McKay & Dennett (M&D) think that systematically adaptive misbeliefs are unstable – that over time they are revealed to be false, much as in the case of the boy who cried wolf. This need not be the case. Systematic falsehood can only be discounted or accommodated if its truth can be investigated effectively. Some claims are particularly difficult to investigate in some contexts, however. The reasons are threefold (explored in Talmont-Kaminski Reference Talmont-Kaminski2009). First, their content may be such as to impede investigation. Supernatural claims, for example, typically involve entities that are invisible, shy, or just far, far away. Second, there may be social taboos against investigating such claims. Durkheim's (Reference Durkheim1912) category of the sacred singles out precisely this kind of social barrier against investigation. Third, which claims can be investigated depends upon which scientific tools are available. Opposition to scientific development ensures that some misbeliefs remain uninvestigated. If people think the wolf is invisible, they won't be surprised when they don't see it. Similarly, if looking for it would break some social mores or would, for example, require heat-sensing cameras they cannot access, people would probably be held back from checking the boy's claims.

Without the possibility of investigation, beliefs become untethered from their truth-value so their popularity and stability are free to be determined by the idiosyncrasies of the human belief-forming system (BFS) as well as any functions that they might have – it is the untestable beliefs that can be most readily moulded to best serve their function. The problem, therefore, is not belief that runs counter to evidence but belief without evidence. Such ungrounded belief is deemed by M&D to fall outside their compass, but their reasons for claiming this are faulty. The first problem is that M&D conflate grounded beliefs with beliefs produced by a BFS that normally produces grounded beliefs. Yet, a BFS that normally produces grounded beliefs may (and in the human case, does) occasionally generate, as a by-product, beliefs that are not properly grounded. The second, and connected, problem is that M&D take Bayes' Theorem to establish what it means for a belief to be grounded. When I have unlimited time and resources available to update beliefs, I'll be sure to do so. In the meanwhile, it is more realistic to use the same standard as that used by Gigerenzer (e.g., Gigerenzer et al. Reference Gigerenzer and Todd1999), to whom M&D repeatedly refer. Bounded rationality theory (Simon Reference Simon1996; Wimsatt Reference Wimsatt2007) sets a realistic standard because it works with the real limitations of the BFS instead of abstracting away from them. Thanks to that, it is also able to explain why the BFS leads to systematic misbelief, that is, systematic bias – such misbelief being the necessary cost of using cognitive heuristics that are fast enough and frugal enough for humans. Indeed, the theory is robust enough to link particular heuristics with particular kinds of systematic misbelief – this is what Rozin & Nemeroff (Reference Rozin, Nemeroff, Gilovich, Griffin and Kahneman2002) does in formulating the contagion heuristic that is responsible for feelings of disgust associated with clean objects. (M&D interpret Rozin in terms of alief instead of belief, but the distinction is not relevant to this point nor does it seem strong enough to carry the weight they place upon it.) Bayes' Theorem cannot cope with any of this except to make post hoc allowances for it.

We now have almost all the tools to see why, despite M&D's less than clear stance, religion provides us with the paradigmatic case of adaptive misbelief (full account to be presented in Talmont-Kaminski, in preparation). Supernatural beliefs, in general, are particularly hard to investigate. Ghosts, fairies, or even just luck are not the kinds of things that would be easy to “catch in the act.” The sacred status that such beliefs often have, as well as the opposition to scientific development that holding them often appears to be connected to, further impede investigation, rendering them effectively untestable. The cognitive science of religion, as developed by Atran (Reference Atran2004), Boyer (Reference Boyer2001) and others, provides us with a plethora of examples such as Barrett's (Reference Barrett2000) hyperactive agent detection device that can be best understood as cognitive heuristics that lead to supernatural beliefs as a by-product. Significantly, this by-product approach seems to explain all manner of supernatural beliefs, including superstitions, without necessarily explaining the specificity of religious beliefs. Life has a habit of putting by-products to work, however.

According to the account of religion put forward by D. S. Wilson (Reference Wilson2002), hardly mentioned by M&D even though it does present religion as adaptive, the job of religious beliefs is to improve group cohesion, thereby allowing religious groups to compete more successfully against other groups. However, just as the by-product account was unable to distinguish religion from superstition, so Wilson is unable to distinguish religion from other group-oriented ideologies. Both problems disappear when we consider religion as a cultural phenomenon that exapts the existing cognitive by-products – an ideology that puts superstitions and the cognitive mechanisms underlying them to work. The result is beliefs that, thanks to their content as well as their social and scientific contexts, are well guarded against falsification and, therefore, relatively free to change as their functions require; that, thanks to the bounded and systematically biased nature of human reason, seem highly plausible; and that motivate people to act in ways adaptive for the group as well as, in the best of times, even for its members – an example of just the kind of gene-culture co-evolution that M&D are looking for.