In critiquing proposed examples of adaptive misbelief, McKay & Dennett (M&D) aim to confirm the assumption that, via evolution, humans “have been biologically engineered to form true beliefs” (sect. 1, para. 2), and conclude that the exchange rate between fitness and truth “is likely to be fair in most circumstances” (sect. 15, last para). We agree with their critiques (Murray & Moore Reference Murray and Moore2009; Schloss Reference Schloss, Bubulia, Sosis, Harris, Genet, Genet and Wyman2007), but the conclusion does not follow.
First, even if many proposals for adaptive misbelief fail, this does not tell us whether adaptive misbeliefs spawned in situations M&D acknowledge as credible are common or, as they claim, limited to “certain rarefied contexts” (sect. 15, last para). For instance, the promiscuous attribution of agency and teleology, or the manifold positive illusions that may be accounted for within error management theory (Johnson Reference Johnson and Levin2009) are extraordinarily plentiful, persistent, and influential. Other kinds of positive illusions, from the placebo effect to magnifying the virtues of beloved people, places, nations, and traditions – consistent with proposals for resource commitment strategies – may be even more plentiful and powerful. And contrary to M&D's tempting suggestion, such positive illusions are not restricted to subjective beliefs that are “not likely to be rudely contradicted by experience” (sect. 14, para. 3). Many of these widespread beliefs entail almost delusional denials of repeated experience. Notions that Eros lasts forever, this time it's real, and (as the sappy song says) “When we're hungry, love will keep us alive” are effective and virtually ubiquitous catalysts for reproductive pairbonding. But by non-reproductive periods of the human life cycle, those Romeos whose romantic illusions have not killed them, have oft' yielded to the wisdom of Friar Lawrence: “These violent delights have violent ends, and in their triumph die…Therefore love moderatedly: long love doth so. Too swift arrives as tardy as too slow.” Yet another category altogether, unexamined by M&D, is selection for cognitive extravagance independent of problem-solving utility (Miller Reference Miller2000; Reference Miller2001). But even granting M&D's conclusion that there are just a few families of adaptive misbelief, we don't yet know enough about their natural history to determine how many species there are or what their carrying capacities and competitive coefficients are relative to true beliefs.
Second, even if reproductively beneficent misbeliefs are rare and most misbeliefs have costs, this does nothing to tell us how well evolution ultimately avoids such costs. Indeed, M&D elegantly acknowledge that functional normativity does not entail statistical normality: In evolution, forgivable malfunctions may be common and achieving proper function may be “positively rare” (sect. 3, para. 5). Thus, even if truth is the evolutionary target as M&D maintain, design constraints, by-product associations, and historical contingencies may make it one that cognition has a low probability of hitting.
Third, many kinds of beliefs – from debates over quantum theory to discussions of metaphysics – have no clear reproductive relevance at all. How, and whether, such beliefs are related to cognitive mechanisms that have been selected for veracity is uncertain (Cromer Reference Cromer1993; Wolpert Reference Wolpert2000). What does not seem uncertain is that manifold beliefs do not influence behaviors or the behaviors they do influence are not reproductively salient. Belief-forming mechanisms generate variety that, analogous to neutral polymorphisms (Kimura Reference Kimura1991), may be unpruned by the adaptive consequences of their truth or falsity. Indeed, the capacity for some degree of cognitive licentiousness may itself be an adaptation to the “uncertain futures problem” (Plotkin Reference Plotkin1997; Wagner Reference Wagner2005).
Finally, M&D's conclusion requires the falsity not only of the above ways in which selection fails to exclude misbelief, but also of the more global but controversial thesis that nothing at all about the process of natural selection serves to favor truth-conducive cognitive tools (Churchland Reference Churchland1987; Plantinga Reference Plantinga and Beilby2002; Stich Reference Stich1990).
On selectionist accounts of the origin of mind, beliefs and belief forming mechanisms are selected by virtue of their capacity to support adaptive behavior or internal states. Thus, belief forming mechanisms will be selected when they yield (i) a representational model that orients organisms towards adaptive behaviors, and/or (ii) a correlational source of arousal or inhibition that serves to motivate adaptive (or inhibit maladaptive) behavior. The question then becomes: Are models that are true better at orienting organisms towards adaptive behaviors, or are true beliefs better at arousing effective desires for adaptive behaviors? From what we know about the action of natural selection, the most prudent answer may be: “There is no reason to think so.”
Why is there no reason to think so? Because (in science, and in belief generally) models need only to “save appearances” in order to be successful. Consider the task of designing “thinking” robots for a competition in which the winners were duplicated (with minor program variations) for future competitions. While one would surely seek to program competing robots to form beliefs that provided an isomorphic “map” of the external environment, would one further seek to program beliefs about that environment that were true? Not obviously. Indeed, there are numerous ways of programming the robot to “conceptualize” its environment that, while representationally biased or even radically false, are nonetheless (a) appropriately isomorphic and (b) reliably adaptive behavior-inducing. Such programs would be adaptive.
What is true of programmed learning robots is true of selection-designed cognition. Dennett has aptly commented, “Lying behind, and distinct from, our reasons are evolutionary reasons, free-floating rationales that have been endorsed by natural selection” (Dennett Reference Dennett2006a, p. 93). Our reasons (in better moments) are truth-seeking; natural selection's are fitness seeking. We cannot know if, in achieving its reasons, selection allows us also to achieve ours.
Of course, one might respond that just because our belief-forming mechanisms are liable to error in these domains does not mean that they are routinely or irremediably unreliable (after all, we often discover our errors, like the cognitive biases mentioned above). But this offers little reassurance, since the seeming discovery of error relies on comparing beliefs to other beliefs which, for all we know, are comparably unreliable, though perhaps for different reasons.
Richard Dawkins has commented that “however many ways there are of being alive, it is certain that there are vastly more ways of being dead” (Dawkins Reference Dawkins1996, p. 9). The same is true of being right and wrong. Natural selection is immensely effective at weeding out ways of not being alive. It is unclear how well it fares in culling ways of not believing truly.
