The Wason card selection task (Wason 1966) is probably the most famous demonstration of illogical human reasoning. It involves showing people four cards, each with a letter on one side and a number on the other, with the rule that “if there is a vowel on one side, there is an even number on the other”:
Participants must test whether this rule is true by selectively turning over one or more cards. Logic would dictate that in the above task, only the A and 7 cards should be turned over, as the K and 4 could not disprove the rule whatever is on the other side, and if they confirm the rule, this does not act as proof that the rule is true. However, only 4% of participants gave this correct answer (Johnson-Laird & Wason 1976).
However, other researchers have pointed out that participants do perform well on the Wason selection task in certain adaptations of the task. For example, if the rule is a familiar one, such as “If a person is drinking beer, then they must be over 18 years old” (with cards: Beer / Coke / 22 / 16), then participants’ performances vastly improve, with 73% giving the right answer (of choosing ‘Beer’ and ‘16’) in this case (Griggs & Cox 1982). Cheng & Holyoak (1985) found that even if the rule is not explicitly familiar, but the task uses thematic material (such as transit status and different innoculations) and participants are given a real-life scenario where they might have to implement the rule (in this case, imagining that they are a border officer checking travellers have the right innoculations), participants do much better (92% correct) than if they are given just the thematic material with no rationale (61% correct).
Cosmides argued that people are better at the above tasks because they involve detecting cheats (people who take benefits without paying for them). This ability of cheat-detection was likely to have been highly evolutionarily adaptive, and hence we may have specially evolved built-in “cheat-detection modules”. Cosmides found that in cost-benefit versions of the Wason task (where there is potential for detecting ‘cheating’), participants perform much better than in non-cost-benefit versions (80% vs 45% correct, in her study). However, the problem with this explanation is that it does not account for all successful reasoning (e.g. participants generally do well with the rule “if you clear up spilt blood, then you must wear rubber gloves”, even though clearing up spilt blood without gloves does not really equal ‘cheating’, as Manktelow & Over 1991 pointed out). Also, there is a danger of inventing new brain modules to explain every aspect of human behaviour, without any neurophysiological evidence to back this up.
It may be that formal logic is not the right normative theory we should be comparing human behaviour to anyway. Probability theory may actually prove more suited to solving real world reasoning problems than logic, as real world problems usually involve elements of uncertainty that can be taken into account by probability theory, but not by formal logic, which requires statements to be either true or false.
Chater & Oaksford (1994) used the probabilistic approach when developing their Optimal Data Selection theory, which explains people’s “illogical” choices on the Wason task as actually being optimal in terms of expected information gain, and hence quite rational.