People did not evolve to argue about tax policy with strangers on the internet. For most of human history, survival depended on a small band, and being cast out could be fatal. The mind that came out of that history is quick to sort people into "us" and "them," and slow to notice that it is doing so. Knowing how this sorting works will not switch it off, but it lets you catch it in the act.

Groups out of nothing

In experiments published in 1970 and 1971, Henri Tajfel and his colleagues at the University of Bristol divided schoolboys into groups on the flimsiest grounds imaginable: whether they overestimated or underestimated the number of dots on a screen, or whether they preferred paintings by Paul Klee or Wassily Kandinsky. The boys then divided small rewards between other boys known only by code numbers, gaining nothing themselves. Even so, they favored members of their own trivially defined group. More striking, many chose to maximize the gap between the groups even when it cost their own group money: they preferred a split that gave their own group fewer points if it put them further ahead of the other group. The lesson of these minimal group experiments is sobering. A label alone, with no history, no conflict and no real stakes, is enough to start favoritism.

In-group love and out-group hate

It would be easy to conclude that humans naturally hate outsiders. The psychologist Marilynn Brewer argued in 1999 that the evidence points somewhere more hopeful. Most discrimination, she found in reviewing the research, comes from warmth toward one's own group rather than hostility toward others. People help, trust and forgive insiders more, without necessarily wishing outsiders harm. Active hostility tends to grow under particular conditions: when groups compete for scarce goods, when one side feels threatened, or when a group comes to see itself as morally superior.

"They are all alike"

A second tendency is the out-group homogeneity effect: people see members of their own group as varied individuals but see members of other groups as more alike than they are. Partisans can name the factions on their own side while picturing the other side as a single type. Once a whole group looks uniform, its most extreme member can stand in for all of them.

When reasoning protects identity

Group loyalty also shapes how people reason. The legal scholar Dan Kahan and colleagues gave Americans a puzzle about interpreting data from a table. When the table was said to show whether a skin cream worked, people who were good with numbers solved it more often, as expected. When the identical numbers were said to show whether a ban on carrying concealed handguns reduced crime, something odd happened: the most numerate people became more polarized, using their skill to reach the answer their political side preferred. Kahan calls this identity-protective cognition: reasoning that protects a person's standing in a group they belong to. For an individual, the cost of being wrong about gun statistics is small. The cost of disagreeing with everyone you love can be large. The mind does the math.

A useful question follows. Before arguing about whether a claim is true, ask yourself: what would it cost me socially to believe the other answer? If the honest answer is "a lot," your reasoning deserves extra scrutiny, and so does the other person's.

Moral conviction

Finally, some opinions are held as matters of taste or practicality, while others are held as moral truths. Linda Skitka and colleagues found that the more people experienced an attitude as a moral conviction, the more they wanted social distance from people who disagreed: they were less willing to have them as neighbors, coworkers or friends. Moral conviction is not a flaw. Abolitionists held their views as moral convictions, and rightly so. But it means that the issues people care about most are the ones on which they are least able to tolerate the people across the line.

Group identity is not bad in itself: it gives people belonging, and every movement for human rights has depended on solidarity. The danger lies in letting a group label do the work that judgment should do.

Worked example: the team chat

Priya works at a company where a heated policy debate has split the staff. In the team chat, a colleague from the other camp posts a statistic. Priya's first reaction is that "those people" always twist numbers. Then she notices three things. She has never actually talked with this colleague about the issue, so her picture of him comes from the loudest voices in his camp, a sign of out-group homogeneity. She also realizes that if the same number had come from her own side, she would have shared it without checking. And when she asks what it would cost her to accept it, she admits that her closest friends at work would be annoyed.

None of this tells Priya whether the statistic is right. What it tells her is that her instant judgment was driven by membership, not evidence. She checks the source, finds it is sound but narrower than the post implied, and replies to the colleague directly, as one person to another, instead of mocking the post in a side channel.