There is robust experimental evidence that Ss who are required to check whether P→Q regularly overlook the relevance of ~Qs. So, Ss asked to verify (1), though they routinely want to know what the under-18s are drinking, only rarely remember to ask the non-Coke drinkers whether they are under 18 (Wason 1966).

1. (1) If someone is under 18 (s)he is drinking Coke.

2. (2) It's required that if someone is under 18 (s)he drinks Coke.

By contrast, Ss who are told that (2) is a regulation and asked to check whether everyone is in compliance reliably remember to ask anyone not drinking Coke how old (s)he is. It appears that what sort of drinking is going on is somehow more salient if you're evaluating (2) than if you're evaluating (1). Why on earth is that?

One explanation, recently widely bruited, is that we are innately equipped with special, domain-specific, modular mechanisms for cheater detection, and that these mechanisms are better at their job than the other circuits we use for coping with hypotheticals. (See Cosmides and Tooby [1992] and references therein.) The reason we have this high-performance equipment available, it is further explained, is that it would have been useful for us to have it back when we were heavily into hunting and gathering. (A similar theory would account for our uncanny innate ability to navigate according to the earth's magnetic field – such a comfort if you're driving home late from a hunt or a gather – except that we haven't got one.