such as “Every time I go to Manchester, I travel by train” (Wason & Shapiro, 1971). However, the results of replication studies are mixed, and facilitation effects may be due to the content cuing memories of past experiences with falsifying cases.22
22 For reviews see Evans et al. (1993) and Dominowski (1995). Performance is typically much improved when deontic rules are used, such as “If you tidy your room, then you may go out to play” (Manktelow & Over, 1991). However, since the task is to ascertain whether people are violating the rule rather than to ascertain whether the rule is true or false, these selection tasks are not directly relevant to the issue of confirmation bias.
92 50
40 45 41 39 30 TA 28 TA,3TC
PARTICIPANTS 20 23 TA,3TC,3FC
OF TA,3FC
%3 10 11 11 0 4 ABSTRACT REALISTIC
Figure 3: Canonical selections made on the four card selection task, from a meta-analysis of 55 studies using abstract material and 44 studies using realistic material. Based on data reported in Ragni et al. (2018), Table 4.
Wason rarely used the term “bias” and it appears that by “bias towards verification” he merely meant a tendency to search for positive instances of the rule. Call this a “positive test strategy”. In contrast, a “negative test strategy” searches for cases that are incompatible with the rule. Wason did not claim that participants desire to show that the rule is true. This would be odd, since the rule is abstract and arbitrarily chosen. Instead, he claimed that participants rely on a positive test strategy to examine whether the rule is true.
Participants who use a positive test strategy are trying to confirm the rule, but only in the weak sense that the strategy involves initiating a search process the goal of which is to identify positive instances which confirm the rule. There is some debate in the wider psychological literature about the extent to which goal activation, goal pursuit, and goal
93 monitoring require conscious awareness and intentional control (Marien, Custers, Hassin,
& Aarts, 2012). For the purposes of this paper, it suffices to note that the goal of the search process figures in a stopping rule for the search and a decision rule for the test strategy. If the goal is attained, the search is terminated and the rule is judged true. If the goal is not attained, the rule may be judged false. I refer to the tendency to use a positive test strategy as “confirmation biasg”, where the subscript abbreviates “goal”. Since confirmation biasg refers to the process that generates the selections, it can be invoked to explain the selections. The modal selection is TA and TC because there is a tendency for participants to search for positive instances. These participants fail to grasp that a positive test strategy filters out the class of potentially falsifying instances associated with turning FC.
Furthermore, they also often fail to grasp that turning TC is redundant, since turning TA will either conclusively verify that at least one positive instance of the rule exists or conclusively falsify the rule.
The term “bias” is also widely used to denote systematic deviation from a normative standard. Here is one natural way to quantify a bias towards confirmation in the selection task. Let the probability of falsification when turning TA and when turning FC be denoted by a and b, respectively. Further, let the probability of finding a confirming positive instance when turning TA and when turning TC be denoted by c and d, respectively. These probabilities are epistemic and do not depend on what the reverse side of the cards actually show, beyond the assumption that every possible combination that conforms to the task instructions has a non-zero chance of occurring. The ratio associated with the logically
94 correct solution (TA, FC) acts as the baseline. For the three most common selections, the ratios of the probability of finding a confirming positive instance to the probability of finding a falsifying case can be expressed as:
TA, FC: TA: TA, TC:
These ratios are ordered TA, FC < TA < TA, TC, since:
< <
I refer to this as a kind of “confirmation biase” where the subscript abbreviates “effect”.
Confirmation biase is a property of the selections, so it cannot be invoked to explain the selections. The biase may be more or less strong according to the degree to which there is deviation from the baseline set by the logically correct selection. Thus, the selection of TA and TC displays a stronger confirmation biase than the selection of TA alone. For example, if 0.5 is assigned to each of the four probabilities, then the ratios are 0.5 (TA, FC); 1 (TA);
2 (TA, TC). Alternatively, if the probability of getting a vowel is changed to 5/26, then the ratios are 0.7 (TA, FC); 1 (TA); 1.4 (TA, TC). The point to note is that the order remains the same.
Henceforth, when I use the term “bias” without a subscript, I intend to leave the meaning ambiguous. Nevertheless, it is often important to distinguish these two senses of
“confirmation bias”. As we will see, it is possible to have “confirmation biase” without
“confirmation biasg”, since another process may give rise to the effect; and it is possible to have “confirmation biasg” without “confirmation biase”, due to the statistical structure of
95 the environment or the operation of another process that blocks the effect. In the remainder of this section, I examine two objections that have been raised against a confirmation bias interpretation of selection task performance. I argue that these objections fail. There is good evidence for the existence of both confirmation biase and confirmation biasg.
Potential falsification is not completely avoided
The first objection is that potential falsification is not completely avoided by any participant. For example, Mercier claims that the TA selection rules out confirmation bias:
the A [TA] card, picked by most participants, can falsify the rule, so that its choice cannot be explained by a confirmation bias. (Mercier, 2017, p. 106)
TA is typically either selected alone or alongside TC. As I have shown above, both kinds of selections display confirmation biase. However, since Mercier takes confirmation bias to explain the selections, it may be that he has confirmation biasg in mind.
Clearly, the selection of TA alongside TC can be explained by confirmation biasg.
This, recall, is the modal selection. Evidence in favor of this explanation is provided by a study in which participants were asked to provide reasons for their selections, as they made them (Goodwin & Wason, 1972).23 For instance, one participant who selected both
23 The materials used in this study were shapes (triangles, squares) and colors (red, blue), rather than letters (vowels, consonants) and numbers (even, odd). For ease of comprehension, in the following quotes I have inserted the equivalent materials from the example we have been using.
96 TA and TC gave the following reason for his TA selection: “Only interested if it has [an even number] on it”. Similarly, another participant stated the reason for his TA and TC selections in terms of positive instances: “We must look at both cards which show one of these features [vowel, even] to see if its other [side] has the second feature required if the statement is true”.
The selection of TA alone can also be explained by confirmation biasg. For example, some participants may initially focus on TA and TC because they potentially provide positive instances, but subsequently have an insight that only cards which potentially falsify the rule should be selected (Wason & Johnson-Laird, 1972, p. 185). Thus, TC will be rejected. However, FC will not be selected alongside TA because at the time of the insight into the importance of falsification FC had already been filtered out from attention by the initial application of a positive test strategy.
Alternatively, TC may simply be overlooked during the positive test strategy. An
“if-then” statement has a temporal directionality in the following senses: linguistically, the antecedent comes before the consequent; and, psychologically, to understand and evaluate the conditional, the antecedent is supposed and the consequent subsequently considered
(Evans, 2007). In the selection task, participants first encounter the shown card faces, and subsequently consider what the hidden faces might show. Thus, it is natural to associate the antecedent of the rule with the shown card faces, and the consequent of the rule with the hidden card faces. As a result, participants may find it cognitively demanding to
97 imagine a possible value on the hidden side of TC and mentally represent it as satisfying the antecedent.
Of course, there are other plausible explanations of the selections that do not invoke a positive test strategy. Moreover, there will be individual variation in how participants respond to the task. For example, some participants seem to think that the rule only applies to TA, perhaps because (as previously mentioned) it is natural to associate the antecedent of the rule with the shown card faces. Note, however, that in this case the participant has misunderstood the setup in such a way that the task is trivialized, so that the selection of TA alone provides little information about how they select evidence. The important point is that, contra Mercier, the selection of TA can be explained by confirmation biasg. Moreover, the fact that TA and TC is the modal selection, combined with the written explanations of the selections given by participants, is good evidence that in the majority of cases confirmation biasg is the best explanation of the selections.
The selections reflect “matching bias”
Perhaps the dominant reason for rejecting the confirmation bias interpretation of the selection task is the belief that the selections actually reflect what has been called
“matching bias”. This is taken to be demonstrated by results in the “negations paradigm”, which uses an additional three variants of the target rule (Evans & Lynch, 1973):
(i) If a card has A on one side, then it has 4 on the other side.
(ii) If a card has A on one side, then it does not have 4 on the other side.
98 (iii) If a card does not have A on one side, then it has 4 on the other side.
(iv) If a card does not have A on one side, then it does not have 4 on the other
side.
A confirmation bias interpretation of the selection task seems to predict that TA and TC will be preferred to FA and FC regardless of the presence of negations. However, averaging across all four rules, the actual order is TA > FC > TC > FA. Moreover, there are more TA selections when the antecedent is affirmative (“has A”); more TC selections when the consequent is affirmative (“has 4”); more FA selections when the antecedent is negated
(“does not have A”); and more FC selections when the consequent is negated (“does not have 4”). This effect was dubbed “matching bias” since participants tend to select cards whose revealed faces have values which are mentioned in the target rule. Due to these results, Wason gave up his initial explanation of selection task performance (Wason &
Evans, 1974, p. 142).
The subsequent literature has made much of the fact that the modal selection for both rule i and rule ii is A and 4. This selection corresponds to TA and TC for rule i, but
TA and FC for rule ii. The lesson often drawn is that the same process underlies the modal selections for both rules, so that with respect to rule ii participants get the correct answer for the wrong reason. For example, Mercier comments that:
although participants use the same mechanisms both with the standard and with the negated rule, only in the latter case do they reach the correct answer.
99 In neither case do they reach this initial answer because of a confirmation bias. (Mercier, 2017, p. 106)
The matching bias interpretation of the selection task has been criticized on the grounds that it does not adequately explain performance. For example, Van Duyne (1973, p. 241) claims that it is a “tautology”, and more recently Gigerenzer (2011) claims that it is a
“circular restatement” of the phenomenon to be explained. These criticisms may reflect a failure to clearly distinguish between biase and biasg. It is difficult to see why a participant undertaking the selection task would consciously use a strategy that has matching as a goal. Thus, matching bias might instead be viewed as an effect, in which case explaining the selections in terms of matching bias will seem circular.
Evans (1984) attempts to provide a rationale for a matching process by proposing that selections across the four rules are influenced by the operation of an unconscious
“matching heuristic” that focuses attention on relevant information before analytic processing is carried out.24 Given a sentence that contains a negation (“not X”), the matching heuristic focuses attention on X rather than the complement of X, or the psychological contrast class of things that are not X. According to Evans, this is because
24 Since the mention of the values in the rule increases their salience, it might be held that the matching effect is due to the operation of an availability heuristic (Tversky & Kahneman, 1973). Evans rules this out on the grounds that matching effects also occur on evaluation tasks, in which participants are presented with all possible cases (thus controlling for availability) and asked to judge whether they conform to the rule, contradict the rule, or are irrelevant to the rule.
100 in everyday conversation negations are typically used to deny presuppositions rather than to assert new information. For example, if I say “I didn’t see the game”, the topic of the conversation is likely the game rather than what I did instead.
However, regardless of how plausible this rationale is, there is an empirical problem with the matching bias explanation: A and 4 is the modal selection only for rule i and rule ii. This point was obscured because Evans originally averaged over all four rules.
When considering each rule in isolation, there are six selections that are diagnostic for the operation of a matching heuristic, a positive testing strategy, and a negative testing strategy (Table 1).
rule i rule ii rule iii rule iv
matching heuristic A, 4 A, 4 A, 4 A, 4
positive testing A, 4 A, not 4 not A, 4 not A, not 4
negative testing A, not 4 A, 4 not A, not 4 not A, 4
Table 3. Selections predicted in the negations paradigm by matching, positive testing, and negative testing. Diagnostic selections are indicated by bold font.
Contra the predictions that are diagnostic for the matching heuristic, there is little evidence of a tendency to select FA on rules iii and iv (Reich & Ruth, 1982). To better capture the data, Evans (1984) amended his account to include an “if-heuristic” that
101 categorizes FA cases as irrelevant before the matching heuristic (sometimes now referred to as the “not-heuristic”) operates. The rationale for the if-heuristic is that the linguistic function of “if” is to provoke hypothetical thinking about cases in which the antecedent is true.
It has largely gone unnoticed that this amendment renders matching indistinguishable from a positive test strategy for rule iii and from a negative test strategy for rule iv. Additionally, under both the old and the new versions of the theory, there are no selections that distinguish between matching and a positive test strategy for rule i, or between matching and a negative test strategy for rule ii. Thus, the empirical evidence does not rule out the possibility that participants switch strategies due to the differing structure of the rules.
Wason and Evans (1976) asked participants to justify their selections for rule ii, and found that they appeared to be able to explain their choices in a logical manner. The researchers concluded that participants were merely rationalizing their choices, since participants tend to make incorrect selections for rule i, even after they made the correct selections for rule ii. That participants make their selections based on unconscious heuristics and then use reason to justify their intuitive choices is key to the case for reinterpreting selection task performance as a form of myside bias:
instead of finding reasons for and against each card, participants find plenty of reasons supporting their initial card choice, neglecting reasons to pick other cards, or reasons to not pick the cards initially chosen … this confirmation bias is in fact best understood as a “myside bias”. (Mercier & Sperber, 2017, p. 217)
102 However, there are grounds to think that rule ii facilitates performance. For example,
Pollard and Evans (1980) found that when participants are told that “if p then q” is true, they judge that “if p then not-q” is false, and vice versa. Thus, Evans and Over (1996, p.
160) suggest that some participants take rule ii to imply the falsity of rule i, and then use a positive test strategy to determine whether rule i is true. In effect, this is an indirect form of negative testing with respect to rule ii.
Alternatively, rule ii may facilitate negative testing by reducing the cognitive load associated with mentally representing negations (Sperber, Cara, & Girotto, 1995) and the set of things that comprise the complement of the 4 card. Positive testing with rule i and negative testing with rule ii require the participant to search for cards that have an A on one side and 4 on the other (Table 1). However, negative testing with rule i and positive testing with rule ii require the participant to search for cards that have an A on one side and not a 4 on the other. That participants find it difficult to mentally represent the complement of the cards is suggested by results from the “explicit negations” paradigm, in which negations are included on non-matching card faces (Evans, Clibbens, & Rood, 1996;
Stahl, Klauer, & Erdfelder, 2008). Using our example, the D card would be replaced with one that reads “a letter which is not A”, and the 7 card would be replaced with one that reads “a number which is not 4” (Figure 4).
103 the$ a$letter$ the$ a$number$ letter$ which$is$ number$ which$is$ A not$A 4 not$4
Figure 4. Example card faces shown in the explicit negations paradigm.
The introduction of explicit negations on the cards is often said to result in a reduction or elimination of matching bias. This is apt to mislead, since all the cards in this paradigm show matching values. What the results show is that for rule ii the modal response switches from TA and FC, to TA and TC. This suggests that for most participants positive testing is the default strategy, which may be dropped in favor of negative testing when the latter is less demanding. If the underlying factor is the conservation of cognitive resources, then contra Wason and Evans there is no need to suppose that participants who use negative testing for rule ii will grasp that the positive test strategy is incorrect and subsequently use a negative test strategy for rule i.
Thus, contrary to what many have supposed, the case for matching bias is underwhelming at best. First, a matching effect is not evident for selections with rules iii and iv. And, second, even if we restrict the analysis to rules i and ii, postulating a heuristic matching process fails to capture more of the data than postulating a switch between positive and negative testing, and has less independent plausibility than postulating facilitation effects under rule ii. Of course, the point remains that confirmation biase is found only on rule i, whereas there is a matching effect for both rule i and rule ii. However,
104 this merely shows that there are conditions under which confirmation biase is reduced or eliminated. This is generally the case for cognitive biases. Indeed, if this were not the case, then there would be no hope of debiasing people.
2.2 The 2-4-6 task
Suppose you are told the following three numbers conform to a rule the researcher has in mind:
GIVEN TRIPLE: (2, 4, 6)
Your task is to discover the rule by repeatedly proposing sets of three numbers and receiving feedback about whether they conform. When and only when you are highly confident you know what the rule is, you state your hypothesis.
Before reading on, take a moment to think of a triple for testing. If you are like most participants, you will have generated a triple that is a positive instance of your hypothesis rather than incompatible with it. Here is a representative exchange (Wason, 1960, p. 137):
TEST TRIPLES: (8, 10, 12) [conforms]
(14, 16, 18) [conforms]
(20, 22, 24) [conforms]
(1, 3, 5) [conforms]
HYPOTHESIS: The rule is that by starting with any number two is added each
time to form the next number. [incorrect]
Wason found that most participants gave an incorrect answer at least once, with over one- quarter never giving a correct answer. Just as a positive test strategy on the selection task
105 determines whether the statement is true by determining whether positive instances of it exist (on the cards), a positive test strategy on the 2-4-6 task determines whether the hypothesis is true by determining whether positive instances of it exist (within the set of things determined by the actual rule). Participants who stated incorrect hypotheses tended to rely on an exclusively positive test strategy. The actual rule is:
RULE: Any increasing series of numbers.
Given that the hypothesis formulated by the participant is consistent with but more restrictive than the rule, an exclusively positive test strategy always leads to confirmation.
The only way to discover the hypothesis is incorrect is to use a negative test strategy. Just as a negative test strategy on the selection task determines whether the statement is false by determining whether incompatible cases exist (on the cards), a negative test strategy on the 2-4-6 task determines whether the hypothesis is false by determining whether incompatible cases exist (within the set of things determined by the actual rule). In a follow-up experiment, participants who stated an incorrect hypothesis were not told so but rather were asked “if you were wrong, how would you find out?”. Most replied that they would continue to generate positive instances and wait for one to not conform. Wason
(1968, p. 170) concluded that these participants were “trying to confirm” their hypotheses and that negative testing is “a totally alien concept”.
Positive testing can falsify
As with the selection task, psychologists have long questioned the confirmation bias interpretation of the 2-4-6 task. The primary objection is that positive testing can result
106 in falsification, and negative testing can result in confirmation (Wetherick, 1962). Indeed, one review claims that there is a widespread conflation of these categories:
an astonishing confusion is still to be found in the literature … Some authors seem to have real problems differentiating between a confirmation bias and a PTS [positive test strategy]. According to the PTS, persons have the tendency to ask questions in such a way that their hypothesis would be confirmed if the answer was affirmative. Those authors seem to miss the conditional clause here and transform it instead into propositions like: “Persons have the tendency to seek only for confirmatory evidence”. (Oswald & Grosjean, 2004, p. 86)
We have already encountered this kind of objection, in relation to the selection task. Again, it is important to distinguish between biase and biasg. Participants who use a positive test strategy initiate a search process the goal of which is to identify positive instances of the rule, which, if discovered, confirm the rule. If the search is unsuccessful, they conclude that their hypothesis is false. It is in this weak sense that participants “seek only for confirmatory evidence”.
Using an exclusively positive test strategy may filter out potentially falsifying instances from examination, resulting in confirmation biase. This is especially egregious in the 2-4-6 task since the relevant class of falsifying instances would, if discovered, provide unique information about the way in which the hypothesis is false (i.e. the conditions specified in the hypothesis are not necessary conditions). Nevertheless, it is widely held that the 2-4-6 task does not provide evidence that confirmation biase will result in typical scenarios. This view largely stems from an influential study by Klayman and
107 Ha (1987) which argues that in most realistic rule-discovery settings, positive testing is more likely than negative testing to lead to falsification. Consider the following hypothesis, where Φ specifies some complex set of conditions:
HYPOTHESIS: All and only stars fulfilling Φ have a planetary system.
Klayman and Ha make two assumptions about actual scientific practice. The first is that hypotheses are typically phrased in terms of minority phenomena. For instance, if
(contrary to fact) it were estimated that most stars have planets, it would be more natural to investigate the rare conditions under which stars do not develop planetary systems. The second is that typically for any hypothesis plausible enough to test, Φ and the target property will be approximately co-extensional. Under these circumstances, falsification is more likely to result from examining stars known to be Φ than from examining stars known to be not-Φ (“+Hypothesis-tests”), though these tests will be blind to any falsifying stars that are not-Φ but have planets. This follows trivially from the fact that there are fewer stars with Φ than not-Φ, and not many falsifying cases. Similarly, falsification is more likely to result from examining stars known to have planets than from examining stars known to not have planets (“+Target-tests”), though these tests will be blind to any falsifying stars that are Φ but do not have planets.
Taken together, these two kinds of positive tests can uncover both kinds of falsifying cases. However, +Target-tests are not available to participants in the 2-4-6 task, as they are not allowed to ask the experimenter to provide them with further triples that
108 conform to the actual rule, and then check whether they fall under their hypothesis. In this respect, the 2-4-6 is unlike typical scientific scenarios. The value of the 2-4-6 task is not that it captures a common real-life scenario, but rather that it provides further evidence that most participants find it natural to use a positive test strategy, and difficult to use a negative test strategy. I will now show that although Hayman and Ka are correct that positive testing is more likely to result in falsification in certain environments, there is still a sense in which it gives rise to a robust confirmation biase across environments.
Suppose the rule being tested has the form “ if P then Q ”. The biconditional will be considered later. A positive test strategy tests cases known to be P and cases known to be
Q. A negative test strategy tests cases known to be P and cases known to be not-Q. Further suppose that for each case you initially know only whether P holds or whether Q holds, where the probability that you know the former equals the probability that you know the latter. This assumption will be relaxed later. Under these conditions, the positive test strategy tests every confirming case but only half of the falsifying cases, whereas the negative test strategy tests every falsifying case but only half of the confirming cases.
One measure of the efficiency of a test strategy is given by computing the ratio of tests that falsify (confirm) to the percentage of falsifying (confirming) cases in the target set of things that are P. Suppose for the sake of illustration that one quarter of the population is (P, Q), one quarter is (not-P, Q), one quarter is (P, not-Q), and one quarter is
(not-P, not-Q). For the positive test strategy, 50% of tests discover confirming positive instances and 25% of tests discover falsifying cases. Thus, the ratio of tests that falsify to
109 the percentage of falsifying cases in the set of P things is 0.5, whereas the ratio of tests
that are positive instances to the percentage of positive instances in the set of P things is
1. Efficiency is important because in everyday life only a limited number of tests will be
performed before the issue is either deemed settled or the investigator loses interest in it.
The positive test strategy is less efficient with respect to falsification as the
percentage of cases that are P decreases and as the percentage of cases that are Q
increases. Negative testing is less efficient than positive testing at discovering falsifying
cases when (but only when) both sets P and Q are small. This is intuitive. When the set of
things that are not-Q is large, but there are not many falsifying cases of the form (P, not-
Q), negative testing will perform many tests that discover irrelevant cases of the form (not-
P, not-Q). In such environments, negative testing is less likely than positive testing to
discover falsifying cases (Figure 5).
Chart1Title
1.2 20%$of$cases$are$P 50%$of$cases$are$P 80%$of$cases$are$P 1 2 2 2 1.8 1.8 1.8 0.8 1.6 1.6 1.6 1.4 1.4 1.4 1.2 1.2 1.2 0.6 1 1 1
ratio 0.8 ratio 0.8 ratio 0.8 0.4 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.2 0 0 0 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 %2of2cases2that2are2Q %2of2cases2that2are2Q %2of2cases2that2are2Q 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95 97 99 !positive!testing!(falsification) !negative!testing!(falsification) !!equivalence
Figure 5. Ratio of falsifying tests to the percentage of falsifying cases in the P set.
110 Nevertheless, there will be a robust confirmation biase across all environments, in the following sense. Restrict attention to test outcomes deemed relevant by participants – i.e. cases which make the antecedent of the rule true. Then for the positive test strategy the percentage of falsifying tests will always be less than the percentage of falsifying cases in the P set, whereas for the negative test strategy the percentage of falsifying tests will always be greater than the percentage of falsifying cases in the P set (Figure 6). Similarly, for the negative test strategy, the percentage of confirming tests will be less than the percentage of confirming cases in the P set, whereas for the positive test strategy the percentage of confirming tests will be greater than the percentage of confirming cases in the P set. Furthermore, the more likely it is that you know initially whether a case is Q rather than whether it is P, the more this form of confirmation biase will be magnified.
Now consider testing a rule of the biconditional form “P iff Q ”. A negative test strategy will test every case (P, Q, not-P, not-Q) and hence all confirming cases will be tested. However, the positive test strategy will again test only cases known to be P and known to be Q. Thus, it will again be blind to half of all potentially falsifying cases. As a result, a positive test strategy with respect to a biconditional will also give rise to a confirmation biase in the sense outlined above.
111 1 Chart1Title 1
1.2 0.8 Chart1Title 0.8 11.2 0.6 0.6 0.8 1
0.4 0.4 0.8 0.6
0.6 %)of)falsifying)tests 0.2 0.2 0.4 %)of)confirming)tests
0.4 0.2 0 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0.2 0 %)falsifying)cases)in)P %)confirming)cases)in)P 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95 97 99 0 1 3 5 7 9 11 13 15 17!positive!testing!(falsification)19 21 23 25 27 29 31 33 35 37 39 41 43 45 47!negative!testing!(falsification)49 51 53 55 57 59 61 63 65 67 69 71 73 75 77!!equivalence79 81 83 85 87 89 91 93 95 97 99 !positive!testing!(confirmation) !negative!testing!(confirmation) !!equivalence
Figure 6. Performance of test strategies, applied to conditionals. with respect to the percentage of tests with a result that makes the antecedent true which falsify and which confirm.
Here is an illustrative example, where the P and Q sets are small. Suppose only 4%
of the population is (P, Q); 16% is (not-P, Q); 16% is (P, not-Q), and 64% is (not-P, not-Q).
The negative test strategy will confirm 4% of the time and falsify 32% of the time; this
mirrors the actual proportions in the population. However, the positive test strategy will
confirm 20% of the time and falsify 80% of the time. For the biconditional, test outcomes
deemed irrelevant will have the form (not-P, not-Q). If attention is restricted to test
outcomes deemed relevant, the negative test strategy will confirm 11% of the time and
falsify 88% of the time; again, this mirrors the actual proportions in the set of things that
are P or Q. However, the positive strategy will confirm 20% of the time and falsify 80% of
the time. Thus, contrary to widespread belief, the 2-4-6 task does provide evidence for a
robust form of confirmation biase.
112 3. Myside Bias
Proponents of the new conception of confirmation bias routinely appeal to a small but widely cited literature on informal reasoning. The term “myside bias” was coined by the educational psychologist David Perkins. In a landmark study, Perkins (1985) asked participants to think alone for five minutes about four controversial questions and reach a conclusion if possible. They were then asked to explain their reasoning. Most participants cited two or three arguments for the conclusion reached (“myside arguments”) and perhaps one argument against the conclusion (“otherside arguments”). However, pilot studies suggested that on both sides of each issue there were at least six plausible lines of argument which were in principle within easy access of the participants (Perkins, Farady,
& Bushey, 1991, p. 90). Call this one-sidedness effect “myside biase”.
A follow up study with high school students replicated this effect: participants produced on average 3.5 myside arguments and 0.7 otherside arguments (Perkins, 1989).25
The experimenter then “scaffolded” each student’s reasoning by guiding them towards further reflection. In particular, participants who did not provide many otherside arguments were prompted to produce more. Finally, every participant was asked to cite five more reasons on each side of the case and then to find grounds to challenge the best
25 A myside effect has subsequently been documented in student’s argumentative writing and debate preparation. However, in essays and debates the task is to persuade others rather than to work out what one’s own view is (Baron, 1995; Wolfe, Britt, & Butler, 2009). I deal with persuasion later.
113 two reasons on each side. This procedure led to a 109% increase in myside arguments and a 700% increase in otherside arguments (Figure 7).
8
7 3.8 6 5 4.9 4 Scaffolded Spontaneous 3 3.5
ARGUMENTS 2 1 0.7 0 MYSIDE OTHERSIDE
Figure 7. Mean number of arguments produced for and against one’s position before (spontaneous) and after (scaffolded) experimenter prompts for both. Data from Perkins (1989, p. 187).
The lesson usually drawn is that people are in principle capable of thinking on both sides of the issue, but do not do so spontaneously. Perkins suggested that participants use a
“makes sense epistemology” such that they judge the conclusion true simply in virtue of it making (often superficial) sense in the light of their beliefs:
Such a thinker’s reasoning is dominated by a strategy of cognitive load minimization, rather than a strategy of truth-testing … the first model we deliberately generate that makes sense often serves perfectly well. When it does not, and we are dealing with a situation in practical terms, we quickly discover that failing through experience. (Perkins et al., 1983, p. 187)
114 Alternatively, Mercier and Sperber propose that humans have an evolved cognitive module dedicated to producing and evaluating arguments. This module spontaneously produces myside arguments rather than otherside arguments because its adaptive function is persuasion and self-justification:
If one’s goal is to convince others, one should be looking first and foremost for supportive arguments. Looking for counterarguments against one’s own claims may be part of a more sophisticated and effortful argumentative strategy geared to anticipating the interlocutor’s response, but, in the experimental setting, there was no back-and-forth to encourage such an extra effort (Mercier & Sperber, 2011, p. 62)
Both explanations agree that participants tend to quickly arrive at an intuitive conclusion, perhaps after considering just one line of argument that is prima facie compelling. To further conserve cognitive resources, subsequent reasoning will almost entirely neglect the opposing case. In this sense, myside biase results from the operation of a process that has the goal of finding myside arguments. Call this “myside biasg”.
It is important to note that myside biasg need not render the reasoning process irrelevant to inquiry into the truth. It may be that searching for myside arguments is akin to the positive testing of conditionals. Participants are likely to think that if the conclusion is true, then there are good arguments in favor of it; in which case, a failure to find myside arguments indicates that the conclusion should be doubted. This rationale would explain why participants think they are conducting an impartial inquiry into the issue and do not
115 realize that their reasoning is biased. However, in the remainder of this section, I raise three objections to the myside biasg interpretation of the reason generation task.
Otherside arguments are irrelevant
The first objection is that participants may deem otherside arguments irrelevant to the task of explaining their reasoning. During the process of deliberation otherside arguments will either have been dismissed outright or judged to be outweighed by myside arguments.
In either case, otherside arguments might be considered unworthy of mention when answering the researcher. Moreover, if the participant does wish to explain why they weren’t convinced by the case for the opposing view, they may think it sufficient to cite only the strongest or most obvious otherside argument they considered.
This is a valid criticism of the original Perkins studies. However, it does not apply to the variant of the reason-generation task introduced by Toplak and Stanovich (2003).
Participants indicated the extent to which they agreed or disagreed with three controversial statements. In this paradigm, participants are not asked to explain the reasoning for their conclusion. Rather, after performing several unrelated tasks, participants are asked to think through the three issues carefully and provide arguments both for and against each position, writing down as many as they can. Across the three issues, the mean number of myside arguments was around 2.9 and the mean number of otherside arguments was around 2.2 (Figure 8).
116 4
3 3.3
2.7 2.6 2 Myside:arguments 2.2 2.1 2.2
RGUMENTS Otherside:arguments A
1
0 TUITION ORGANS GASOLINE
Figure 8. Mean number of arguments produced for and against one’s position on three issues, when asked for both. See Toplak and Stanovich (2003, p. 855).
Although the average number of myside arguments generated is comparable to what was found in the initial Perkins study, the average number of otherside arguments generated is more than double. This further supports the criticism that in the Perkins study participants deem otherside arguments irrelevant.
Toplak and Stanovich conclude that, despite the fact that participants were instructed to produce both kinds of arguments, they displayed a “tendency to not give even-handed consideration to both sides of an issue” (2003, p. 858). However, it is noteworthy that, averaging across all three issues, there were only approximately 32% more myside arguments; and, moreover, this is almost precisely the percentage difference found after scaffolding in the Perkins study. While this modest effect is consistent with the operation of a process that has the goal of finding myside arguments, arguably it is not
117 unlikely under the assumption that myside biasg is not the correct explanation of task performance. This brings us to the second objection.
Myside bias is a memory effect
Suppose a person is willing to fairly deliberate about a controversial issue for which there are an equal number of plausible lines of argument on each side. Which arguments actually come to mind will depend on a multitude of factors such as their background knowledge, recent exposure to the arguments, and so forth. Thus, fair-minded individuals will not tend to generate a 1:1 ratio of arguments for and against any given position. For example, even after the process of scaffolding in the Perkins study, participants produced on average 1.7 more myside arguments than otherside arguments. If a person is aware of a greater number of plausible arguments for a position than against it, then, ceteris paribus, they will be more likely to adopt that position. As a result, it is to be expected that even fair-minded individuals will tend to recall slightly more myside arguments than otherside arguments when subsequently asked to think about the issue. Hence, that participants typically cite three or four myside arguments, but only one or two otherside arguments, is to be expected even in the absence of myside biasg.
The memory confound may be circumvented by utilizing a choice blindness paradigm, in which participants are asked to give a response and are then tricked into believing that they gave a different response. Hall et al. (2012) gave participants a two- page questionnaire and asked them to rate their agreement or disagreement (from 1
“completely disagree” to 9 “completely agree”) with 12 statements about morality. They
118 were then asked to read aloud some of their answers from the first page of the questionnaire, and explain the reasoning behind their ratings. However, unbeknownst to the participants, by turning the first page they had attached it to a sticker on the underside of the clipboard that contained opposite versions of two of the original statements.
The results show that 69% of participants accepted at least one of the two altered statements. Furthermore, when asked to explain their reasoning, 53% of participants argued in favor of at least one of the two altered statements. Mercier (2017, p. 108) claims that this paradigm provides “the most conclusive evidence for the myside bias”. This is because myside biase cannot be explained away as a side-effect of the reasoning process that led the participant to adopt the view, since there was no such process.
Strictly speaking, what these results show is that there is a tendency for some participants to provide arguments in favor of what they mistakenly think is their view.
Hence, this is a peculiar form of “myside” bias that reflects a process of post hoc rationalization rather than deliberation. The authors of the study suggest that the results support the theory that the adaptive function of reasoning is persuasion and self- justification (Hall et al., 2012, p. 7). Given that the participant treats the altered statement as their own view, the reason module takes the statement as an input and outputs myside arguments that support it. This brings us to the third and final objection.
119 Myside “bias” is a motivational effect
When asked to explain their reasoning, participants may deliberately attempt to please or appear rational to the researcher, by justifying their conclusion or persuading the researcher that the conclusion is correct. Interpreting this as myside bias runs into two problems. First, in psychology and behavioral economics, to describe something as a “bias” is to portray it as unintentional. For example, in selection task studies, participants who use a positive test strategy do not realize that they are ignoring potential sources of falsification. In contrast, a lawyer who deliberately searches for evidence in support of the hypothesis that their client is innocent would not traditionally be counted as displaying a cognitive bias. Thus, for example, a seminal review of the literature clarifies that:
As the term is used in this article and, I believe, generally by psychologists, confirmation bias connotes … unwitting selectivity in the acquisition and use of evidence. The line between deliberate selectivity in the use of evidence and unwitting molding of facts to fit hypotheses or beliefs is a difficult one to draw in practice, but the distinction is meaningful conceptually. (Nickerson, 1998, p. 175)
The second problem is that “bias” is typically defined relative to a normative standard. In the Perkins studies, it is assumed that during the process of coming to a conclusion the correct thing to do is objectively consider both sides of the issue; thus, when explaining their reasoning, participants should cite myside and otherside arguments in approximately equal numbers. However, if in responding participants aim to please or appear rational to the researcher, then they are performing a different task, for which producing a case in favor of their view is the correct response. Similarly, since the aim of
120 a lawyer is not to ascertain the truth, but rather to defend their client, it is correct for them to present a case that exclusively supports their client.
Wason’s studies are interesting precisely because participants are trying to determine whether the rule is true or false, though they might misunderstand the nature of the rule or the cards. Similarly, Perkin’s studies are interesting insofar as participants are trying to explain their reasoning. If participants are not trying to explain their reasoning, but rather are trying to justify their conclusion, then it is entirely unsurprising that they provide myside arguments. Of course, this raises interesting questions about why participants might want to please or appear rational to the researcher. However, it is not evidence in favor of the existence of a cognitive module that evolved for persuasion and self-justification, and hence spontaneously produces myside arguments.
4. Concluding Remarks
Contra those who hold that confirmation bias should be understood as “myside bias”, there is good evidence that when people test hypotheses they tend to display a bias towards confirmation. Further, I have argued that there is not good evidence that people tend to unintentionally search for evidence that favors their view. Thus, if I am right, proponents of the new conception have the situation exactly backwards.
However, I wish to conclude by suggesting a way forward for the empirical literature on myside bias. Suppose that a choice blindness experiment asked participants to provide reasons for and against the altered position. In this scenario, any motivation to
121 please or appear rational to the researcher will not be satisfied by merely providing myside arguments. If such an experiment were to show that participants tend to produce more myside arguments than otherside arguments relative to the answer they mistakenly think they gave, then this would constitute myside biase and provide compelling evidence in favor of myside biasg.
122 Chapter 4 Reasoning, Rules, and Representation
1. Introduction
Regress arguments have had a long and influential history within the philosophy of mind, and the cognitive sciences. They are especially commonplace as a bulwark against representational or intentional theories of psychological capacities. For instance, arguments of this sort played a prominent role in debates concerning the theory of transformational grammar (Chomsky, 1969b, 1969a; Harman, 1967, 1969), the language of thought hypothesis (Fodor, 1975, 1987; Laurence & Margolis, 1997), the massive modularity hypothesis (Collins, 2005; Fodor, 2000), and intentional accounts of intelligent activity quite broadly (Dennett, 1978; Fodor, 1968; Ryle, 1949). Typically, the regress is presented as one horn of a dilemma:
To explain the manifestation of some kind of capacity, C, the theorist postulates an
(intentional) psychological process of kind P. But, the critic suggests, the successful
operation of any P process itself depends upon some prior manifestation of C. Thus:
Either it is necessary to postulate a second psychological process of kind P, and so
123 on, ad infinitum, or alternatively, one must grant that C can be explained without
positing P.
Thus, the proposed intentional theory is either broken-backed or redundant. Or so proponents of regress arguments would have us believe.
Recently, a similar argument – which we simply call the Regress – has surfaced in philosophical debate regarding the nature of reasoning. Participants in this debate are not concerned with everything that gets called “reasoning”. Rather, they focus on a relatively circumscribed range of reasoning-like phenomena – which they call active reasoning or inference26 – a kind of person-level, conscious, voluntary activity, which at least in paradigmatic instances, results in the fixation of belief. It is widely assumed that active reasoning in this sense is fairly pervasive amongst human beings; that it can involve attitudes with markedly differing contents; that simple, consciously made, deductive inferences are a prototypical case; and that errors in active reasoning are both possible and, indeed, fairly commonplace. For philosophers interested in active reasoning, then,
26 Three comments regarding terminology. First, as is common in the present context – but see Broome (2013, p. 292) – we use “active reasoning” and “inference” interchangeably. Second, although it is slightly infelicitous to use “inference” in this restricted sense, it should be read as such unless explicitly modified –e.g. as in sub-personal inference. Finally, as is typical, we take it to be true, more-or-less by definition, that a process or activity is active only if it is person-level. As such, we count no sub-personal processes as active.
124 the core explanatory challenge is to provide an illuminating account of the nature of this psychological capacity.
Within this context, the presumed significance of the Regress is that it (allegedly) undermines a family of highly influential accounts of inference – what might be called
Intentional Rule-Following theories (or IRF’s). To a first approximation, such theories make a pair of commitments. First, they suppose that inference essentially involves following rules concerning the premises from which one reasons:
(Rule-Following View): All active reasoning involves rule-following operations.
In addition, they impose the following necessary condition on rule-following:
(Intentional View): All rule-following involves intentional states which represent
the rules being followed.
In brief, the Regress purports to show that if such accounts were correct any instance of active reasoning – no matter how apparently simple – would be a super-task involving an infinite number of rule-following operations. In which case, contrary to fact, it would be impossible for finite creatures like us to actively reason.
If the Regress were sound, it would have serious implications for philosophical debate regarding the nature of inference. What may be less obvious is that it would also have significant consequences for scientific theories of reasoning, and cognition more broadly. Within the psychology of reasoning, quite generally, and the psychology of deductive reasoning in particular, it is commonplace to suppose that reasoning relies on
125 mentally-represented rules.27 This commitment is perhaps most apparent in mental logic accounts, where it is explicitly hypothesized that there are “deduction rules that construct mental proofs in the system’s working memory” (Rips, 1994, p. 104). But the commitment is also apparent amongst dual-process theorists who routinely suppose that System 2 processes involve intentional states that represent rules (e.g. Sloman, 1996). Moreover, we suspect – though won’t argue here – that even those who explicitly reject the mental logic approach also presuppose the existence of intentional states that represent rules. For example, mental models accounts seem to presuppose the existence of such states, albeit where the presumed rules are for the manipulation and inspection of iconic models denoting possibilities, as opposed to the construction of mental proofs through chaining linguistic entities such as sentences (e.g. Johnson-Laird, 2008).
Of course, such accounts of reasoning are contentious and may turn out to be false.
But on the face of it, this should be an empirical issue, addressed by empirical means. If the Regress is sound, however, such theories should be rejected a priori. Further, as we will show, since the Regress does not turn essentially on assumptions about the nature of active reasoning per se, the argument, if sound, would apply to a far broader class of phenomena. Specifically, as we will see, it would apply, with minimal modification, to
27 It is worth noting, in this regard, that philosophers writing on inference have tended to focus on cases in which we reason in accordance with logical rules, such as modus ponens. Further, Boghossian (2008, p. 499) claims that whereas denying the Rule-Following View of reasoning in general seems false, denying the Rule-Following View of deductive reasoning in particular seems “unintelligible”.
126 processes that are unconscious and sub-personal and, hence, not active. If sound, then, the
Regress would have ramifications for a wide array of theories in many regions of cognitive science, including theories of perception.
Fortunately, the Regress is not sound. Formulations of the argument are invariably underspecified; and once presented in suitably perspicuous fashion, it becomes clear that the Regress relies on assumptions no sensible version of IRF should endorse. The primary burden of this chapter is to show why this is so.
Here’s how we proceed. In §I, we explain the IRF account of reasoning in more detail, and set out some of its prima facie virtues. In §II, we aim to explain the general structure of the Regress, and provide the most charitable formulation of the argument that we can. In §III, we discuss a standard – and we think correct – response to this original Regress: to posit sub-personal inferential processes. We show that this response provides a plausible way to block the original Regress. But following suggestions from
Boghossian and others, we also a) show how to develop a Revenge Regress, which targets
IRF’s about sub-personal processes, and b) explain how to use this result to develop a
Strengthened Regress, which fills the gap in the original argument. Finally, in §IV we explain why the Strengthened Regress is still subject to a serious objection, and, in §V, we address two responses to this objection.
127 2. The Virtues of Intentional Rule-Following Accounts of Inference
The IRF is not so much a single account of inference, as a family of proposals that share a common commitment to the Rule-Following View of inference, and to the Intentional View of rule-following. In our view, such proposals merit serious consideration because they possess a host of explanatory virtues. We are especially sympathetic to variants of IRF that incorporate some form of computationalism about mental processes – a class that includes the sort of ‘classicism’ advocated by Fodor and Pylyshyn (1988), versions of connectionism (e.g. Smolensky, 1988), and some recent Bayesian approaches to cognitive modelling (e.g. Perfors, Tenenbaum, & Regier, 2011). These frameworks are amongst the most plausible extant approaches to the study of higher cognition in general, and reasoning in particular.
Although this is not the place to discuss the virtues of IRF’s in detail, a brief reminder should make clear that much is at stake, if the Regress is sound. First, consider some of the prima facie explanatory virtues that accrue merely as a result of adopting the
Rule-Following View (cf. Boghossian, 2014, pp. 4, 12):
• A theory of reasoning should discriminate reasoning from mere causation by
belief (and other intentional states). Not all instances of beliefs causing other
beliefs are inferences. Notoriously, there are “deviant” causal chains involving
beliefs that are obviously non-inferential.28 The rule-following account helps to
28 Example: Suppose John believes that he’s late for class, and that this realization makes him
128 explain the difference. Very roughly, in the case of inference, the influence of
belief is wholly mediated by rule following operations; in the other cases, not.
• Since not all reasoning is good reasoning, we should prefer, on grounds of
generality, an account that covers both the good and the bad. Rule-following
accounts can capture this desideratum. On such views, one can reason badly,
either by following a bad rule, or by making mistakes in one’s attempt to follow
good rules. In contrast, good reasoning only occurs when one correctly follows a
good rule.
• A theory of reasoning ought to explain the sorts of generality that are exhibited
by inference. For example, it is widely recognized by philosophers and
psychologists that we are capable of reasoning about an exceedingly broad array
of topics – roughly, any topic for which we possess concepts. Moreover, our
inferences often exhibit similar patterns or “logical forms” across these various
topical domains. Rule-following accounts provide promising explanations of
such phenomena. Specifically, if some inferential rules are akin to logical rules
in being largely “content independent” or “formal”, then we have a partial
explanation of why we are able to reason about so many different subject
matters. Further, if we suppose that humans follow these rules in lots of
sweat. If on the basis of this experience he came to believe that he was sweating, we could have a case of causation by belief, but not inference.
129 different contexts, we will have an explanation of why inferences in different
domains exhibit similar forms.
• A theory of active reasoning should both subsume, and explain the difference
between, deductive and inductive reasoning. Once again, the rule-following
picture offers a natural account. When reasoning deductively the relevant rule-
following operations involve deductive rules; and when one reasons inductively
the relevant rules are inductive ones.
No doubt there are other issues that the Rule-Following View might help address, but let’s turn to the Intentional View. As we see it, there are two deep and closely related explanatory motivations for this view. The first is what we call the Guidance Problem. The above explanatory virtues of the Rule-Following View all turn on the assumption that rules can in some sense guide our cognitive activities. But how is this possible? After all, a rule qua rule is “just an abstract object” and so presumably incapable of exerting any causal influence (Boghossian, 2014, p. 13).
Here’s where the Intentional View enters the picture. Although rules as such cannot guide cognition, intentional states that encode or represent such rules can. For in addition to their representational properties, intentional states have other properties that are causally relevant – various physical and structural properties, for example. On the
130 Intentional View, then, rules guide behavior in an attenuated sense: They are the contents of intentional states – rule-representations – that are causally implicated in reasoning.29
A second, related virtue of the Intentional View is that it helps resolve a very old problem for rule-following accounts of cognition. In brief, such accounts presuppose a distinction between following a rule, and mere accordance with a rule (Hahn & Chater,
1998, p. 203f.). Without such a distinction, rule-following per se will be of little use in explaining what is distinctive of reasoning. For it will turn out that all processes describable by a rule – that is, all processes that display regularity in their behavior – are rule-following processes. In which case, it will be no more true of reasoning that it involves rule-following, than it will be of, say, the planets that they “follow” a rule when conforming to Kepler’s Laws of planetary motion (Fodor, 1975). Again, we think that the Intentional
View provides a credible approach to this problem. According to this approach:
29 Although it does not require endorsing the Intentional View in its full generality, it is also worth noting that the idea that rules are encoded by intentional states helps explain what is otherwise a puzzling fact about human beings, namely: We are capable of learning rules that influence our behavior, on the basis of “one-shot” instruction or linguistic communication. For example: If, at passport control, the guard tells me “Stand behind the yellow line, until you are called”, I stand behind the yellow line and wait to be called! On the basis of one exposure to instruction, my behavior is modified so that I follow the rule. This is readily explained on the assumption that, on the basis of linguistic processing, I come to possess one or more intentional states that represent the content of the guard’s utterance – i.e. the rule.
131 What distinguishes what organisms do from what the planets do is that a representation of the rules they follow constitutes one of the causal determinants of their behavior. (Fodor, 1975, p. 74)
In contrast, where the planets are concerned:
At no point in a causal account of their turnings does one advert to a structure which encodes Kepler's laws and causes them to turn. The planets might have worked that way, but the astronomers assure us that they do not. (ibid.)
In summary: If the Intentional View is correct, we have a prima facie plausible way both to resolve the Guidance Problem, and to draw the rule-following/rule-accordance distinction. Moreover, since a solution to these problems is a prerequisite for the Rule-
Following View to have any explanatory value, it is exceedingly attractive to combine the
Intentional and Rule-Following views in the manner proposed by IRF’s.
Of course, all of the above is defeasible; and matters would be quite different if there were powerful independent reasons to reject IRF’s. With this in mind, we turn to the
Regress.
3. The Regress
Although a number of theorists have invoked variants of the Regress, Boghossian’s discussion strikes us as the most perspicuous, to date; and for this reason, we focus primarily on it here. In 2.1 we lay out Boghossian’s general strategy. In 2.2 we explain how he aims to establish a crucial premise of the argument, what we call the Rule
132 Application Condition. Then, in 2.3 we sketch the Regress itself; and in 2.4 we provide a more regimented formulation of the argument.
3.1 The General Strategy
It is important to distinguish the Regress, advocated by Boghossian and others, from a range of superficially similar worries. The relevant regress is not an epistemic one. It is not, for example, a regress with respect to justification, or reasons for belief. Nor is it a regress concerning the determination of meanings or contents, of the sort associated with
Kripke’s Wittgenstein (Kripke, 1982). Finally, it is not a definitional regress wherein the definiendum – “inference” – is to be defined in terms of “rule-following”, which in turn is to be defined in terms of “inference”, and so on. Rather, the problem allegedly posed by the
Regress is a regress of mental operations. The worry, in brief, is that IRF’s place active reasoning beyond the grasp of finite creatures by turning every instance of inference into a supertask: an infinite sequence of rule-following operations to be performed in a finite period of time.
Here is the general strategy: To generate the desired regress, Boghossian seeks to show that the IRF entails the following interlocking pair of conditions:
Condition 1: Each inference Ii requires a rule-following operation Rj
Condition 2: Each rule-following operation Rj requires some further inference Ij.
Given these conditions, we can generate a regress of mental operations by cycling between them:
Suppose I draw inference I1;
133 By Condition 1: I perform a rule-following operation R1;
By Condition 2: I draw an inference I2.
By Condition 1: I perform a rule-following operation R2;
By Condition 2: I draw an inference I3.
And so on…
A regress of mental operations ensues. In which case, if the conjunction of the Rule-
Following View and Intentional View entail these conditions, then IRF’s turn all inferences into supertasks.
3.2 Establishing Condition 2
How does Boghossian seek to establish that IRF’s are committed to Conditions 1 and 2?
Since the Rule-Following View asserts that all inference involves rule-following operations, Condition 1 is easily secured. Condition 1 just is a rendering of the Rule-
Following View. In contrast, neither the Rule-Following View nor the Intentional View asserts Condition 2. Boghossian’s main argumentative burden, then, is to show that they entail it.
How is this to be done? Rather than focusing on rules of inference, Boghossian initially discusses a simple decision rule with the aim of drawing out some general morals regarding what, on the Intentional View, would be required for active, person-level, rule- following:
Suppose I receive an email and that I answer it immediately. When would we say that this behavior was a case of following the: (Email Rule) Answer any email
134 that calls for an answer immediately upon receipt! as opposed to just being something that I happened to do that was in conformity with that rule? Clearly, the answer is that it would be correct to say that I was following the Email Rule in replying to the email, rather than just conforming to it, when it is because of the Email Rule that I reply immediately. (2014, p. 13)
Of course, this immediately raises an instance of the Guidance Problem: What is it to follow this rule, as opposed merely to conforming to it? Since the Email Rule “qua rule, is just an abstract object” it cannot directly guide behavior. Instead, if the Intentional View is correct:
… my behavior is to be explained via some state of mine that represents or encodes that rule. (ibid.)
Merely positing such an intentional state does not, however, fully explain this particular instance of rule-following activity. There also needs to be a process in which this rule- representation might figure so as to guide my behavior. And according to Boghossian, it is plausible that this process conforms to the following pattern:
… I have grasped the rule, and so am aware of its requirements. It calls on me to answer any email that I receive immediately. I am aware of having received an email and so recognize that the antecedent of the rule has been satisfied. I know that the rule requires me to answer any email immediately and so conclude that I shall answer this one immediately. (ibid.)
Of course, this is only one specific instance of rule-following activity. Nevertheless,
Boghossian takes it to illustrate what, on the Intentional View, active rule-following in general would require:
135 On this Intentional construal of rule-following, then, my actively applying a rule can only be understood as a matter of my grasping what the rule requires, forming a view to the effect that its trigger conditions are satisfied, and drawing the conclusion that I must now perform the act required by its consequent. (ibid.)
Notice – and this is the crucial point – that this appears tantamount to claiming that “on the Intentional view of rule-following, rule-following requires inference” (ibid.). More precisely, Boghossian appears to be insisting that, on the Intentional View:
Rule Application Condition: For a person-level rule-following process to utilize a
rule-representation, it must contain an inferential sub-process – an inference from
the rule to what the rule calls for under the circumstances.
And, of course, if this is true, then so too is Condition 2. That is, the Rule Application
Condition entails that each rule-following operation Rj requires some further inference Ij.
3.3 The Regress Within Reach
If the above is correct, then an intentional rule-following account of inference is committed to both Conditions 1 and 2. As Boghossian puts it:
On the one hand, we have the Intentional View of rule-following, according to which applying a rule always involves inference. On the other hand, we have the Rule-Following picture of inference according to which inference is always a form of rule-following. (2014, p. 14)
Further, if this is so, then it would seem that any instance of person-level rule-following must involve an infinite series of further rule-following operations. If, for example, I actively follow the Email Rule, then I must draw an inference, in order to follow it; and
136 since, by assumption, this involves rule-following, I must draw another inference, which requires another instance of rule-following... And so on ad infinitum. Thus, Boghossian concludes that:
These two views… can’t be true together. Combining the two views would lead us to conclude that following any rule requires embarking upon a vicious infinite regress in which we succeed in following no rule. (2014, p. 14)
Boghossian is not alone in drawing this pessimistic conclusion. For instance, for the same reason, Wright claims that if rule-following requires a state that carries a content that licenses the inferential transition, then it is “uncertain that any coherent – regress-free – model can be given of what inferring actually is” and hence:
… we must drop the idea that inference is, everywhere and essentially, a kind of rule-following. That, in outline, is the solution to the problem of the Regress. (2014, p. 32f.)
Similarly, Broome claims that if rule-following requires an explicit representation of a rule then:
… you would have to determine whether each particular case of potential reasoning falls under the rule. Doing so would require reasoning, which would again require following a rule. There would be a circle. (2014, p. 632)
In short: some very influential philosophers maintain that the Regress undermines IRF.
137 3.4 The Regress Regimented
With the above exegetical work complete, we are now in a position to set out the Regress in full dress. Our aim is to capture the details and spirit of Boghossian’s discussion as charitably as possible, though without logical lacunae. What follows is our best effort.
The Regress proceeds from the assumption that IRF is true, to the untenable conclusion that active reasoning – or inference – is impossible for finite creatures. And since the
Regress targets IRF’s about active reasoning, it is natural to formulate the Rule-Following
View and Intentional View in person-level terms. That is:
(1) Any process of inference is a kind of person-level rule-following.
(2) Any process of person-level rule-following utilizes a person-level rule-
representation.
Here (1) and (2) clearly characterize a version of IRF about inference, or active reasoning.
But as we saw earlier, without additional premises, they do not suffice to generate a regress. Rather, one must further maintain a version of what we earlier called the Rule
Application Condition:
(3) For a person-level rule-following process to utilize a personal-level rule-
representation, it must contain an inferential sub-process – a person-level inference
from the rule to what the rule calls for under the circumstances.
138 Here “sub-process” is to be understood as referring to a proper part of the person-level rule- following process.30 It follows from (1), (2), and (3) that:
(4) Any inferential process involves an inferential sub-process.
Since any such inferential sub-process is itself an inference, it will also involve an inferential sub-process; and so on, ad infinitum. Hence, by iteration on (4) we have:
(5) Any inference requires infinitely many inferential sub-processes.
But given that the performance of infinitely many inferences cannot be carried out in finite time, it follows from (5) that:
(6) Inference is impossible for finite beings like us.
Yet active reasoning is possible for creatures like us. At any rate, this is what Boghossian,
Broome, Wright, and almost everyone – including the present authors – suppose.31 In which case, on the assumption that the IRF is true, we appear to have reason to reject the
IRF.
30 This is required to block an interpretation of (3) on which the rule-following process is identified with the inference from the rule to what the rule calls for. For if such an identification is made, (3) will not generate a regress of operations. 31 Presumably some eliminativists, behaviorists, and the like would deny this.
139 4. Sub-Personal Processes, Revenge, and the Strengthened Regress
4.1 Getting Sub-Personal
How should proponents of IRF’s respond to the Regress? If premises (1)–(3) are true, then regress ensues. But it is plausible to reject premise (2) in favor of a weaker requirement.
Specifically, proponents of IRF’s may allow that person-level rule-following sometimes involves person-level rule-representations, whilst insisting that sub-personal rule- representations may also play the requisite role. The resulting variant of the Rule-
Following View can be formulated as follows:
(2**) Any process of person-level rule-following utilizes a rule-representation that
is either personal or sub-personal.
This modification evades the Regress. Moreover, it does so in an independently motivated, and independently plausible fashion. Although the personal/sub-personal distinction is a notoriously vexed one (see Drayson, 2012), for present purposes, the crucial requirement is – as Boghossian recognizes – that person-level processes are “processes of which we are, in some appropriate sense, aware” (2008, p. 483). In contrast, sub-personal states are “not consciously accessible to the thinker” (Boghossian, 2014, p. 15). Yet if this is how we are to draw the personal/sub-personal distinction, it should be clear that any plausible IRF will need to insist that person-level rule-following quite typically involves sub-personal rule-representations. This is because, as Boghossian, Broome and many others recognize, active reasoners very typically lack conscious awareness of following a rule. In which case, proponents of IRF’s have exceedingly good reason to insist that rule-representing states
140 are often sub-personal. Moreover, this has nothing to do with regress worries, per se.
Rather, it is mandated by the antecedent assumption that active reasoning is a commonplace cognitive activity, along with the overwhelmingly plausible empirical claim that active reasoners very frequently lack conscious awareness of any relevant rule, or rule-representing state. More generally, the point is that once a theory posits representational states to explain a cognitive phenomenon, the hypothesized states must be sub-personal, if the agent lacks conscious awareness of them. In this regard, proponents of IRF’s are in the same predicament as psycholinguistics who posit sub-personal representations of syntactic rules, or vision scientists who posit sub-personal representations of edges. And, in our view, this is not bad company to keep.
4.2 The Revenge Regress
We have argued that weakening (2) in the proposed manner both evades the above
Regress, and is independently plausible. Nonetheless, the proponent of IRF is not out of the woods yet. For, as critics note, a closely analogous sub-personal regress can be generated. Thus, Boghossian maintains:
In the present context, going sub-personal presumably means identifying rule- acceptance...not with some person-level state, such as an intention, but with some sub-personal state...Let us say that [such a state] is some sub-personal intentional [i.e., representational] state in which the rule’s requirements are explicitly encoded. Then, once again, it would appear that some inference (now sub-personal) will be required to figure out what the rule calls for under the circumstances. And at this point the regress will recur. (2008, p. 498)
141 The core insight of the above passage is that merely extending the Intentional View to cover sub-personal rule following does little to alter the overall structure of the IRF. In which case, one might think that if utilizing a person-level rule-representation requires an inferential sub-process, then utilizing a sub-personal rule-representation will also require an inferential sub-process – albeit a sub-personal one. And if this is so, then we can generate a Revenge Regress that mirrors the original:
(1*) Any sub-personal inference is a kind of sub-personal rule-following.
(2*) Any process of sub-personal rule-following utilizes a sub-personal rule-
representation.
(3*) For a sub-personal rule-following process to utilize a rule-representation, it must
contain an inferential sub-process – a sub-personal inference from the rule to
what the rule calls for under the circumstances.
From (1*- 3*) it follows that:
(4*) Any sub-personal inference involves a sub-personal inferential sub-process.
And since any such inferential sub-process is itself a sub-personal inference, by iteration on (4*) we may infer:
(5*) Any sub-personal inference requires infinitely many sub-personal inferential
sub-processes.
Finally, given that the performance of infinitely many sub-personal inferences cannot be carried out in finite time, it follows from (5*) that:
(6*) Sub-personal inference is impossible for finite sub-personal systems like ours.
142 4.3 Regress Strengthened
No doubt this conclusion will be welcome to those already suspicious of intentional explanations of the sort found in cognitive science. For the proponent of IRF, however, the
Revenge Regress is exceedingly unfortunate. Supposedly, by allowing for sub-personal rule-representation, IRF’s have an independently plausible way to escape the original
Regress. But if the Revenge Regress is sound, the escape route is blocked, and the IRF is left without a way to account for active reasoning.
In our experience, the significance of the Revenge Regress is not always clearly appreciated. One problem is that it targets a different phenomenon from the earlier
Regress – i.e. sub-personal inference. Why, then, should it be relevant to theories of person- level inference? Another problem is that proponents of the Regress never spell out in detail how the Revenge Regress interacts with the original one in order to further strengthen the case against IRF’s. In view of this, it would be helpful to fill the gap by showing how to combine the Revenge Regress with the original argument in order to develop a
Strengthened Regress. Again, here is our best effort. First, assume the Rule-Following
View:
(1) Any process of inference is a kind of person-level rule-following.
Next, in view of the response to the original Regress, reject (2) and replace it with:
(2**) Any process of person-level rule-following utilizes a rule-representation that
is either personal or sub-personal.
143 Now we require two variants of the Rule Application Condition. The first we retain from the original argument:
(3) For a person-level rule-following process to utilize a personal-level rule-
representation, it must contain an inferential sub-process – a person-level
inference from the rule to what the rule calls for under the circumstances.
However, the replacement of (2) by (2**) requires that we supplement it with another variant of the Rule Application Condition:
(3**) For a personal-level rule-following process to utilize a sub-personal rule-
representation, it must contain an inferential sub-process – a sub-personal
inference from the rule to what the rule calls for under the circumstances.
The crucial difference between (3) and (3**) is, of course, that the former specifies what is involved in using person-level rule-representations, whereas the latter specifies what is involved when actively reasoning with sub-personal rules. These premises commit the intentionalist not to (4) but to:
(4**) Any person-level inference involves either a person-level or sub-personal
inferential sub-process
Moreover, the Revenge Regress still commits the intentionalist to:
(4*) Any sub-personal inference involves a sub-personal inferential sub-process.
Suppose we try to carry out a process of active reasoning. By (4**) it involves a sub-process of either active reasoning or sub-personal inference. If it involves the former, then (4**) will also apply to that sub-process. Thus, if successive iterations were always to lead to a
144 further sub-process of active reasoning, they would generate the original regress. But if at any stage active reasoning involves a sub-personal inference, then by iteration on (4*) the
Revenge Regress is generated. So, we have shown not (5) but rather:
(5**) Any inference requires infinitely many person-level or sub-personal
inferential sub-processes.
Hence, for the by-now familiar reason:
(6) Inference is impossible for finite beings like us.
QED.
5. Rejecting the Strengthened Regress
Although the Regress is widely supposed to show that IRF’s are untenable, we maintain that such a view is unwarranted. Even in its strengthened form, the Regress is unsound.
Our first pass response to the Strengthened Regress is to reject (4*) – the claim that any sub-personal inference involves additional inferential sub-processes. We take it to be obvious that any remotely plausible theory of sub-personal inference – intentionalist, or otherwise – must reject this commitment, since it is viciously regressive all by itself. But, of course, (4*) is a consequence of premises (1*) – (3*) of the Revenge Regress. So, if we are to reject (4*), it must be because one of those premises is false. Further, since our response to the original Regress was to advocate an IRF for sub-personal inference, we are committed to rejecting (3*), since (1*) and (2*) simply describe the Rule-Following and
145 Intentional Views as they apply to sub-personal inference. Our challenge, then, is to argue that it is legitimate to reject (3*).
How is this to be done? (3*) is a sub-personal version of the Rule-Application
Condition. It maintains that sub-personal rule-following of the kind envisaged by the
Intentional View requires an inferential sub-process from the rule to what the rule calls for under the circumstances. The obvious way to justify the rejection of (3*), then, is to explain how a sub-personal process might utilize a rule-representation, without thereby containing an inferential sub-process. We think that this challenge can be met; and indeed, that the right response is an exceedingly familiar one.
5.1 The “Primitivist” Strategy
Causal-explanatory regress is amongst the most commonplace theoretical challenges to intentional theories in cognitive science. It should come as no surprise, then, that cognitive scientists have a routine strategy for quashing such worries. In our view, this strategy works extremely generally, including for IRF’s about sub-personal inference.
Obviously, the intentionalist about rule-following processes cannot maintain – on pain of regress – that rule-guided psychological processes always involve further rule- guided psychological processes. Yet the intentionalist need not make this commitment.
Instead, they can – and often do – posit a level of primitive processing mechanisms. Such processors may take rule-representations as inputs. In which case, the primitive processes they subserve will, in a sense, be rule-guided – though only in the thin sense that a rule- representation is causally implicated in the process because it is an input to the processor.
146 In contrast to non-primitive processes, however, primitive ones are not rule-guided in a richer sense. That is, they are not rule-guided in the sense that they involve further rule- guided, or inferential sub-processes. Thus, if non-primitive processes – such as those involved in active reasoning – ultimately decompose into primitive ones, then we have a general view of psychological processes on which rule-application regresses cannot occur.
5.2 Primitive Processes and Reflexes
The above primitivist proposal is, of course, exceedingly well-known (Block, 1995; Dennett,
1978; Fodor, 1968, 1987; Fodor & Pylyshyn, 1988; Pylyshyn, 1980). But to see how it helps address the Regress, it is useful to clarify the notion of primitive processes. We think that this is usefully done by comparing them with prototypical (monosynaptic) reflexes.32
Primitive processes are closely analogous to prototypical reflexes in two crucial respects, and importantly disanalogous in another. A first point of similarity is their automaticity.
Given the relevant input conditions, a reflex generates a fixed behavioral output. Knock a knee, and it flexes. Analogously, provide input to a primitive processor, and it too flexes automatically – though not to lift a knee, but to output a representation.
A second similarity is that, in contrast to non-primitive intentional processes, the input-output relations of both reflexes and primitive processes are not inferentially mediated. Given a blow to the knee, it flexes; and as far as we know, no intervening stage of the process involves an inference, or rule-following operation. The same holds for
32 Or, at any rate, a caricature of reflexes.
147 primitive psychological processes. They too have no intermediate stages that involve further inference or rule-following.
Yet there is, of course, an important difference between prototypical monosynaptic reflexes, and primitive psychological processes. In the case of reflexes, such as the patellar or corneal reflexes, the input is not a representation. Crudely put, it is a mere physical magnitude – a stimulus. In contrast, for primitive processes to play their assigned role within an intentional psychology, it is necessary that their inputs are representational.
Indeed, in the cases of interest here, it is necessary that they represent rules. Primitive processors of the relevant sort, then, must be automatic, non-inferentially mediated, rule- applicators. By virtue of being rule-applicators they can underwrite an intentional account of sub-personal inference; and by virtue of being automatic and non-inferentially mediated, they evade the concern that the application of any rule requires a further inferential step. They thus provide an alternative model of sub-personal rule-application, which permits the proponent of IRF to reject (3*).
5.3 Primitive Processes and Stored Program Computers
The above might well sound rather mysterious, were we to lack any model of how sub- personal inference could “bottom out” in processes that are reflex-like and yet rule-guided in the thin sense outlined above. But we do possess a model of such processes. For what we are describing is closely akin to a core aspect of standard, stored program computers.
Such computers take programs (rules) as input; and many of their sub-processes themselves involve rule-governed sub-processes (inferences). But computers are organized
148 in such a manner that sooner or later all this rule-governed activity decomposes in a set of reflex-like operations, which themselves do not rely on any further inferential activity.
Indeed, their possession of this characteristic is amongst the central reasons that the concept of a stored program computer became so important to cognitive science. For it provides a model of how a system can be rule-guided without thereby succumbing to regress problems (Dennett, 1982; Fodor, 1975, 2000). The notion of a primitive process is simply a generalization of this aspect of stored program computers, formulated in a manner that remains neutral regarding the precise nature of the processors and operations involved in human cognition.
5.4 Primitive Processes and Cognitive Architecture
It is worth stressing that the above is little more than standard background theory in much of cognitive science. This is because a hypothesized set of primitive processes and operations is a core facet of what, by deliberate analogy with computer science, is ordinarily called cognitive architecture.
As Zenon Pylyshyn noted a very long time ago, a cognitive architecture consists, at least in part, of “those functions or basic operations of mental processing that are themselves not given a process explanation” (1980, p. 126). That is, they are psychological functions and operations that are not themselves to be explained in terms of other psychological processes – specifically, processes that deploy rules and representations
(ibid.). In this respect, Pylyshyn continues, they are quite unlike “cognitive functions in general… [which] are…explainable…in terms of rules and representations” (ibid.).
149 Instead, primitive processes and operations are “appealed to in characterizing cognition” and “are themselves explainable biologically, rather than in terms of rules and representations” (ibid.)
By broad consent, it is an empirical matter which specific cognitive processes and operations are primitive. That there are such processes and operations is, however, widely
– and we think correctly – assumed to be a presupposition of any sensible intentional psychological science; and for the very same reason that primitive operations are a prerequisite for any sensible version of IRF. Without such operations, regress ensues.
Again, as Pylyshyn observed long ago, the positing of primitive processes and operations avoids “a regress of levels of interpreters, with each one interpreting the rules of the higher level and each in turn following its own rules of interpretation” (ibid.).
6. Counter-Arguments
Positing primitive sub-personal processes allows proponents of IRF’s to reject the Rule-
Application Condition – (3*) – and thereby neutralize the Regress, even in its strengthened form. Yet, as already noted, this regress-blocking strategy is an exceedingly familiar one from cognitive science. So, it is somewhat surprising that it receives so little attention in the literature on active reasoning.
Why might this be so? One obvious possibility is that the cost of primitivism is in some way too high – that it staves off the Regress, only to raise other no less serious
150 problems for IRF’s. In this section, we conclude by briefly considering two possible problems of this sort, which are hinted at in the literature on active reasoning.
Positing primitive processes addresses the Strengthened Regress only at the expense of succumbing to well-known Kripkensteinian rule-following problems
Primitive rule-applicators take rule-representations as inputs. But one might find it deeply puzzling how such inputs could have rules as their contents. How, for example, might an input determinately represent modus ponens as opposed to some other rule? The obvious suggestion is that it represents the rule in virtue of the effects it has on the processor itself – that it induces modus ponens-like behavior in the processor. Yet this suggestion appears to raise Kripke’s familiar Wittgensteinian worries about rule- following. Out of the frying pan, and into the fire.
We are tempted to give the above worry short shrift. The Regress, as understood by its advocates (and by us), is entirely independent of Kripke’s problem. Our aim here has been to address the Regress. If Kripke’s problem remains unaddressed, so be it. That’s a problem for another day.
Of course, if primitivism generated special Kripkensteinian worries for IRF’s, this quick-fire response would ring hollow. But we deny that it has such consequences. First,
Kripke’s problem is orthogonal to the issue of whether one adopts the primitivist proposal.
If Kripke’s problem is a serious one for IRF, then it applies equally to primitive processes and non-primitive ones. Kripke’s problem concerns the possibility of internalizing a determinate rule, given that it is supposed to cover a potential infinity of cases. Assuming
151 the Intentional View of rule-following, this reduces to the issue of what it is for an intentional state to determinately represent a specific, infinitary rule. Further, if IRF is correct, then both primitive and non-primitive processes rely on rule-representations of very much the same sort. In which case, it is hard to see why Kripkean concerns would not arise equally for both sorts of process. In short: positing primitive processes should make no difference to whether or not Kripke’s problem undermines IRF.
Second, we deny that Kripke’s problem is especially troublesome for IRF’s as such, whether or not they endorse primitivism. To be clear: IRF’s are theories about a class of psychological processes – i.e. inferential ones. In contrast, as Boghossian notes:
Kripke’s problem arises against the backdrop of a naturalistic outlook relative to which it is difficult to see how there could be determinate facts about which infinitary rule I have internalized. (2014, p. 13)
For proponents of IRF, to “internalize” a rule is to represent it. In which case, Kripke’s problem clearly arises for IRF’s only when one further demands a naturalistic account of rule-representation. In contrast, the problem has no traction if one “waives naturalistic constraints” –e.g. by allowing for primitive facts regarding the content of rule-representing states.
Speaking personally, we are not much inclined towards this sort of non-naturalism.
But that’s beside the point. Our point is that Kripke’s problem is not a problem for primitivism as such, or even IRF’s as such. Indeed, it is not a problem about psychological processes at all. Rather, it is a problem for naturalistic theories of content. Moreover, it is
152 one that arises for them entirely independently of issues to do with rule-following. If one accepts that we can so much as think about determinate, infinitary rules – e.g. modus ponens – then the problem arises for naturalistic theories of content.
Although positing primitive processes might save an IRF for sub-personal inference, it does so only at the expense of rendering IRF untenable for active reasoning
At one stage, Boghossian considers a proposal about sub-personal inference which may appear to resemble primitivism to a considerable degree. It goes like this:
I consider [the premises] (1) and (2). I do so with the aim of figuring out what follows from these propositions, what proposition they support. A sub-personal mechanism within me “recognizes” the premises to have a certain logical form. This activates some sub-personal state that encodes the MP rule which then puts into place various automatic, sub-personal processes that issue in my believing [the conclusion] (3). (2014, p. 15)
Setting aside the challenge posed by the Revenge Regress, which he maintains is
“importantly correct”, Boghossian is prepared to imagine that some reasoning works like this. However, he continues:
that is not the sort of reasoning that this paper is about—rather, it is about person-level reasoning, reasoning as a mental action that a person performs, in which he is either aware, or can become aware, of why he is moving from some beliefs to others. No such process of reasoning can be captured by a picture in which (a) reasoning is a matter of following rules with respect to the contents of our attitudes and (b) our following rules with respect to the contents of our attitudes is a matter of automatic, subconscious, sub- personal processes moving us from certain premises to certain conclusions.
153 If this is so, then it may seem that, by introducing such sub-personal processes, we fail to account for the sort of active reasoning we sought to understand in the first place. And since this sort of sub-personal process looks much like primitive rule-application, it may further appear that endorsing primitivism thereby undermines the prospect of an IRF about active reasoning.
Appearances, at least in this instance, are misleading. As we noted in section 3, it is widely if not universally supposed that active reasoners often, though not invariably, lack conscious awareness of the rules they are following (cf. Boghossian, 2014, p. 12).33 In which case, it ought to be common ground that there are two different kinds of active reasoning:
AR1: Active reasoning for which there is conscious awareness of the premises and
the conclusion, but not the rule.
AR2: Active reasoning for which there is conscious awareness of the premises, the
conclusion, and the rule.
If the representation of the rule is sub-personal, we have AR1. If the representation of the rule is person-level, we have AR2 – the sort of reasoning that Boghossian says he seeks to explain.
33 Furthermore, that children can engage in active reasoning is taken to be an important reason to avoid conceptual over-sophistication in their accounts. See e.g. Broome (2013, pp. 229, 236); Boghossian (2014, p. 6).
154 In our view, a theory of reasoning ought to capture both these kinds of active reasoning. And, as far as we can tell, our response to the Regress in no way prevents us from doing so. If we insisted that all rule-application was sub-personal, then we could not.
But we make no such commitment. Indeed, we allow for at least four different sorts of rule-governed process.
• We allow for AR2 because we accept that, in some cases, people are conscious of the
rules they follow, as well as their premise and conclusion attitudes.
• We allow for AR1 because we posit rule-representations that are sub-personal and,
hence, not the subject of conscious awareness.
• We allow for non-primitive, sub-personal inferences where one has no conscious
awareness of premises, conclusion or rule.
• Finally, we allow for primitive sub-personal rule-application processes, which
involve no inferential sub-processes, but are rule-governed, in a thin sense, by
virtue of taking rule-representations as input.
The main point of positing this hierarchy of processes was to allow for dependencies that block the Regress in a plausible fashion. Thus, for example, in some cases, AR2 may rely on AR1, so that a consciously accessible rule is applied via an inferential sub-process whose rule is not, itself, consciously accessible. Further, we allow that such AR1 processes may rely on non-primitive, sub-personal inferences. And of course, we insist that all such cascades must at some point rely on primitive rule-application processes. Far from failing
155 to accommodate AR2, we maintain that we accommodate it, and many other sorts of inference beside.
6. Conclusion
We started by touting the prima facie explanatory virtues of IRF’s. We then argued that, by positing a cascade of different sorts of rule-application processes, IRF’s can accommodate active reasoning in a non-regressive fashion. Specifically, we argued that these resources allow proponents of IRF’s to address the Regress even in its strengthened form. We concluded by suggested that our solution does not generate any obviously untenable consequences for IRF’s. In view of this, and contrary to the opinion of many, we conclude that the Regress fails to undermine intentional rule-following accounts of reasoning.
156 References
Abelson, R. P. (1985). A variance explanation paradox: When a little is a lot. Psychological Bulletin, 97, 129–133.
Ackerman, R., & Thompson, V. A. (2017). Meta-reasoning: Monitoring and control of thinking and reasoning. Trends in Cognitive Sciences, 21, 607–617.
Allen, M. (1998). Comparing the persuasive effectiveness of one-and two-sided messages. In M. Allen & R. W. Preiss (Eds.), Persuasion: Advances Through Meta-Analysis (pp. 87–98). Cresskill, NJ: Hampton Press.
Alperson-Afil, N., Goren-Inbar, N., Herzlinger, G., & Wynn, T. (2020). Expert retrieval structures and prospective memory in the cognition of Acheulian hominins. Psychology, 11, 173–189.
Ambrose, S. H. (2012). Obsidian dating and source exploitation studies in Africa: Implications for the evolution of human behavior. In I. Liritzis & C. Stevenson (Eds.), Obsidian and Ancient Manufactured Glasses (pp. 56–72). Albuquerque: University of New Mexico.
Anderson, C. A., Lepper, M. R., & Ross, L. (1980). Perseverance of social theories: The role of explanation in the persistence of discredited information. Journal of Personality and Social Psychology, 39, 1037–1049.
Andrews, P. W., Gangestad, S. W., & Matthews, D. (2002). Adaptationism—How to carry out an exaptationist program. The Behavioral and Brain Sciences, 25, 489–504; discussion 504- 553.
Baron, J. (1995). Myside bias in thinking about abortion. Thinking & Reasoning, 1, 221–235.
Berwick, R. C., & Chomsky, N. (2016). Why Only Us: Language and Evolution. MIT Press.
Blancke, S., Boudry, M., & Braeckman, J. (2019). Reasonable irrationality: The role of reasons in the diffusion of pseudoscience. Journal of Cognition and Culture, 19, 432–449.
Bland, S. (2018). Cognitive bias, situationism, and virtue reliabilism. Synthese. https://doi.org/10.1007/s11229-018-02031-6
Blegen, N. (2017). The earliest long-distance obsidian transport: Evidence from the ∼200 ka Middle Stone Age Sibilo School Road Site, Baringo, Kenya. Journal of Human Evolution, 103, 1–19.
157 Block, N. (1995). The mind as the software of the brain. In An Invitation to Cognitive Science (2nd Ed.). Thinking: An Invitation to Cognitive Science, Vol. 3: Thinking (pp. 377–425). Cambridge, MA: The MIT Press.
Boghossian, P. (2008). Epistemic rules. The Journal of Philosophy, 105, 472–500.
Boghossian, P. (2014). What is inference? Philosophical Studies, 169, 1–18.
Bosco, F. A., Aguinis, H., Singh, K., Field, J. G., & Pierce, C. A. (2015). Correlational effect size benchmarks. Journal of Applied Psychology, 100, 431–449.
Bottero, J., Herrenschmidt, C., Vernant, J.-P., & Zabbal, F. (2000). Ancestor of the West: Writing, Reasoning and Religion in Mesopotamia, Elam and Greece. (T. L. Fagan, Trans.). Chicago: University of Chicago Press.
Boyette, A. H., & Hewlett, B. S. (2018). Teaching in hunter-gatherers. Review of Philosophy and Psychology, 9, 771–797.
Brooks, A. S., Yellen, J. E., Potts, R., Behrensmeyer, A. K., Deino, A. L., Leslie, D. E., … Clark, J. B. (2018). Long-distance stone transport and pigment use in the earliest Middle Stone Age. Science, 360, 90–94.
Broome, J. (2013). Rationality Through Reasoning. Wiley-Blackwell.
Broome, J. (2014). Normativity in reasoning. Pacific Philosophical Quarterly, 95, 622–633.
Brown, F. H., McDougall, I., & Fleagle, J. G. (2012). Correlation of the KHS Tuff of the Kibish Formation to volcanic ash layers at other sites, and the age of early Homo sapiens (Omo I and Omo II). Journal of Human Evolution, 63, 577–585.
Brown, K. S., Marean, C. W., Jacobs, Z., Schoville, B. J., Oestmo, S., Fisher, E. C., … Matthews, T. (2012). An early and enduring advanced technology originating 71,000 years ago in South Africa. Nature, 491, 590–593.
Carmody, R. N., & Wrangham, R. W. (2009). The energetic significance of cooking. Journal of Human Evolution, 57, 379–391.
Chomsky, N. (1969a). Comments on Harman’s reply. In S. Hook (Ed.), Language and Philosophy: A Symposium (pp. 152–159). New York: New York University Press.
Chomsky, N. (1969b). Linguistics and philosophy. In S. Hook (Ed.), Language and Philosophy: A Symposium (pp. 51–94). New York: New York University Press.
158 Collins, J. (2005). On the input problem for massive modularity. Minds and Machines, 15, 1–22.
Coon, C. (1971). The Hunting Peoples. Atlantic Little Brown.
Darwin, C. (1871). The Descent of Man and Selection in Relation to Sex. London: John Murray.
Dawkins, R. (2006). The Blind Watchmaker. London: Penguin Press.
Dawkins, R., & Krebs, J. R. (1978). Animal signals: Information or manipulation? In J. R. Krebs & N. B. Davies (Eds.), Behavioural Ecology: An Evolutionary Approach (pp. 282–309). Sutherland, MA: Sinauer Associates, Inc. de Boer, B. (2017). Evolution of speech and evolution of language. Psychonomic Bulletin & Review, 24, 158–162.
De Neys, W. (2018a). Bias, conflict, and fast logic: Towards a hybrid dual process future? In W. De Neys (Ed.), Dual Process Theory 2.0 (pp. 47–65). New York: Routledge.
De Neys, W. (2018b). Dual Process Theory 2.0. Routledge.
Dennell, R. (1997). The world’s oldest spears. Nature, 385, 767–768.
Dennett, D. C. (1978). Brainstorms: Philosophical Essays on Mind and Psychology. Cambridge, Mass.: The MIT Press.
Dennett, D. C. (1982). Styles of mental representation. Proceedings of the Aristotelian Society, 83, 213–226.
Diez-Martín, F., Sánchez Yustos, P., Uribelarrea, D., Baquedano, E., Mark, D. F., Mabulla, A., … Domínguez-Rodrigo, M. (2015). The origin of the Acheulean: The 1.7 million-year-old site of FLK West, Olduvai Gorge (Tanzania). Scientific Reports, 5, 17839.
Dira, S. J., & Hewlett, B. S. (2016). Learning to spear hunt among Ethiopian Chabu adolescent hunter-gatherers. In H. Terashima & B. S. Hewlett (Eds.), Social Learning and Innovation in Contemporary Hunter-Gatherers: Evolutionary and Ethnographic Perspectives (pp. 71–81). Tokyo: Springer Japan.
Ditto, P. H., & Lopez, D. F. (1992). Motivated skepticism: Use of differential decision criteria for preferred and nonpreferred conclusions. Journal of Personality and Social Psychology, 63, 568–584.
159 Doherty, M. E., Mynatt, C. R., Tweney, R. D., & Schiavo, M. D. (1979). Pseudodiagnosticity. Acta Psychologica, 43, 111–121.
Dominowski, R. L. (1995). Content effects in Wason’s selection task. In Perspectives on Thinking and Reasoning: Essays in Honour of Peter Wason (pp. 41–65). Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc.
Drayson, Z. (2012). The uses and abuses of the personal/subpersonal distinction. Philosophical Perspectives, 26, 1–18.
Dunbar, R. I. M. (2014). How conversations around campfires came to be. Proceedings of the National Academy of Sciences, 111, 14013–14014.
Dunbar, R. I. M. (2016). Human Evolution: Our Brains and Behavior. Oxford, New York: Oxford University Press.
Dunbar, R. I. M. (2017). Group size, vocal grooming and the origins of language. Psychonomic Bulletin & Review, 24, 209–212.
Dutilh Novaes, C. (2018). The enduring enigma of reason. Mind & Language, 33, 513–524.
Edwards, K., & Smith, E. E. (1996). A disconfirmation bias in the evaluation of arguments. Journal of Personality and Social Psychology, 71, 5–24.
Elqayam, S., & Over, D. E. (2016). From is to ought: The place of normative models in the study of human thought. Frontiers in Psychology, 7:628. https://doi.org/10.3389/fpsyg.2016.00628
Endicott, K. (1988). Property, power and conflict among the Batek of Malaysia. In T. Ingold, D. Riches, & J. Woodburn (Eds.), Hunters and Gatherers 2: Property, Power and Ideology (pp. 110–127). Oxford: Berg Publishers.
Evans, J. (1984). Heuristic and analytic processes in reasoning*. British Journal of Psychology, 75, 451–468.
Evans, J. (1989). Bias in Human Reasoning: Causes and Consequences. Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc.
Evans, J. (2007). Hypothetical Thinking: Dual Processes in Reasoning and Judgement. New York, NY, US: Psychology Press.
160 Evans, J. (2017). A brief history of the Wason selection task. In The Thinking Mind: A Festschrift for Ken Manktelow (pp. 1–14). New York, NY, US: Routledge/Taylor & Francis Group.
Evans, J. (2019). Reflections on reflection: The nature and function of type 2 processes in dual- process theories of reasoning. Thinking & Reasoning, 25, 383–415.
Evans, J., Clibbens, J., & Rood, B. (1996). The role of implicit and explicit negation in conditional reasoning bias. Journal of Memory and Language, 35, 392–409.
Evans, J., & Lynch, J. S. (1973). Matching bias in the selection task. British Journal of Psychology, 64, 391–397.
Evans, J., Newstead, S. E., & Byrne, R. M. J. (1993). Human Reasoning: The Psychology of Deduction. Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc.
Evans, J., & Over, D. E. (1996). Rationality and Reasoning. Oxford: Taylor & Francis.
Evans, J., & Wason, P. C. (1976). Rationalization in a reasoning task. British Journal of Psychology, 67, 479–486.
Felton, M., Crowell, A., & Liu, T. (2015). Arguing to agree: Mitigating my-side bias through consensus-seeking dialogue. Written Communication, 32, 317–331.
Fitch, W. T. (2011). The evolution of syntax: An exaptationist perspective. Frontiers in Evolutionary Neuroscience, 3. https://doi.org/10.3389/fnevo.2011.00009
Fitch, W. T. (2017). Empirical approaches to the study of language evolution. Psychonomic Bulletin & Review, 24, 3–33.
Fodor, J. A. (1968). The appeal to tacit knowledge in psychological explanation. Journal of Philosophy, 65, 627–40.
Fodor, J. A. (1975). The Language of Thought. Harvard University Press.
Fodor, J. A. (1987). Psychosemantics: The Problem of Meaning in the Philosophy of Mind (1st edition). Cambridge, Massachusetts London: A Bradford Book.
Fodor, J. A. (2000). The Mind Doesn’t Work That Way: The Scope and Limits of Computational Psychology. Cambridge, MA.: A Bradford Book.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition, 28, 3–71.
161 Gamble, C. (1993). Exchange, foraging and local hominid networks. In C. Scarre & F. Healy (Eds.), Trade and Exchange in Prehistoric Europe (pp. 35–44). Oxbow Books in association with The Prehistoric Society and the Société Préhistorique Française.
Ganapini, M. B. (2019). Confabulating reasons. Topoi, 1–13.
Garfield, Z. H., Garfield, M. J., & Hewlett, B. S. (2016). A cross-cultural analysis of hunter- gatherer social learning. In H. Terashima & B. S. Hewlett (Eds.), Social Learning and Innovation in Contemporary Hunter-Gatherers: Evolutionary and Ethnographic Perspectives (pp. 19–34). Tokyo: Springer Japan.
Gigerenzer, G. (2011). Surrogates for theory. APS Observer, 22. Retrieved from https://www.psychologicalscience.org/observer/surrogates-for-theory
Goodwin, R. Q., & Wason, P. C. (1972). Degrees of insight. British Journal of Psychology, 63, 205–212.
Grove, M., Pearce, E., & Dunbar, R. I. M. (2012). Fission-fusion and the evolution of hominin social systems. Journal of Human Evolution, 62, 191–200.
Hahn, U., & Chater, N. (1998). Similarity and rules: Distinct? exhaustive? empirically distinguishable? Cognition, 65, 197–230.
Hahn, U., & Harris, A. J. L. (2014). What does it mean to be biased: Motivated reasoning and rationality. In B. H. Ross (Ed.), Psychology of Learning and Motivation (Vol. 61, pp. 41– 102). Academic Press.
Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108, 814–834.
Haidt, J. (2012a). The Righteous Mind: Why Good People Are Divided by Politics and Religion. Knopf Doubleday Publishing Group.
Haidt, J. (2012b, February 3). Jonathan Haidt explains our contentious culture. Retrieved from https://vimeo.com/36128360
Hall, L., Johansson, P., & Strandberg, T. (2012). Lifting the veil of morality: Choice blindness and attitude reversals on a self-transforming survey. PLOS ONE, 7, e45457.
Harman, G. (1967). Psychological aspects of the theory of syntax. The Journal of Philosophy, 64, 75–87.
162 Harman, G. (1969). Linguistic competence and empiricism. In S. Hook (Ed.), Language and Philosophy: A Symposium (pp. 143–151). New York: New York University Press.
Henrich, J. (2015). The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter. Princeton: Princeton University Press.
Hewlett, B. S. (2016). Social learning and innovation in hunter-gatherers. In H. Terashima & B. S. Hewlett (Eds.), Social Learning and Innovation in Contemporary Hunter-Gatherers: Evolutionary and Ethnographic Perspectives (pp. 1–15). Tokyo: Springer Japan.
Heyes, C. (2018). Cognitive Gadgets: The Cultural Evolution of Thinking. Cambridge, Massachusetts: Belknap Press: An Imprint of Harvard University Press.
Hill, K., & Kintigh, K. (2009). Can anthropologists distinguish good and poor hunters? Implications for hunting hypotheses, sharing conventions, and cultural transmission. Current Anthropology, 50, 369–378.
Hill, K., Walker, R. S., Božičević, M., Eder, J., Headland, T., Hewlett, B., … Wood, B. (2011). Co- residence patterns in hunter-gatherer societies show unique human social structure. Science, 331, 1286–1289.
Hindley, P. C. (2015). Nominal and imperative iconic gestures used by the Khoisan of North West Botswana to coordinate hunting. African Study Monographs, 35, 149–181.
Hollitscher, W. (1939). The concept of rationalization. The International Journal of Psycho- Analysis, 20, 330–332.
Howard, M. V. A., Brewer, N., & Williams, K. D. (2006). How processing resources shape the influence of stealing thunder on mock-juror verdicts. Psychiatry, Psychology and Law, 13, 60–66.
Jenyns, L. (1860). Letter to Charles Darwin (January 4th, 1860). In F. Burkhardt, S. Evans, & A. Pearn (Eds.), Evolution: Selected Letters of Charles Darwin (1860-1870) (pp. 1–3). Cambridge: Cambridge University Press.
Johnson-Laird, P. (2008). How We Reason. Oxford: Oxford University Press.
Johnson-Laird, P. N., & Tagart, J. (1969). How implication is understood. The American Journal of Psychology, 82, 367–373.
163 Jonas, E., Schulz-Hardt, S., Frey, D., & Thelen, N. (2001). Confirmation bias in sequential information search after preliminary decisions: An expansion of dissonance theoretical research on selective exposure to information. Journal of Personality and Social Psychology, 80, 557–571.
Jones, E. (1908). Rationalization in every-day life. The Journal of Abnormal Psychology, 3, 161– 169.
Jones, N. B., & Konner, M. J. (1976). !Kung knowledge of animal behavior (or: The proper study of mankind is animals). In R. Lee & I. DeVore (Eds.), Kalahari Hunter-Gatherers: Studies of the !Kung San and Their Neighbors (pp. 325–348). Harvard: Harvard University Press.
Kennedy, K. A., & Pronin, E. (2008). When disagreement gets ugly: Perceptions of bias and the escalation of conflict. Personality and Social Psychology Bulletin, 34, 833–848.
Klayman, J. (1995). Varieties of confirmation bias. In J. Busemeyer, R. Hastie, & D. L. Medin (Eds.), Psychology of Learning and Motivation (Vol. 32, pp. 385–418). Academic Press.
Klayman, J., & Ha, Y. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94, 211–228.
Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Reasons for confidence. Journal of Experimental Psychology: Human Learning and Memory, 6, 107–118.
Kripke, S. A. (1982). Wittgenstein on Rules and Private Language: An Elementary Exposition. Harvard University Press.
Kuhn, D. (1991). The Skills of Argument. New York: Cambridge University Press.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–498.
Laland, K. N. (2018). Darwin’s Unfinished Symphony: How Culture Made the Human Mind (Reprint edition). Princeton University Press.
Laurence, S., & Margolis, E. (1997). Regress arguments against the language of thought. Analysis, 57, 60–66.
Layton, R., O’Hara, S., & Bilsborough, A. (2012). Antiquity and social functions of multilevel social organization among human hunter-gatherers. International Journal of Primatology, 33, 1215–1245.
164 Leakey, L. S. B., Tobias, P. V., & Napier, J. R. (1964). A new species of the genus Homo from Olduvai Gorge. Nature, 202, 7–9.
Lee, N. Y. L., & Johnson-Laird, P. N. (2013). Strategic changes in problem solving. Journal of Cognitive Psychology, 25, 165–173.
Lee, R. B. (1979). The !Kung San: Men, Women and Work in a Foraging Society. Cambridge University Press.
Lewis, D. M. G., Al-Shawaf, L., Conroy-Beam, D., Asao, K., & Buss, D. M. (2017). Evolutionary psychology: A how-to guide. American Psychologist, 72, 353–373.
Lew-Levy, S., Reckin, R., Lavi, N., Cristóbal-Azkarate, J., & Ellis-Davies, K. (2017). How do hunter-gatherer children learn subsistence skills? Human Nature, 28, 367–394.
Liebenberg, L. (2012). The Art of Tracking, the Origin of Science. Claremont, South Africa: New Africa Books.
Lombao, D., Guardiola, M., & Mosquera, M. (2017). Teaching to make stone tools: New experimental evidence supporting a technological hypothesis for the origins of language. Scientific Reports, 7, 1–14.
Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37, 2098–2109.
Lundholm, H. (1933). Repression and rationalization. British Journal of Medical Psychology, 13, 23–50.
MacDonald, K. (2007). Cross-cultural comparison of learning in human hunting. Human Nature, 18, 386–402.
Manktelow, K. I., & Over, D. E. (1991). Social roles and utilities in reasoning with deontic conditionals. Cognition, 39, 85–105.
Marien, H., Custers, R., Hassin, R. R., & Aarts, H. (2012). Unconscious goal activation and the hijacking of the executive function. Journal of Personality and Social Psychology, 103, 399–415.
Marlowe, F. (2010). The Hadza: Hunter-gatherers of Tanzania. University of California Press.
Marshall, L. (1959). Marriage among !Kung Bushmen. Africa, 29, 335–365.
165 Marshall, L. J. (1976). The !Kung of Nyae Nyae. Cambridge, Mass: Harvard University Press.
Mercier, H. (2017). Confirmation bias – myside bias. In R. Pohl (Ed.), Cognitive Illusions: Intriguing Phenomena in Judgement, Thinking and Memory (pp. 99–114). Psychology Place.
Mercier, H. (2020). Not Born Yesterday: The Science of Who We Trust and What We Believe. Princeton University Press.
Mercier, H., & Sperber, D. (2009). Intuitive and reflective inferences. In K. Frankish & J. Evans (Eds.), In Two Minds: Dual Processes and Beyond (pp. 149–170). Oxford University Press.
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory. Behavioral and Brain Sciences, 34, 57–74.
Mercier, H., & Sperber, D. (2017). The Enigma of Reason. Harvard University Press.
Moshman, D., & Geil, M. (1998). Collaborative reasoning: Evidence for collective rationality. Thinking & Reasoning, 4, 231–248.
Mynatt, C. R., Doherty, M. E., & Tweney, R. D. (1977). Confirmation bias in a simulated research environment: An experimental study of scientific inference. The Quarterly Journal of Experimental Psychology, 29, 85–95.
Nestler, S. (2010). Belief perseverance. Social Psychology, 41, 35–41.
Newton, E., & Roberts, M. (Eds.). (2005). Methods of Thought: Individual Differences in Reasoning Strategies. Hove: Psychology Press.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2, 175–220.
Nielsen, R., Akey, J. M., Jakobsson, M., Pritchard, J. K., Tishkoff, S., & Willerslev, E. (2017). Tracing the peopling of the world through genomics. Nature, 541, 302–310.
Nussbaum, E. M., & Kardash, C. M. (2005). The effects of goal instructions and text on the generation of counterarguments during writing. Journal of Educational Psychology, 97, 157–169.
Ohnuma, K., Aoki, K., & Akazawa, A. T. (1997). Transmission of tool-making through verbal and non-verbal communication: Preliminary experiments in Levallois flake production. Anthropological Science, 105, 159–168.
166 O’Keefe, D. J. (1999). How to handle opposing arguments in persuasive messages: A meta- analytic review of the effects of one-sided and two-sided messages. Annals of the International Communication Association, 22, 209–249.
O’Keefe, D. J. (2009). Persuasive effects of strategic maneuvering: Some findings from meta- analyses of experimental persuasion effects research. In F. H. van Eemeren (Ed.), Examining Argumentation in Context: Fifteen Studies on Strategic Maneuvering (pp. 285–296).
Oliver, R. T. (1936). Human motivation: Intellectuality, emotionality, and rationalization. Quarterly Journal of Speech, 22, 67–77.
Oswald, M., & Grosjean, S. (2004). Confirmation bias. In R. Pohl (Ed.), Cognitive Illusions: A Handbook on Fallacies and Biases in Thinking, Judgement and Memory (pp. 79–96). Psychology Press.
Perfors, A., Tenenbaum, J. B., & Regier, T. (2011). The learnability of abstract syntactic principles. Cognition, 118, 306–338.
Perkins, D. N. (1985). Postprimary education has little impact on informal reasoning. Journal of Educational Psychology, 77, 562–571.
Perkins, D. N. (1989). Reasoning as it is and could be: An empirical perspective. In Thinking Across Cultures: The Third International Conference on Thinking (pp. 175–194). Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc.
Perkins, D. N., Allen, R., & Hafner, J. (1983). Difficulties in everyday reasoning. In W. Maxwell, Thinking: The Expanding Frontier (pp. 177–189). Philadelphia: The Franklin Institute Press.
Perkins, D. N., Farady, M., & Bushey, B. (1991). Everyday reasoning and the roots of intelligence. In Informal Reasoning and Education (pp. 83–105). Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc.
Peters, U. (2019). Illegitimate values, confirmation bias, and Mandevillian cognition in science. The British Journal for the Philosophy of Science, Forthcoming. https://doi.org/10.1093/bjps/axy079
Piddington, R. (1928). Reasoning and rationalization. Australasian Journal of Psychology and Philosophy, 6, 42–54.
167 Pollard, P., & Evans, J. (1980). The influence of logic on conditional reasoning performance. Quarterly Journal of Experimental Psychology, 32, 605–624.
Putt, S. S., Wijeakumar, S., Franciscus, R. G., & Spencer, J. P. (2017). The functional brain networks that underlie Early Stone Age tool manufacture. Nature Human Behaviour, 1, 1–8.
Putt, S. S., Woods, A. D., & Franciscus, R. G. (2014). The role of verbal interaction during experimental bifacial stone tool manufacture. Lithic Technology, 39, 96–112.
Pylyshyn, Z. W. (1980). Computation and cognition: Issues in the foundations of cognitive science. Behavioral and Brain Sciences, 3, 111–132.
Ragni, M., Kola, I., & Johnson-Laird, P. N. (2018). On selecting evidence to test hypotheses: A theory of selection tasks. Psychological Bulletin, 144, 779–796.
Reeve, H. K., & Sherman, P. W. (2007). Why measuring reproductive success in current populations is valuable: Moving forward by going backward. In S. W. Gangestad & J. A. Simpson (Eds.), The Evolution of Mind: Fundamental Questions and Controversies (pp. 86–94). Guilford Press.
Reich, S. S., & Ruth, P. (1982). Wason’s selection task: Verification, falsification and matching. British Journal of Psychology, 73, 395–405.
Richter, D., Grün, R., Joannes-Boyau, R., Steele, T. E., Amani, F., Rué, M., … McPherron, S. P. (2017). The age of the hominin fossils from Jebel Irhoud, Morocco, and the origins of the Middle Stone Age. Nature, 546, 293–296.
Rips, L. J. (1994). The Psychology of Proof: Deductive Reasoning in Human Thinking. MIT Press.
Robinson, J. H. (1921). The Mind in the Making: The Relation of Intelligence to Social Reform. New York, NY, US: Harper & Brothers.
Ryle, G. (1949). The Concept of Mind. Chicago: University of Chicago Press.
Sandgathe, D. M., & Berna, F. (2017). Fire and the genus Homo: An introduction to Supplement 16. Current Anthropology, 58, S165–S174.
Scalise Sugiyama, M. (2017). Oral storytelling as evidence of pedagogy in forager societies. Frontiers in Psychology, 8. https://doi.org/10.3389/fpsyg.2017.00471
168 Schaeken, W., Vooght, G. D., Vandierendonck, A., d’Ydewalle, G., & d’Ydewalle, G. (Eds.). (1999). Deductive Reasoning and Strategies. Mahwah, N.J: Routledge.
Schwarz, N. (2004). Metacognitive experiences in consumer judgment and decision making. Journal of Consumer Psychology, 14, 332–348.
Semaw, S., Rogers, M. J., Quade, J., Renne, P. R., Butler, R. F., Dominguez-Rodrigo, M., … Simpson, S. W. (2003). 2.6-Million-year-old stone tools and associated bones from OGS-6 and OGS-7, Gona, Afar, Ethiopia. Journal of Human Evolution, 45, 169–177.
Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3–22.
Smith, D., Schlaepfer, P., Major, K., Dyble, M., Page, A. E., Thompson, J., … Migliano, A. B. (2017). Cooperation and the evolution of hunter-gatherer storytelling. Nature Communications, 8, 1853.
Smith, K. M., & Apicella, C. L. (2019). Hadza hunter-gatherers disagree on perceptions of moral character. Social Psychological and Personality Science, 1948550619865051.
Smolensky, P. (1988). On the proper treatment of connectionism. Behavioral and Brain Sciences, 11, 1–23.
Sperber, D. (1994). The modularity of thought and the epidemiology of representations. In Mapping the mind: Domain specificity in cognition and culture (pp. 39–67). New York, NY, US: Cambridge University Press.
Sperber, D. (2000). Metarepresentations in an evolutionary perspective. In D. Sperber (Ed.), Metarepresentations: A Multidisciplinary Perspective (pp. 117–137). Oxford University Press.
Sperber, D. (2001a). An evolutionary perspective on testimony and argumentation. Philosophical Topics, 29, 401–413.
Sperber, D. (2001b). In defense of massive modularity. In Language, brain, and cognitive development: Essays in honor of Jacques Mehler (pp. 47–57). Cambridge, MA, US: The MIT Press.
Sperber, D. (2013). Speakers are honest because hearers are vigilant: Reply to Kourken Michaelian. Episteme, 10, 61–71.
169 Sperber, D., Cara, F., & Girotto, V. (1995). Relevance theory explains the selection task. Cognition, 57, 31–95.
Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G., & Wilson, D. (2010). Epistemic vigilance. Mind & Language, 25, 359–393.
Stahl, C., Klauer, K. C., & Erdfelder, E. (2008). Matching bias in the selection task is not eliminated by explicit negations. Thinking & Reasoning, 14, 281–303.
Stanovich, K. E., & West, R. F. (2008). On the failure of cognitive ability to predict myside and one-sided thinking biases. Thinking & Reasoning, 14, 129–167.
Stanovich, K. E., West, R. F., & Toplak, M. E. (2016). The Rationality Quotient: Toward a Test of Rational Thinking. Cambridge, Massachusetts: The MIT Press.
Sterelny, K. (2018). Why reason? Hugo Mercier’s and Dan Sperber’s The Enigma of Reason: A New Theory of Human Understanding. Mind & Language, 33, 502–512.
Stibbard-Hawkes, D. N. E., Attenborough, R. D., & Marlowe, F. W. (2018). A noisy signal: To what extent are Hadza hunting reputations predictive of actual hunting skills? Evolution and Human Behavior, 39, 639–651.
Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50, 755–769.
Taylor, W. S. (1923). Rationalization and its social significance. The Journal of Abnormal Psychology and Social Psychology, 17, 410–418.
Tennie, C., Call, J., & Tomasello, M. (2009). Ratcheting up the ratchet: On the evolution of cumulative culture. Philosophical Transactions of the Royal Society B: Biological Sciences, 364, 2405–2415.
Thieme, H. (1997). Lower Palaeolithic hunting spears from Germany. Nature, 385, 807–810.
Tomasello, M. (2001). The Cultural Origins of Human Cognition. Cambridge, Mass.: Harvard University Press.
Toplak, M. E., & Stanovich, K. E. (2003). Associations between myside bias on an informal reasoning task and amount of post-secondary education. Applied Cognitive Psychology, 17, 851–860.
170 Tryon, C. A., McBrearty, S., & Texier, P.-J. (2005). Levallois lithic technology from the Kapthurin Formation, Kenya: Acheulian origin and middle stone age diversity. African Archaeological Review, 22, 199–229.
Turnbull, C. M. (1965). Wayward Servants: The Two Worlds of the African Pygmies. Greenwood Press.
Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79, 281– 299.
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232.
Van Duyne, P. C. (1973). A short note on Evans’ criticism of reasoning experiments and his matching response hypothesis. Cognition, 2, 239–242.
Waldron, V. R., Caughlin, J., & Jackson, D. (1995). Talking Specifics: Facilitating Effects of Planning on AIDS Talk in Peer Dyads. Health Communication, 7, 249–266.
Walker, E. R. (1928). I. Reasoning and rationalization. Australasian Journal of Philosophy, 6, 149–151.
Walton, D. (2009). Anticipating objections in argumentation. In H. Ribeiro (Ed.), Rhetoric and Argumentation in the Beginning of the XXIst Century (pp. 87–109). Coimbra: Coimbra University Press.
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. The Quarterly Journal of Experimental Psychology, 12, 129–140.
Wason, P. C. (1966). Reasoning. In B. Foss (Ed.), New Horizons in Psychology (pp. 135–151). Harmondsworth: Penguin Books.
Wason, P. C. (1968). “On the failure to eliminate hypotheses...” – a second look. In P. C. Wason & P. N. Johnson-Laird (Eds.), Thinking and Reasoning: Selected Readings (pp. 165–174). Penguin.
Wason, P. C., & Evans, J. (1974). Dual processes in reasoning? Cognition, 3, 141–154.
Wason, P. C., & Johnson-Laird, P. N. (1972). Psychology of Reasoning: Structure and Content. Oxford, England: Harvard University Press.
171 Wason, P. C., & Shapiro, D. (1971). Natural and contrived experience in a reasoning problem. Quarterly Journal of Experimental Psychology, 23, 63–71.
Wetherick, N. E. (1962). Eliminative and enumerative behaviour in a conceptual task. Quarterly Journal of Experimental Psychology, 14, 246–249.
White, A. (2017). A model-based analysis of the minimum size of demographically-viable hunter- gatherer populations. Journal of Artificial Societies and Social Simulation, 20, 9.
Wiessner, P. (2014). Embers of society: Firelight talk among the Ju/’hoansi bushmen. Proceedings of the National Academy of Sciences of the United States of America, 111. https://doi.org/10.1073/pnas.1404212111
Williams, G. C. (1966). Adaptation and Natural Selection: A Critique of Some Current Evolutionary Thought. Princeton, N.J., Princeton University Press.
Williams, K. D., Bourgeois, M. J., & Croyle, R. T. (1993). The effects of stealing thunder in criminal and civil trials. Law and Human Behavior, 17, 597–609.
Winterhalder, B. (1986). Diet choice, risk, and food sharing in a stochastic environment. Journal of Anthropological Archaeology, 5, 369–392.
Wolfe, C. R., & Britt, M. A. (2008). The locus of the myside bias in written argumentation. Thinking & Reasoning, 14, 1–27.
Wolfe, C. R., Britt, M. A., & Butler, J. A. (2009). Argumentation schema and the myside bias in written argumentation. Written Communication, 26, 183–209.
Woodburn, J. (1982). Egalitarian societies. Man, 17, 431–451.
Wrangham, R., & Conklin-Brittain, N. (2003). Cooking as a biological trait. Comparative Biochemistry and Physiology Part A: Molecular & Integrative Physiology, 136, 35–46.
Wright, C. (2014). Comment on Paul Boghossian, ‘“What is inference.”’ Philosophical Studies, 169, 27–37.
Wynn, T., & Coolidge, F. (2011). How levallois reduction is similar to, and not similar to, playing chess. In A. Nowell & I. Davidson (Eds.), Stone Tools and the Evolution of Human Cognition (pp. 83–103). Boulder, Colorado: University Press of Colorado.
172