Confirmation bias, or why your reading always proves you right

The work of the psychologist Peter Wason, in the 1960s, showed that faced with a rule to verify, participants almost always look for cases liable to confirm it, rather than cases liable to refute it. The experiment used series of numbers and cards to be turned over, with nothing political, identity-related or financial at stake.

That is what makes the result important. If the bias shows itself where nobody has an interest in being right, then it needs no passions to be triggered. Passions do not create it, they amplify its effects and make correcting it more costly.

A point of vocabulary is called for. Confirmation bias is neither lying nor intellectual laziness. A very well-informed, very rigorous and perfectly honest person is subject to it, and all the more so the better they master the arguments of their own camp.

The first bears on the search for information. We read, listen to and keep company with what goes our way, often without deliberate intention: through the choice of sources, through the recommendations that follow from it, through the composition of those around us. The sample of arguments we are exposed to is therefore not representative of what exists.

The second bears on evaluation. At equal quality, an argument that contradicts us is examined more demandingly than an argument that comforts us: one looks for a methodological flaw in it, questions the source, notes what it does not demonstrate. The same level of demand, applied to the favourable argument, would often produce the same verdict of fragility.

The asymmetry is what makes the phenomenon invisible from the inside. We all remember having taken apart a badly built opposing argument, which feeds the sense of being demanding. We far less often remember having accepted without checking a piece of reasoning that suited us, precisely because we did not check it.

Environments where content is ranked according to the engagement it provokes make selective exposure easier: they make it more likely to encounter what pleases, and less likely to encounter what disturbs. The tendency existed before them, in the choice of a newspaper or a circle of friends; the scale and the speed are new.

One must nevertheless guard against too convenient an explanation. Research on the actual diversity of the sources consulted gives results more nuanced than the image of a sealed bubble, and some work observes contradictory exposure more frequently online than offline. The problem is not only encountering the opposing argument, it is treating it with the same demandingness as one's own.

In other words, the architecture of platforms acts above all on the first effect, that of searching. The second, that of evaluation, remains entire, and it is the one that decides what we do with contrary information when it finally arrives.

Telling oneself “I am going to stay objective” has no measurable effect. Intention is not enough, and confidence in one's own impartiality even tends to worsen the bias, since it dispenses one from checking. The useful remedies are procedural, not moral.

Three procedures have proved more robust than good will. Writing the best version of the opposing position, the one its most competent supporter would defend, obliges one really to encounter it. Explicitly looking for what could make you change your mind shifts attention from the conclusion to the evidence. Deciding in advance on the criterion that would make you acknowledge an error turns a position into a hypothesis.

This last procedure has a particular merit, including when it fails. If no imaginable fact could move you, it is useful to know it: it means that what you are defending belongs to a value, which is perfectly legitimate, but is not discussed like a piece of data. Many conversations would gain from beginning with that clarification.

After answering an opinion questionnaire, go back over the two or three items on which your answer was the most immediate. Those are the ones where your position is most firmly installed, and therefore the ones where the bias has the most opportunities to act unnoticed.

For each, write out the strongest objection you know of, and note what would have to be observed for it to prevail. The exercise does not change minds, and that is not its purpose: it brings out the difference between the positions you hold as findings and those you hold as principles.