In October, I had the pleasure of attending a symposium on the “Argumentative Theory of Reasoning” at which the keynote speaker was psychologist Hugo Mercier, co-author with Dan Sperber of a landmark paper published in Behavioral and Brain Sciences in 2011 entitled “Why do humans reason? Arguments for an argumentative theory.” Mercier’s presentation was organized by the Centre for Research in Reasoning, Argumentation and Rhetoric at the University of Windsor (Ontario). In this post, I’ll summarize in my own words the key points and lines of argument in Mercier’s presentation. (More details can be found in the original Sperber-Mercier paper, available online).

What is Reasoning?

According to Mercier, reasoning is the cognitive process in which reasons are used to weigh decisions and justify beliefs. As such, reasoning is not the dominant mode of cognition.  Much of our everyday mental activity is involved in intuitive or common sense inference, like “it’s raining, so I’ll need to take an umbrella.” This sort of immediate inference is contrasted to reasoning, in which the link between one thought and another is intermediated by at least one consciously held proposition. This proposition facilitates the mental transition from one belief, or a piece of data, to a conclusion. In many situations, we cannot justify to ourselves (or to others) the acceptance of a conclusion without the help of these intermediaries. So part of the job of the faculty of reason is to find tenable propositions, i.e., reasons, which will help us justify our beliefs and decisions. Reasoning is therefore associated with the creation, articulation, and evaluation of arguments.

The Function of Reasoning

The “Classical View” of reasoning’s function is that it helps the “lone reasoner” arrive at better beliefs and make better decisions, typically through self-scrutiny and the correction of mistaken intuitions. Descartes’ conception of reasoning as a means to acquire knowledge based on indubitable first principles represents this view, as does Daniel Kahneman’s view that the purpose of reasoning is to correct mistakes of intuition.

If this really were the function of reasoning, we would expect that human beings generally reason quite well. We are the “rational animal,” after all.  (To Mercier’s point, if the primary function of teeth is to chew food, we would expect that animals with teeth are generally successful at chewing and ingesting food for subsequent digestion. What sense would it make to assert that the primary evolved function of teeth is chewing if animals chewed badly, regularly choking on their food, leading to poor nutrient utilization or even asphyxiation?)

The Irrational Animal?

But how well do humans reason? By the standards implied by the classical view of reasoning, we do not reason very well. Cognitive psychologists have established empirically that for a wide range of tasks, we actually perform really badly — badly enough that the classical view of reasoning becomes difficult to sustain. The point is not that reasoning cannot correct mistakes of intuition. Surely it can. The point is that it does not typically do so.

In his talk, Mercier reviewed the well-known Wason selection task experiment, and the usual dismal results obtained. (I discussed the selection task in an earlier post). Although expert logicians get the correct answer to this problem two or three times more often than undergraduates, a 30% success rate cries out for an explanation. Mercier also gave the bat-and-ball problem as an example of an apparently simple challenge for which the wrong answer is often given. With this problem, we are given two facts: (i) that a bat and ball together cost $1.10, and (b) the bat costs $1.00 more than the ball. The question is “how much does the ball cost?” (See Kahneman’s Thinking Fast and Slow, p. 44 for the answer.)

The Wason selection task and the bat-and-ball problem are challenges to which people often respond inappropriately, that is, by intuition, rather than logic. If reason’s function were to correct mistaken intuitions, we would predict that reason would regularly detect and correct them. But the opposite usually happens. That is, reason looks for arguments that would confirm, and not falsify, an initial belief based on an intuition. In other words, our reasoning faculty has a built-in mechanism of confirmation bias!

Confirmation Bias

Confirmation bias is the tendency to only look for arguments that support our ideas and hypothesis. The classical view of reasoning suggests that our faculty of reason should be testing and vetting intuitive beliefs, not coming up with reasons in support of them. And even smart, open-minded people who are highly motivated to get the right answer (i.e., they get paid) have this bias.

What Mercier does at this point is intriguing. Rather than taking the standard view that confirmation bias is a “bug” in our cognitive apparatus, he says just the opposite. Confirmation bias is a symptom of reason performing its primary evolved function, that is, looking for reasons to justify beliefs and decisions. And from this perspective reason functions very well!  That is, if the rational animal evolved reasoning as a mechanism for gathering reasons to put together arguments in order to justify beliefs and decisions, we would predict that confirmation bias would be a common phenomenon.

Yet confirmation bias is just the beginning of the problem for the classical theory. We observe reasoning acting to bolster rather than challenge intuition, whether it is right or wrong, while exhibiting additional phenomena such as belief perseverance, polarization, and overconfidence.

If isolated ratiocination is a context in which we don’t adequately challenge ourselves, perhaps it is not the faculty of reason to blame, but the context in which reasoning occurs. The “argumentative theory of reasoning” predicts that in social contexts of reasoning, in particular when there is disagreement between interlocutors and a critical discussion ensues, we can reason well. What is the basis of this prediction?

Reasoning Is For Arguing

Mercier starts with the premise that for communication between two people, the exchange of messages must by-and-large be beneficial to both parties, i.e., the sender of a message and the receiver of the message. If it were not, then we would not observe communication to be a stable feature in the human population. The second premise is that the sender of a message can benefit from lying, cheating, and deceiving the other person. So the receiver must have a means to vet incoming information, and adopt the attitude of epistemic vigilance. The third premise is the observation that while mechanisms of epistemic vigilance such as trust calibration (adjusting trust depending on perceived benevolence and competence) and coherence checking (checking of new information against background beliefs) are effective, they tend to reject too much information and as a result, throw out some good information along with the bad. Therefore, there would need to be a solution to the problem of excessive skepticism on the part of the receiver, given that senders can lie, etc., and that communication is nonetheless stable.

The solution is that senders can provide reasons supporting the message such that the receivers can evaluate those reasons and then decide whether to accept or reject the message based on them. The argumentative theory of reasoning postulates that reasoning evolved to help people communicate by exchanging reasons, i.e., by dialog.

Now, if a sender has an interest in persuading an epistemically vigilant receiver to accept a given piece of information, we would predict that senders would try to find reasons to support their standpoint, i.e., reasons that confirm their side of an issue. So confirmation bias is really “my-side” bias, and we would expect confirmation bias of this sort if reasoning is use to find arguments for our own positions on issues. Yet, it would be a mistake to think of confirmation bias as a cognitive limitation. To turn people into falsifiers, we just have to give them something to disagree about!

It is also a mistake to think of confirmation bias as a side effect of people getting emotionally charged in a debate over some contentious topic. It turns out that confirmation bias is no more prevalent in discussions of politics than it is in discussions about something mundane like the correct solution to the Wason selection task.

The Social Context of Effective Reasoning

The argumentative theory predicts that in contexts of solitary ratiocination, people tend not to challenge their own reasons for beliefs and decisions. Using our “introspective eye” is neglected because people don’t really care if those reasons are good or bad; the production of reasons is intended merely to convince others, or provide self-reinforcing rationalizations for us. On the other hand, once dubious reasons are served up in a social context, others may demur. The argumentative theory predicts that confirmation bias can be defeated by cooperative or collaborative inquiry, and that reasoning should work better when people are placed in (social) situations where they are challenged to evaluate arguments.

For example, in one Wason selection task study, it was found that 18% of the participants got the right answers when working on the problem in isolation. Things changed dramatically when the participants were then put in a group of four or five, however. In this situation, one of the people who figured out the correct solution convinced others (who had come up with various invalid solutions) that he had the right answer. After discussing the various solutions — some wrong, one right — about 80% of the participants came around to accept the correct solution. However, in a group of people who came up only with various invalid solutions, group discussion did not appreciably alter the outcome.

A Consilience of Inductions

Mercier claims that the argumentative theory predicts and explains a number of phenomena, such as:

  • when and why reasoning fails to correct misguided intuitions
  • the prevalence of confirmation bias
  • the occurrence of motivated reasoning
  • why reasoning drives people to decisions they can justify, not necessarily better decisions
  • when and why groups outperform individuals on reasoning tasks

Because of the theory’s ability to explain diverse phenomena, and because of the convergence of empirical results from different areas of psychology (social, developmental, moral, cross-cultural, etc.), Mercier claims substantial support for the theory.

Whether or not you accept the Sperber-Mercier theory or the evolutionary psychology upon which part of its case depends, I think you have to admit that it is thought-provoking and raises some serious challenges to common assumptions about reasoning and argumentation.