… an absence of reasoning is to be expected when people already hold some belief on the basis of perception, memory, or intuitive inference, and do not have to argue for it. Say, I believe that my keys are in my trousers because that is where I remember putting them. Time has passed, and they could now be in my jacket, for example. However, unless I have some positive reason to think otherwise, I just assume that they are still in my trousers, and I don’t even make the inference (which, if I am right, would be valid) that they are not in my jacket or any of the other places where, in principle, they might be. In such cases, people typically draw positive rather than negative inferences from their previous beliefs. These positive inferences are generally more relevant to testing these beliefs. For instance, I am more likely to get conclusive evidence that I was right or wrong by looking for my keys in my trousers rather than in my jacket (even if they turn out not to be in my jacket, I might still be wrong in thinking that they are in my trousers). We spontaneously derive positive consequences from our intuitive beliefs. This is just a trusting use of our beliefs, not a confirmation bias (see Klayman & Ha 1987). [...]
One of the areas in which the confirmation bias has been most thoroughly studied is that of hypothesis testing, often using Wason’s rule discovery task (Wason 1960). In this task, participants are told that the experimenter has in mind a rule for generating number triples and that they have to discover it. The experimenter starts by giving participants a triple that conforms to the rule (2, 4, 6). Participants can then think of a hypothesis about the rule and test it by proposing a triple of their own choice. The experimenter says whether or not this triple conforms to the rule. Participants can repeat the procedure until they feel ready to put forward their hypothesis about the rule. The experimenter tells them whether or not their hypothesis is true. If it is not, they can try again or give up.
Participants overwhelmingly propose triples that fit with the hypothesis they have in mind. For instance, if a participant has formed the hypothesis “three even numbers in ascending order,” she might try 8, 10, 12. As argued by Klayman and Ha (1987), such an answer corresponds to a “positive test strategy” of a type that would be quite effective in most cases. This strategy is not adopted in a reflective manner, but is rather, we suggest, the intuitive way to exploit one’s intuitive hypotheses, as when we check that our keys are where we believe we left them as opposed to checking that they are not where it follows from our belief that they should not be. What we see here, then, is a sound heuristic rather than a bias.
This heuristic misleads participants in this case only because of some very peculiar (and expressly designed) features of the task. What is really striking is the failure of attempts to get participants to reason in order to correct their ineffective approach. It has been shown that, even when instructed to try to falsify the hypotheses they generate, fewer than one participant in ten is able to do so (Poletiek 1996; Tweney et al. 1980). Since the hypotheses are generated by the participants themselves, this is what we should expect in the current framework: The situation is not an argumentative one and does not activate reasoning. However, if a hypothesis is presented as coming from someone else, it seems that more participants will try to falsify it and will give it up much more readily in favor of another hypothesis (Cowley & Byrne 2005). The same applies if the hypothesis is generated by a minority member in a group setting (Butera et al. 1992). Thus, falsification is accessible provided that the situation encourages participants to argue against a hypothesis that is not their own. [...]
When one is alone or with people who hold similar views, one’s arguments will not be critically evaluated. This is when the confirmation bias is most likely to lead to poor outcomes. However, when reasoning is used in a more felicitous context – that is, in arguments among people who disagree but have a common interest in the truth – the confirmation bias contributes to an efficient form of division of cognitive labor.
When a group has to solve a problem, it is much more efficient if each individual looks mostly for arguments supporting a given solution. They can then present these arguments to the group, to be tested by the other members. This method will work as long as people can be swayed by good arguments, and the results reviewed in section 2 show that this is generally the case. This joint dialogic approach is much more efficient than one where each individual on his or her own has to examine all possible solutions carefully. The advantages of the confirmation bias are even more obvious given that each participant in a discussion is often in a better position to look for arguments in favor of his or her favored solution (situations of asymmetrical information). So group discussions provide a much more efficient way of holding the confirmation bias in check. By contrast, the teaching of critical thinking skills, which is supposed to help us overcome the bias on a purely individual basis, does not seem to yield very good results (Ritchart & Perkins 2005; Willingham 2008).
See also Mercier & Sperber 2011 on confirmation bias: