R2. Oversimplify and self-monitor

As several commentators (e.g., Boyer; Sutton) point out, cognitive systems are necessarily compromises that have to honor competing demands in one way or another. Since time is of the essence, the speed-accuracy tradeoff is critical; cost also matters so “fast and frugal” systems or methods (Gigerenzer & Goldstein Reference Gigerenzer and Goldstein1996; Gigerenzer et al. Reference Gigerenzer, Todd, the and Research1999) are often the order of the day. But these can generate errors in abundance, so if the animal can afford it, it is good to have a meta-system of one kind or another in place, monitoring the results, discarding bad outputs when they arise, and shifting methods if possible. Good advice, then, in both animal design and artifact design, is oversimplify and self-monitor (Dennett Reference Dennett1984a).

The BBS format enables this strategy, and we followed it in our target article. Our deliberately oversimplified definition of belief set the table for a variety of useful commentaries showing just how complicated these issues truly are. We had a lot of ground to survey, so we decided, pragmatically, to paint with broad strokes, and to come back later (in this response to commentary) with the called-for corrections. As several commentators (e.g. Cokely & Feltz; Gjersoe & Hood; Liddle & Shackelford; Wereha & Racine) point out, our hyper-general definition of belief, as “a functional state of an organism that implements or embodies that organism's endorsement of a particular state of affairs as actual” (target article, sect. 1, para. 1), blurs the oft-proposed boundaries between a range of arguably distinct types of cognitive states. (It is also worth remembering that in the working vocabularies of many people, the everyday term “belief” is restricted to matters of great moment only – religious belief, political creed, and other topics of capital-b Belief – and would not be used to discuss one's current perceptual state or whether there was beer in the fridge.) We did discuss, and approve of, Gendler's (2008) alief/belief distinction, and Ainslie puts it to good use, also reminding us (in personal correspondence) of Gendler's useful mnemonic characterization, which we should have quoted in the target article:

But we did not even mention, as Frankish points out, the acceptance/belief distinction, which, he argues, may turn out to play a key role: A pragmatic acceptance is not, strictly speaking, a misbelief at all; and our prime candidates for adaptive misbeliefs, positive illusions, may be voluntarily adopted policies, not involuntarily imposed biases – in us, if not in other animals incapable of such “metacognitive” evaluations. This leads Frankish to a sketch of an experimental paradigm well worth pursuing. Flanagan and Konečni raise similar objections. Flanagan comments on the strategic role of statements of belief in competitive contexts, but notes that there is nothing epistemically disreputable about believing that one can win: “‘can’ does not entail ‘will.’” Further on, however, he makes a telling slip: “Both players, if they are any good, go into the match believing that they can win, indeed that they will win” (our emphasis). Flanagan is right that there is no mistake in believing that one can win, or in hoping that one will win. But where both players believe that they will win, we have misbelief (although not necessarily unreasonable misbelief; each may have compelling reasons for expecting to win). Insofar as such misbelief boosts confidence and enables honest signaling of such confidence, it may be adaptive. Like Frankish and Flanagan, Konečni suggests that positive illusions may represent doxastically uncommitted action policies. Haselton & Buss and Johnson, however, take roughly the opposite view, arguing that genuine (mis)beliefs may generate adaptive behavior more effectively than cautious action policies. We return to their commentaries further on.

It is tempting to re-baptize acceptance as c-lief, since acceptance stands to belief roughly as belief stands to alief, a more sophisticated and expensive state, reserved now for just one species, humans. (Cf. Dennett's Reference Dennett1978 belief/opinion distinction, which is explicitly modeled on betting on the truth of a sentence which one believes [not alieves] to be true.) But this won't help resolve all the confusions, since, as Krebs & Denton observe, collaborative positive illusions (e.g., “I'm OK, you're OK”) may begin as pragmatic policies or acceptances – we are, in Haidt's (Reference Haidt2001) nice observation, intuitive lawyers, not intuitive [truth-seeking] scientists – but among the effects in us are unarticulated cognitive tendencies that may be best seen as akin to aliefs – except for not being antecedent to all other cognitive attitudes.

R3. Is our evolutionary thinking naïve?

Several commentators challenge our frankly adaptationist reasoning as naïve, and our “reverse engineering” perspective on misbelief is seen as blinkered or distorting. The points they raise are instructive, but serve rather to expose the weaknesses of various standard objections to adaptationism. Wilks, for instance, sees ghosts of Lamarck and Sheldrake's morphic resonances (!) in our project, and suggests that even Chomsky's curious views on evolution have more plausibility than ours. Wilks “cannot see what all this has to do with evolution, understood as natural selection of traits inherited through the genome,” and indeed, from that pinched perspective, it is not surprising that he would miss the point. Natural selection is not just about “traits inherited through the genome.” Perhaps he was misled by the fact that we carefully distinguished – as some do not – between genetic fitness and human happiness; but we also went to some lengths to note the role of gene-culture coevolution, and nowhere did we restrict natural selection to genetic evolution (see sects. R5 and R7). Wilks expresses doubts about how “brain modifications might conceivably affect the gametes,” ignoring the Baldwin Effect (Deacon Reference Deacon1997; Dennett Reference Dennett1991; Reference Dennett1995a; Reference Dennett, Weber and Depew2003b), a particularly clear path by which surprisingly specific talents can migrate from brain modifications into the genome. (Fear of Lamarckian heresy has prevented many from taking the Baldwin Effect seriously; it is not heretical biology.) His comparison with Fodor's notorious example of an innate concept of telephone is simply a straw man. There are plenty of well-proven cases in which ecologically significant contents of considerable specificity are genetically transmitted. The fear of snakes exhibited by laboratory-raised monkeys and small children who have never seen a snake (Mineka et al. Reference Mineka, Davidson, Cook and Keir1984; LoBue & DeLoache Reference LoBue and DeLoache2008), for example, or the species-specific nest-building dispositions of birds that have never seen such a nest being built should temper his incredulity.

Especially in the case of behavior regulators, there is typically an interplay, a coordination, between genetically transmitted features and culturally (or “socially”) transmitted elements. Marcus is right that it does not follow that if something could be learned, it must be learned; it might well be innate, but also vice versa, as Avital and Jablonka (Reference Avital and Jablonka2000) show: Many long-presumed innate animal “instincts” turn out to be learned behaviors, copied in one way or another from parents' behavior, not part of their genetic legacy, as a host of cross-fostering studies demonstrate. The genes fix the disposition to attend to what the parents do, but the rest is up to environmental transmission. What this fact brings to our attention is that Mother Nature is not a gene-centrist! Where genetic evolution leaves off and developmental and, indeed, “empiricist” (Marcus) psychological learning takes over, is an entirely open option, with large differences between fairly closely related species. (Consider the wide variation in the extent to which species-typical bird song is innate.) Marcus's example of “learning to walk” is useful, since, as he says, there is an innate stepping reflex in humans that exists at birth. On top of this reflex comes something we can still call learning to walk. His point is not that walking is innate in humans – it isn't, when compared with, say, the walking (indeed, running) skills found in a newborn antelope. Where innate instinct leaves off and learning begins is not a line that can be, or need be, sharply drawn. It goes without saying, we thought, that belief-generating mechanisms depend critically on environmental input, but we should have said it anyway, as several commentators (e.g., Dunning; Dweck) chide us for underestimating the importance of environmental variation. So we agree with Dweck, Liddle & Shackelford, and Wilks that individual, isolated beliefs are unlikely to be the target of genetic selection, but that does not imply that quite specific biases could not be incorporated into our genetically transmitted equipment. We may in effect be primed to imprint on whatever in the environment fills a certain fairly specific doxastic role, much as newly hatched ducklings imprint on the first large moving thing they see and follow it.

Similarly, as a number of commentators reveal, the line between by-product and adaptation is not sharp at all. Every adaptation, after all, must emerge from something that varies “randomly” (under no selection) or from some prior arrangement that persists for other reasons, and what had heretofore been a by-product is brought into focus and enhanced and exploited by selective pressure. Showing that something is (likely) a by-product does not rule out the possibility that there is (already, as it were) opportunistic selective pressure on it. The bright colors of autumn foliage of deciduous trees in New England are probably just a by-product of the chemistry of chlorophyll loss after leaf death (though this has recently been challenged by evidence that it signals either inhospitality or vigor to aphids looking for a winter home; see Yamazaki Reference Yamazaki2008); but whether or not aphids are attracted to, or repelled by, bright autumn colors, assuredly there is now selective advantage to having brilliant autumn color in New England. The economies of Vermont, New Hampshire, and Maine benefit significantly from the autumn “leaf-peepers” (foliage enthusiasts) that invade, and hence there is a pronounced bias against cutting down handsome trees and for planting, or encouraging the growth of, the most colorful variants.

Adaptationists know – or should know, since the classic work of George Williams (Reference Williams1966) – that the evidential demands for establishing an adaptation are greater than the demands for discovering a mere by-product. As Millikan says, “If certain kinds of errors are common and also systematically useful, it does not follow that they are common because they are useful.” It does not follow, but fortunately there are ways of testing to see if and when such adaptationist hypotheses are true. Sometimes, however, the tests are too impractical to carry out (they might require a few thousand years of observation of evolution, for instance), and often the adaptation is so obvious, once discovered, that nobody bothers challenging the claim. It is interesting that the charge of “Just So Story” leveled at adaptationists is almost entirely reserved for hypotheses dealing with features of human evolution. A brief canvassing of textbooks of biology will find literally thousands of examples of confidently asserted adaptationist claims that have never been challenged and never been thoroughly tested – claims about the functions of enzymes, the functions of organs, the functions of behaviors (of protists, animals, plants…). People get touchy when their own organs and behaviors are analyzed from an adaptationist perspective, but unless they are prepared to dismiss the mountains of insight to be found in the rest of biology, they should stop treating “Just So Story” as a handy-dandy wild-card refutation-device. It is no such thing.

Critics of adaptationism are right, however, that there is a perilous amount of free scope in the range of permissible hypotheses. For instance, why does nature so often counteract one bit of flawed design with another, compensatory one, instead of just “fixing” the first? Maybe there is a constraint – so far unknown – that renders the latter course impossible or more expensive (see discussion of Haselton & Buss further on, and see McKay & Efferson [under review] for a discussion of constraints in the context of error management theory). Such chains of reasoning are not just flights of fancy, since there are differential consequences that can usually be tested for, but until such tests are conducted, we are left with merely plausible conjectures. This open-endedness haunts the discussions further on in this response – see, for example, our discussion of Johnson – since the question to which commentators continually return is whether it is cheaper or easier for the mind to deceive itself with misinformation than to provide accurate information and adjust its prudential policies to fit the risk. Until we can assess this by evaluating known cognitive mechanisms and their evolutionary costs, this question must remain unsettled.

Wereha & Racine chant the standard evo-devo mantra, claiming that “by reverse engineering the beliefs of adult humans” we forgo a developmental analysis – which is true enough, but does it matter in this case? They are right, of course, that environmental interactions, especially those that engage language, are crucial, and for that reason cultural-genetic interactions are necessary. What we disagree with is their claim that evo-devo considerations obviate or even blunt the effectiveness of reverse-engineering approaches. One simply has to do one's reverse-engineering with more attention to the myriad possibilities raised by developmental demands. One way of putting Wereha & Racine's main claim is this: Because development, from embryo on, is a process that has to protect the robustness of the organism at every stage, later (e.g., “adult”) features could just as easily be leftovers, fossil traces, of features that paid for themselves in infancy as features that pay for themselves in adulthood. That is, indeed, a distinct possibility that needs to be considered. And Gjersoe & Hood provide a possible example: the entrenchment phase in hypothesis formation in childhood development. This oversimplification strategy has a huge payoff: Oversimplify and (eventually) self-monitor, but in those who are not particularly reflective, a tendency to cling uncritically to one's first hypothesis might be a residue of a particularly adaptive bias in childhood that has outlived its usefulness.

R4. Adaptive oversimplifications

Oversimplifications that make cognitive life easier are also proposed by Zawidzki, who notes that Dennett himself (1991) has argued that the concept of a self is a benign “user illusion.” Zawidzki also notes that much of the specificity of self-interpretation may be an artifact of societal demand, adaptive in the context of complex sociality. We indeed overlooked the role of such oversimplifications as instances of adaptive misbelief, probably because, like the concept of a center of gravity, they can quite readily be recast as something like strategic metaphors rather than falsehoods (consider the widespread understanding that there is nothing pejorative in the everyday understanding of the “user illusion” that makes laptops so user-friendly). We agree that they make an illuminating further category to explore (see also the discussion of free will later).

In this light, Bertamini & Casati could be seen as suggesting that naïve physics is also an instance of oversimplify and self-monitor, on a hugely different time scale. We are only in recent centuries beginning to discover the falsehoods latent in our everyday conception of the world, a conception that is, as they say, “prima facie veridical” in that it does not “interfere with our interactions with the world.” This pragmatic effectiveness, of course, is the evolutionary rationale for the default assumption that true beliefs are adaptive and misbeliefs are not. As Millikan observes, “it is getting straight about what is in front of our noses that is the first order of importance for us.” (Boyer says that what matters for adaptive design is “that the circumstances in question be such that decision-making does not lead to excessive vulnerability.”) Not bumping into dangerous things, and finding food, shelter, and mates requires a certain amount of effective information-gathering – and not misinformation-gathering. Wilson & Lynn view the fact that our senses give us only a narrow window on the physical variation in the available environmental stimuli as “deception,” but that is unwarranted; sensory systems that provide a truncated or edited message that informs may not give the whole truth, while giving, normally, nothing but the truth. The obvious norm for information-gathering is not to be deflected by the motivational system, since wishful thinking is typically unrealistic and sometimes catastrophically so (Ainslie). At the same time, as various commentators note, there can be overriding reasons for editing the information-gathering to accomplish various palliative ends. If the truth hurts too much, it will disable, not enable, the intentional agent.

R5. Illusions and collusions

Although a number of commentators (Ackerman et al.; Brown; Gjersoe & Hood; Krebs & Denton) endorse our claim that positive illusions represent sound candidates for adaptive misbelief, others are skeptical. First, there are methodological and statistical concerns. For example, although Cokely & Feltz argue that adaptive misbeliefs are in general much more widespread than we allowed, they point out that better-than-average effects can represent statistical artifacts. The fact that driving ability has a negatively skewed distribution means that most drivers simply are better-than-average; the mean is clearly an inappropriate measure of central tendency in this case. We do not dispute this, but we note that better-than-average effects are also documented for normally distributed traits, and (as Kruger et al. note) the effects replicate when other, similar methodological points are taken into consideration.

A different sort of concern has to do with the contexts in which positive illusions are observed. Some commentators (Dweck, Flanagan, Konečni, and Kruger et al.) appear to suggest that if such illusions are a product of genetic evolution, they should not be confined to a particular culture or to a particular historical epoch. Moreover, as Dunning observes, they should be particularly evident in tasks with adaptive significance. No one has performed the latter analysis across task contexts, as Dunning notes, but there are data about the cross-cultural replicability of positive illusions. Unfortunately, however, there does not appear to be consensus on this issue: Brown states that “positively-biased self-perceptions are a pervasive, cross-cultural phenomenon,” but Dweck, Flanagan, and Kruger et al. express doubts about the cultural universality of positive illusions, noting that they are more reliably documented in Western societies. In any case, we note here that cultural variability is by no means a decisive datum against the evolutionary claim: if cultural evolution plays a coevolutionary role, there may be, in effect, cultural subspecies of evolved misbelief.

Konečni claims that positive illusions are a feature of a particular historical period: “the recently terminated era of easy credit.” We enjoyed his commentary, and await empirical substantiation of this claim. We are confused, however, by his methodological critique of studies that purport to demonstrate positive illusions regarding participants' children. On the one hand, Konečni complains that such “studies have presumably not polled the opinions of the parents (including potential ones) who terminated pregnancies – or committed infanticide, physical and/or sexual abuse.” His implication at this point seems to be that offspring-directed positive illusions are an artifact of biased sampling. He goes on, however, to suggest that had such parents been polled, they would have demonstrated positive illusions regarding their children; this, he suggests, would undermine the suggestion that biased offspring appraisals facilitate parental care. We think Konečni is trying to have his cake and eat it too. Leaving aside pregnancy terminations (which were presumably uncommon in ancestral environments), we agree that demonstrations of offspring-directed positive illusions in abusive parents would undermine the evolutionary argument, but we doubt that such parents would harbor these illusions. This doesn't mean, however, that the finding of widespread offspring-directed positive illusions is a statistical artifact – that depends on whether parental care is normally distributed (abusive parents may represent a “bump” at the lower end of the distribution), and on how much of the distribution was sampled by the relevant studies. But if it is an artifact, it's nevertheless a telling one, because it implicates a positive correlation between offspring-directed positive illusions and parental care, consistent with our evolutionary suggestion.

Kruger et al. raise more serious concerns in their measured and informative commentary. They question whether positive illusions are the norm for healthy individuals, and they point out the many instances of systematic negative (self) illusions. (Ackerman et al. also speak of negative illusions, but they refer to illusions that are negative with respect to others.) We appreciate this point, but we note that the existence of negative self illusions is not in itself problematic for claims about the adaptive significance of positive self illusions – although it may, as Bertamini & Casati recognize, demonstrate that the relevant mechanisms are domain-specific rather than domain-general. As Haselton and colleagues have noted, a tendency toward false positives may be adaptive in certain adaptive contexts (as in the male sexual overperception bias that Haselton & Buss describe; see sect. R7 for further discussion), whereas a tendency toward false negatives may be adaptive in others (as in the female commitment underperception bias that Haselton & Buss report elsewhere; see Haselton & Buss Reference Haselton and Buss2000; see also Ackerman et al.'s comments about the benefits of being “hard to get”). Different domains will call for biases in different directions. It is worth citing Hartung's (Reference Hartung, Lockard and Paulhus1988) speculations about the adaptive value of negative self illusions in certain circumstances, what he calls “deceiving down.” It remains to be demonstrated, of course, that the negative illusions that Kruger et al. mention are domain-specific adaptations.

Dunning also emphasizes the role of environmental context (as does Dweck), noting that misbeliefs often arise because the environment fails to furnish the information needed to form accurate judgments. Illusions, on this view, reflect forgivable design limitations rather than design features. Pessimistic predictions about the trustworthiness of others may persist not because they are fitness-enhancing, as Ackerman et al. suggest, but because they are liable to confirmation but not to refutation. Marcus makes a similar point, noting that illusions may reflect the operation of a general confirmation-bias mechanism rather than dedicated domain-specific machinery. We note that even if a general confirmation-bias mechanism generates illusions as a well-entrenched subclass of outputs, the serendipitous benefits that those outputs provide might “protect” the confirmation bias mechanism (which does, after all, output a lot of mistaken cognition) from counter-selection, helping to “pay for” its persistence. If underestimations of others' trustworthiness are less costly than overestimations (Ackerman et al.), then a mechanism that generates underestimations (initially) as a by-product may be a candidate for exaptation.

Wilson & Lynn also mention the confirmation bias, linking it to the motive force of strong or “hot” affect. Marcus suggests that positive illusions are potentially underpinned by motivated reasoning, but we see this possibility as a potential generalization of – rather than necessarily an alternative to – the evolutionary claim we defended. As Ainslie discusses, in an extraordinarily rich and compressed commentary, motivational and affective forces represent the proximal mechanisms by which natural selection tethers belief to survival. Our abilities to appraise evidence dispassionately may be selectively sabotaged (motivationally biased) in adaptive domains, yielding positive illusions. Motivated reasoning might not be adapted, however – it might be largely a by-product of the selection for increasing intelligence that Ainslie describes, a process which enabled our ancestors to discover the intervening carrots and sticks of reward, and to begin devising ways of getting the carrots without going to the trouble of checking on the world. Human imagination was born, with all its costs and benefits. The result has been “the unhitching of reward from adaptiveness” and in the ensuing holiday of imagination, we have had to create methods of epistemic self-control to protect ourselves from our own freedom. Exploring the further wrinkles Ainslie draws to our attention will have to wait for another occasion.

Wilson & Lynn point out that inflated self-esteem can come at a cost. We acknowledge that the connections between self-esteem and adaptive behaviours are complex, but we didn't suggest that unrestrained self-esteem would be adaptive, and we cited Baumeister (Reference Baumeister1989) on the “optimal margin of illusion.” In fact, because self-esteem is such a heterogeneous concept we avoided using the term at all in our target article. The large survey that Wilson & Lynn cite points out that the category “self-esteem” encompasses a range of subtypes. For example, Jordan et al. (Reference Jordan, Spencer, Zanna, Hoshino-Browne and Correll2003) characterized the defensive subtype as involving a discrepancy between high explicit (conscious) and low implicit (unconscious) self-esteem. These authors found that individuals with this discrepancy were significantly more narcissistic than individuals high in both explicit and implicit self-esteem. Discrepancies between implicit and explicit self-esteem have also been implicated in the formation of persecutory delusions (Bentall & Kaney Reference Bentall and Kaney1996; Kinderman & Bentall Reference Kinderman and Bentall1996; Reference Kinderman and Bentall1997; McKay et al. Reference McKay, Langdon and Coltheart2007b; Moritz et al. Reference Moritz, Werner and von Collani2006). To the extent that illusory positive self-views are adaptive, therefore, we would predict them to be held at both conscious and unconscious levels.

The above-discussed commentaries provide valuable correctives to our enthusiasm for positive illusions and the evolutionary implications thereof. We acknowledge that more research is needed to clarify whether illusional beliefs are reliably observed in specific adaptive contexts, and whether they trend in the expected directions. We also note, however, that a number of commentaries complement and extend our analysis of positive illusions. We have already mentioned Gjersoe & Hood's suggestion that the developmental phase of theoretical entrenchment involves an adaptive positive illusion – “overconfidence in the generalisability of one's theory.” Brown and Krebs & Denton detail the important role that people play in validating and perpetuating the illusions of others (Wilson & Lynn make this point about false beliefs more generally). In our target article we noted how the positive illusions of parents with respect to their co-parents and their children could strengthen familial bonds and facilitate parental care. Brown, however, notes that infants also benefit by internalizing the positive illusions of their parents with respect to themselves. Krebs & Denton point out that individuals may manipulate others into validating their own positive self illusions, but they also appreciate (as does Brown) that this process can be collaborative and mutually beneficial.

Boyer and Sutton describe how our own memory systems can be co-conspirators in the maintenance of adaptive illusions. Both commentators note that selection does not indulge abstract epistemic concerns – memories need be accurate, therefore, only insofar as they are fitness-enhancing. Memories that are accurate for accuracy's sake are a biological luxury, so adaptive considerations may frequently trump epistemic considerations. The result is that many of our memories, as beliefs about past occurrences, may be examples of adaptive misbelief.

R6. Delusions and doxastic shear pins

To provide a framework for our discussion, we developed a tentative taxonomy of misbelief. We began by distinguishing two general types: misbeliefs arising in the course of normal doxastic functioning, and misbeliefs resulting from some kind of break in normal functioning. Liddle & Shackelford draw our attention to a similar analysis by Wakefield (Reference Wakefield1992). Wakefield's (1992; see also Reference Wakefield1999a; Reference Wakefield1999b) concern is to provide a rigorous theoretical grounding for the concept of (mental) disorder. His analysis incorporates both a value component (disorders are harmful, where “harm” is judged by the standards of the relevant culture) and an evolutionary component (disorders reflect the failures of internal mechanisms to carry out their naturally selected functions). We endorse Wakefield's analysis – and regret not previously being aware of it – but note that his project is wider and more general than ours, distinguishing function from dysfunction in naturally selected mechanisms insofar as this distinction can guide decisions about candidates for disorder, while we are interested in belief specifically.

As “disorders of belief,” delusions represent the key area of overlap between Wakefield's analysis and ours. Our emphasis was on delusions as the output of belief-formation mechanisms that have ceased to perform their normal (naturally selected) functions – Wakefield's evolutionary criterion. His value criterion, however, is clearly also important: Delusions are harmful insofar as they occasion distress and insofar as they jeopardize the social and occupational functioning of individuals who hold them (this is the “clinical significance” criterion in the DSM-IV-TR; American Psychiatric Association 2000). In our target article we were wary of considering delusions as adaptive, and indeed we labeled them instances of “doxastic dysfunction” (although we didn't clearly discriminate between biological and social conceptions of dysfunction). We did, however, speculate about a class of misbeliefs enabled by the action of system components designed to break: doxastic shear pins. Several of our commentators (Langdon; Liddle & Shackelford; Millikan; Mishara & Corlett) pick up on this concept.

Langdon notes that if doxastic shear pins exist, their shearing should involve some kind of neurocognitive “short-circuit” rather than a stable neuropsychological impairment. We agree with this point. She also distinguishes neuropsychological (deficit) and motivational answers to the question of why deluded individuals cling to their delusions. Mishara & Corlett consider this distinction an “overly strict conceptual schism,” and it is true that motivational and deficit hypotheses need not be mutually exclusive (Langdon is well aware of this). Although we found it useful to distinguish between misbeliefs that represent functionless departures from normal operation (“culpable design limitations”) and those that incorporate some functional component, we did acknowledge the porous nature of such conceptual boundaries. Nevertheless, we remain open to the possibility of misbeliefs with purely “deficit” aetiologies. Mishara & Corlett, however, favor the doxastic shear pin perspective: delusions accommodate aberrant prediction error signaling, disabling flexible conscious processing and enabling the preservation of habitual responses in the context of impaired predictive learning mechanisms. As such they serve a functional, even biologically adaptive, role. We appreciate this perspective, and we think that work on prediction errors represents a key avenue of research into delusions. Inferences about biological adaptiveness, however, may be unjustified here: As Millikan notes, the existence of doxastic shear pins “does not imply that failures to function properly are helpful, but only that in some circumstances it is best not to attempt to function at all.”

Coltheart raises issues of truth and groundedness with respect to delusions, and asks us to clarify whether we consider well-grounded false beliefs to be misbeliefs. The short answer to this is, Yes. Misbeliefs are simply false beliefs – they may be grounded or ungrounded. Grounded misbeliefs reflect forgivable design limitations: in contexts of imperfect information (we may be underinformed or even deliberately misinformed), misbeliefs are inevitable. Ungrounded misbeliefs, on the other hand, may result from culpable failures in naturally selected belief mechanisms (delusions), but they might also reflect designed features of such mechanisms (the adaptive misbeliefs we sought in the target article). Talmont-Kaminski claims that ungrounded beliefs fall outside our compass, but he seems to have misunderstood our expository strategy. Ungrounded beliefs are within our compass, but only insofar as such beliefs are false. This was perhaps overly stipulative, but it made our discussion manageable (for example, it allowed us to skirt moral beliefs and beliefs about norms more generally). As we stated in our target article, we do not expect adaptive misbeliefs to be generated by mechanisms designed to produce beliefs that are false per se. Rather, we implicate evolved tendencies for forming domain-specific ungrounded beliefs. Where these beliefs are (contingently) false, we will see adaptive misbelief. Where they are (contingently) true, they fall outside our purview.

Not all ungrounded beliefs, of course, are adaptive: once again, we argue that such beliefs often reflect breakdowns in belief formation machinery, and where such beliefs are harmful (Wakefield's value criterion), they constitute delusions. But here, too, ungrounded beliefs can be contingently true, as in the delusional jealousy example that Coltheart elaborates. We are quite happy for such cases of “serendipitously” true belief (or “accidentally” true, as Coltheart prefers) to count as instances of delusion; they are just not instances of misbelief. Where misbelief is concerned, truth is the critical feature, simply by (our) definition; where delusion is concerned, truth may ultimately be irrelevant (see Leeser & O'Donohue Reference Leeser and O'Donohue1999; Spitzer Reference Spitzer1990). In cases such as the delusional jealousy scenario that Coltheart outlines, truth or falsity may be difficult to establish – a feature that may contribute to the incorrigibility of such beliefs (many religious beliefs also have this feature; see discussion later). It is worth noting, however, that delusions can resist the presentation of manifestly contradictory evidence; indeed, as Mishara & Corlett show, such evidence may even strengthen delusional conviction through the process of reconsolidation (see Corlett et al. 2009).

Sperber makes the interesting point that most human beliefs are acquired via communication with others. Because of this, he doubts that most human beliefs are grounded in the sense of being “appropriately founded on evidence and existing beliefs.” We appreciate Sperber's general point, but our view is that the testimony of others (whether oral or written) is ultimately just another source of evidence that should be weighed up when forming beliefs. We look up at the sky and form a belief about whether it will rain; later we listen to the weather forecast and revise our belief accordingly. The evidence of testimony may be easier to override than direct perceptual evidence (Langdon discusses the idea that delusions involve a loss of the ability to override the latter), but it is evidence that can ground belief nevertheless. We don't see any reason to consider beliefs acquired via communication to be ungrounded. Sperber notes that “from a cognitive and social science point of view, a definition of ‘belief’ that excludes most religious beliefs renders itself irrelevant.” We agree with this, and we think the same of any definition of “grounded” that excludes beliefs acquired by communication. Such a definition would guarantee its own irrelevance.

R7. Error management theory and religion

Haselton & Buss and Johnson pick up on our point about how adaptive behavioral biases need not reflect adaptive biases in belief. We do not doubt that when the costs of relevant errors in a given domain are recurrently asymmetric, selection should implement a bias toward committing less costly errors (Haselton & Nettle Reference Haselton and Nettle2006). Our point was that such biases need not involve a systematic departure from Bayesian belief revision, but merely judiciously biased action policies (see McKay & Efferson [under review] for a more thorough, technical treatment of these issues). A second point we made was that even when selection in accordance with the error management principle plausibly results in biased belief-forming processes, such processes may produce misbeliefs as tolerable by-products rather than as adaptations (such biased systems may be adaptive not by virtue of the misbeliefs they produce, but by virtue of the fact that they minimize misbeliefs of a certain type). Millikan endorses this point.

Haselton & Buss provide a valuable counterpoint to our skepticism regarding whether certain error management examples might qualify as examples of adaptive misbelief. They observe that a demonstration that selection can solve such adaptive problems without misbelief is not a demonstration that selection has solved such problems without misbelief. Selection might have followed any number of design trajectories, subject to the physical, economic, historical, and topographical constraints that we mentioned; it is an empirical question which trajectory was in fact followed. Haselton & Buss go on to suggest several reasons why biased beliefs might have featured in the solution to such adaptive problems. We are not convinced by their first suggestion, that such beliefs “could provide the motivational impetus for courtship behavior.” Judicious action policies, after all, would also provide that. Their next suggestion, that male misbeliefs about the sexual intent of women might help allay fears of rejection, is not obviously different from their first: presumably this is also a point about motivational impetus. It is not clear why selection would go to the trouble of instilling fears of rejection and then installing biased beliefs to allay those fears, but again, it is an empirical matter which trajectory was in fact followed. Haselton & Buss's final suggestion seems more promising to us: The confidence boost that biased beliefs provide might be attractive to females in and of itself. In a related analysis, Ackerman et al. imply that female misbeliefs about the commitment intentions of men might heighten the desires of potential suitors, leading to increased male investment and ultimately boosting the romantic returns to the females concerned. We have already discussed the similar points that Brown and Krebs & Denton make about how misbeliefs can transform the psychological states of others.

Johnson takes an error management approach to supernatural belief, and argues that such belief is adaptive. His claim is that selection should favor belief in supernatural agents because such beliefs would yield exaggerated estimates of the risk of one's social transgressions being detected. In our target article we indicated that we did not think there was strong evidence for this theory. Johnson has several points to make about the priming evidence we reviewed, but none of these points seem to help his case. First, he notes that the religious primes used by researchers tend to be culturally specific – typically derived from Western Judeo-Christian traditions. The issue of cultural specificity is important, especially as regards genetic evolutionary claims (see our remarks earlier concerning the cross-cultural validity of positive illusions), but how should it apply here? We did not contest the findings that religious primes increase prosocial behavior – instead we queried whether such primes exert their effects by activating reputational concerns involving supernatural agents, and we also queried whether such effects are mediated by religious belief. Johnson then states that experiments may not “differentiate the behavior of ‘believers’ and ‘non-believers’ – Joe Bloggs may be an avowed atheist who, on his way to Las Vegas, is nevertheless very concerned about seeing a black cat or wearing his lucky jacket or what his grandmother would have said.” We are not sure we follow this – we don't see the relevance of such superstitious beliefs to the supernatural watcher hypothesis that Johnson advocates. We do, however, acknowledge Johnson's point that many different belief systems might play the role of his “supernatural watcher” – karmic beliefs in comeuppance might inhibit social transgressions just as effectively as beliefs in personal punitive deities.

A further point that Johnson makes concerns the conclusions that can be drawn from priming studies. Evidence that supernatural primes promote prosocial behavior does not, he says, prove that supernatural beliefs are adaptive – such effects “could be evidence that religious primes turn people into suckers who give away precious resources.” We are confused by this point. The supernatural watcher hypothesis states that belief in supernatural agents inhibits antisocial behavior and is adaptive by virtue of that fact. Priming studies enable demonstrations of a causal link between religious priming and prosocial behavior. What kind of evidence would Johnson think relevant if not this? He doesn't specify. Perhaps the problem is that Shariff and Norenzayan (Reference Shariff and Norenzayan2007) reported an increase in prosocial behavior (Dictator Game donations) following religious priming, whereas Johnson's theory requires a decrease in antisocial behavior. If so, we draw attention to Randolph-Seng and Nielsen's (Reference Randolph-Seng and Nielsen2007) study, which found that participants primed with religious words cheated significantly less than controls on a subsequent task. The problem, from our perspective, is that this study could not empirically adjudicate between the supernatural watcher hypothesis and an alternative, behavioral priming, interpretation (Randolph-Seng does not appear to dispute this point). The same limitation, we argued, applies to the studies of Pichon et al. (Reference Pichon, Boccato and Saroglou2007) and Shariff and Norenzayan (Reference Shariff and Norenzayan2007).

Norenzayan, Shariff, & Gervais (Norenzayan et al.) pick up on this point, noting that supernatural watcher and behavioral priming mechanisms need not be mutually exclusive; they might well operate in tandem, and could even be mutually reinforcing. Nevertheless, these authors marshal evidence that provides support for the supernatural watcher account and yet resists a behavioral-priming interpretation. We appreciate their reference to the study of Dijksterhuis et al. (Reference Dijksterhuis, Preston, Wegner and Aarts2008), although we worry that the baby is discarded with the bathwater here: This study disambiguates the felt presence of a supernatural agent from prosocial outcomes, certainly, but only by dispensing with a prosocial component altogether (this is not a gripe about the study itself, but about its interpretation vis-à-vis the supernatural watcher hypothesis). In general, however, we find the arguments of Norenzayan et al. to be quite persuasive. In particular, we are impressed by the results of the Gervais and Norenzayan (Reference Gervais and Norenzayan2009) study that they mention. The finding that religious primes activate public self-awareness is exactly the kind of result that is needed to substantiate the supernatural watcher hypothesis. We are keen to learn whether such reputational awareness moderates the magnitude of the primes' effect on prosocial behavior.

Norenzayan et al. attribute (mis)belief in supernatural agents to cultural rather than genetic evolution. Although, by their lights, religion does not therefore supply a case of evolved misbelief, we did not intend to restrict our analysis of adaptive misbelief to cases of genetic evolution. On the contrary, we are open, at least in principle, to the possibility that culturally selected religious beliefs constitute adaptive misbeliefs. Talmont-Kaminski, Wilson & Lynn, and Zawidzki provide related analyses. The accounts of Talmont-Kaminski and Wilson & Lynn are in fact almost identical – like Norenzayan et al., they view “religion as a cultural phenomenon that exapts existing cognitive by-products” (Talmont-Kaminski). Wilson & Lynn thus suggest that the tension between by-product and adaptation explanations of religion can be defused: Both camps might be right – the by-product proponents where genetic evolution is concerned and the adaptation proponents where cultural evolution is concerned.

Along with Johnson, Talmont-Kaminski remarks upon the lack of falsifiability of religious beliefs, and outlines several barriers, physical and social, to the exposure of religious belief as false. As Talmont-Kaminski notes, it is precisely because of such barriers to testability that supernatural beliefs are well suited to serving a functional role. Sperber provides an indispensable analysis of an additional class of barriers: barriers to comprehension. For a belief to be open to epistemic evaluation, he notes, it must have a propositional content, a truth value. Many religious beliefs, however, have only “semi-propositional” content – they are mysterious and obscure, permitting manifold exegeses. (Sperber's concept of semi-propositional attitudes has not, alas, been influential among the philosophers who have devoted their careers to elucidating “classical” propositional attitudes. We can hope that a new generation of more empirically minded philosophers will eventually see the utility, indeed the inescapability, of acknowledging this set of at least belief-like phenomena.) According to Sperber, such beliefs are better suited to playing an adaptive role than many beliefs with ordinary propositional content: “content unproblematically open to epistemic evaluation might either raise objections within the relevant social group, or, on the contrary, be too easily shared beyond that group.”

Bulbulia & Sosis propose yet another variety of beliefs (or belief-like states) whose function is not strictly to inform (or misinform) the believers about the layout of their world: cooperative commitments. Following Schelling, they suggest that a certain sort of commitment problem might be solved by something like a group myth that gets everybody on the same page, as one says. The commitment problem is this: Getting individuals to cooperate can be like herding cats, but if the cats can be transformed into something more like sheep, by inculcating a religious myth in them all, this may create points of salience that engender the sorts of uniformity of attitude and synchrony of response that make large scale cooperative projects feasible. Once initiated, such a phenomenon might become more or less self-sustaining without any knowing supervision. Indeed, too much knowingness might subvert the whole enterprise, breaking the spell and tumbling everyone back into their feline individuality. It is important to note that if such a phenomenon did evolve (mainly by cultural evolution, one must suppose, with perhaps some genetic predisposition favoring it), individuals could be strongly motivated to resist any developments that threatened to undermine their obliviousness to the motivational source of their “conviction” or “faith” – without needing to know why they were so motivated. As usual, those who were blessed (by natural selection) with the disposition to behave in this way would be the beneficiaries of this clever arrangement without anybody needing to understand the cleverness of it all – until Schelling came along.

Wilson & Lynn give a vivid account of the ubiquity of deception in human culture, but seem to forget the adaptiveness of deceiving others. In the target article we set this topic aside as too obvious to need more than a brief review: Of course, it is often “adaptive” for kings to deceive their subjects, for generals to deceive their troops, for everyone to deceive their enemies. Who benefits – cui bono (Dennett Reference Dennett1995a; Reference Dennett2006a) – from the false religious and social propaganda that they describe? Wilson & Lynn apparently assume that if it is not the individuals themselves whose fitness is enhanced by believing these falsehoods, it is the groups to which they belong – an instance of group selection utilizing cultural, not genetic, evolution. Again, we are open to this possibility – but we note that these authors overlook the other possibility proposed and defended by Dennett (Reference Dennett1995a, Reference Dennett2006a): It may be the memes' own fitness that is enhanced by these adaptations, in which case these are instances of other-deception or host-manipulation, not group selection at all. One of the benefits of the memetic perspective is that it exposes the non sequitur in any argument that claims that some features are ubiquitous among groups and (hence) must be adaptive to those groups that have them. In order to establish religion as a case of (culturally selected) adaptive misbelief, one must show that individuals or groups that acquire religious cultural variants have an advantage over those not similarly “infected.” We think the jury is still out, and await evidence of this selective advantage.

R8. Truth or consequences

The ways in which the truth of beliefs can be divorced from their consequences for survival may be myriad, but they do not extend as far as Schloss & Murray propose. Like some other commentators, they think the case for adaptive (or fitness-neutral) misbelief is stronger than we allow. Our view, Schloss & Murray claim, “requires the falsity of” the radical claims of Churchland, Plantinga, and Stich, and we agree; we think those views are clearly false, for reasons presented elsewhere (on Churchland and Plantinga, see Dennett Reference Dennett2009 and forthcoming; on Stich, see Dennett Reference Dennett1981; Reference Dennett1985). As Millikan notes, to say – as Stich does – that natural selection does not care about truth, is like saying that:

Schloss & Murray also, we think, underestimate the force of Quine's observations on systematic falsehood discussed by us, and their thought experiment about the robot competition can nicely expose the issue:

Schloss & Murray are apparently imagining something like this: First the roboticist writes a program that captures all the relevant information in a behavioral “map” – and, to make the software development easier, all the nodes and action-representations are given true labels (“cliff” means cliff and “wall” means wall and “go left” means go left, etc.); and then, once the system is up and running and well tested, the roboticist goes back and systematically replaces “cliff” with “street” and “go left” with “jump” and so forth, for all the terms in the program. Now, it seems, the robot believes that when it reaches the street it should jump, where before it believed that when it reached the cliff it should go left – but since the “isomorphism” is preserved, it actually turns left when approaching the cliff, just as before – it is like the nearsighted Mr. Magoo, only more so! All its “false beliefs” conspire to keep it out of harm's way. But, as Quine (among others) observed, what the nodes mean, what content they actually have, is not determined by their labels, but by their myriad connections with each other and the world. The robot still has mainly true beliefs, but they are misleadingly “expressed” in the imagined internal labels. We think it is failure to appreciate this point that underlies much of the skepticism about the force of our default presumption. The explanation of the behavioral success of any successful organism must be in terms of how its sense organs inform it about its behavioral environment. Misinformation can only “work” against a broad background of information.

A final example: Suppose people started saying, to everybody they encountered, “You're the most wonderful person I've met!” Perhaps initially this would have a benign effect, perking everyone up a little, but of course the effect would soon fade and the utterance would become the one-word synonym for “hello”: “urthemoswunnerfulpersonimet.” Why? Because utterances can only mean, in the long run, what their hearers take them to mean, and when utterers can no longer reasonably expect their hearers to take them to mean what their words “literally” mean, they can no longer have the intention of communicating by those words what the words used to mean, and then the words can no longer mean what they used to mean (“literally”). There is no way of divorcing what the subject believes, overall, from how the subject acts, so if an internal “danger to the left!” warning reliably leads the animal to jump left, not right, then the meaning of “left” and “right” in the animal's representation system must have reversed – or it must have inverted its “policies” somehow. So for evolution to discover a move, a design, that reliably misleads an organism (in an adaptive direction) it must be that the organism for one reason or another cannot make the Quinean adjustment, or it is evolutionarily cheaper, more robust, for the organism to actually lie to itself than to make the policy adjustments that would do the adaptive thing, given the truth about the situation.

R9. The “illusion of conscious will”?

We had initially hoped to devote space in the target article to belief in free will as a candidate for adaptive misbelief, but the topic is huge and space limitations obliged us to postpone it altogether, so we are pleased that Mishara & Corlett and Randolph-Seng raise the issue. As in our treatment of the “user-illusion” (see earlier, on Zawidzki), we think that there is a strong case to be made that this is best seen not as a useful falsehood, an enabling myth that we expose at our periļ but rather, as simply an important true belief, once it is properly unpacked and laundered of obsolete connotations.

Some (e.g., Blackmore Reference Blackmore1999; Crick Reference Crick1994; Wegner Reference Wegner2002) have argued that science has shown that we don't have free will. Others are compatibilists (e.g., Dennett Reference Dennett1984b; Reference Dennett2003a; Fischer Reference Fischer1994; Fischer & Ravizza Reference Fischer and Ravizza1998; Frankfurt Reference Frankfurt1988; Mele Reference Mele1995). Dennett, for instance, has argued that although there are varieties of free will that science has plausibly shown not to exist, there are others that are unscathed, and they are the varieties that matter. Belief in them is indeed crucial to our mental health (to put it crudely) but these are true beliefs, compatible with what science has discovered, and is likely to discover, about the mechanisms of human choice. Does what one believes about the reality of free will make a discernible difference? Vohs and Schooler (Reference Vohs and Schooler2008) show that students who read a passage (from Crick Reference Crick1994) assuring them that free will is a myth are more likely to cheat in a subsequent opportunity to win money. Like Dweck's results, this finding might motivate a policy of deliberate myth-making – to try to preserve whatever shreds of responsibility remain in the wake of scientific self-knowledge – but since myth-maintenance is probably a losing battle even in the short run, for the reasons we have reviewed, a more stable policy might be to wean ourselves from the brittle traditional concepts; so that Crick's message turns into a socially bland observation about the emptiness of an obsolete concept, not a subversive blow to the integrity of our self-image as responsible agents. The fact that this healthy perspective is a hard sell, perennially challenged by the all too obvious intuition that “real” free will requires something like a miracle, may be indirect evidence that we are not just “natural-born dualists” (Bloom Reference Bloom2004) but natural-born believers in incompatibilist versions of free will as well. Such (false) beliefs may indeed have been adaptive in the past, enabling our ancestors to face life's decisions unburdened by misbegotten worries about causation and fatalism, but that does not make them necessary for mental health or effectiveness today.