Every large atrocity in modern history was preceded by words. In the months before the 1994 genocide in Rwanda, a popular radio station called Tutsi people inyenzi, "cockroaches." Nazi propaganda showed Jews as rats spreading through Europe. Words like these do not cause killing on their own, but they remove an obstacle to it: the ordinary human reluctance to harm someone who is recognizably like yourself. Psychologists call this process dehumanization, and they have learned that it comes in more forms, and in milder versions, than most people assume.

Two ways of denying humanity

The Australian psychologist Nick Haslam proposed in 2006 that people can deny others' humanity in two different ways, because there are two different things we mean by "human." One is what separates humans from animals: refinement, self-control, reason and moral sense. Denying these produces animalistic dehumanization, in which others are seen as crude, impulsive and animal-like. This is the dehumanization of the "cockroach" broadcasts, and of slurs that compare people to apes, dogs or vermin.

The other thing we mean by "human" is what separates people from objects and machines: warmth, emotion, curiosity and individuality. Denying these produces mechanistic dehumanization, in which others are treated as cold, interchangeable objects. This is the dehumanization of the bureaucracy that sees a refugee as a case number, of the doctor who talks over a patient as if they were a broken machine, or of the online commenter who calls opponents "bots" and "NPCs." It is quieter, but it also makes harm easier.

The everyday version

Dehumanization is not only something extremists do. A Belgian team led by Jacques-Philippe Leyens found in research beginning around 2000 that people readily grant their own group complex, uniquely human feelings such as remorse, nostalgia, hope and admiration, while being slower to see the same feelings in outsiders. Outsiders are allowed simple emotions such as anger and fear, which animals share. Leyens called this infrahumanization: seeing other groups as capable of basic emotions but less capable of uniquely human ones.

Measuring the blatant kind

For years researchers assumed that open, blatant dehumanization had largely disappeared from modern democracies. In 2015 Nour Kteily and colleagues tested that assumption with a strikingly direct tool: a slider beneath the familiar "ascent of man" drawing that shows a stooped ape gradually becoming an upright human. Participants placed various groups on the slider, from 0 to 100. Many Americans openly placed some groups lower on the scale than their own, and the lower they placed a group, the more likely they were to support harsh treatment of that group, such as torture of suspects or barring them from entry. Blatant dehumanization, it turned out, was alive and predicted real policy views.

How good people do harm

Albert Bandura, one of the most cited psychologists of the twentieth century, asked a related question: how do people who hold ordinary moral standards bring themselves to hurt others? His answer, set out in 1999, was moral disengagement, a set of mental moves that switch off a person's conscience without changing their values. Dehumanization was one. Others included euphemistic labeling, which softens harmful acts with gentle words, so that bombing civilians becomes "collateral damage" and firing workers becomes "rightsizing." Another was displacement of responsibility, in which people see their harmful acts as dictated by an authority: "I was only following orders." Others spread blame across a crowd, minimize the consequences, or blame the victims for what happened to them.

Bandura's list works as a checklist: when you hear your own side using these moves, a conscience is being switched off.

Warning signs in your own speech

For the personal practice of disagreement, the most useful lesson is to watch for language that describes people with words for vermin, disease or filth. Calling a political movement a "cancer," its members "parasites," or a group of migrants an "infestation" borrows directly from the vocabulary that has preceded mass violence. People who use these words are rarely planning violence. But they are rehearsing a way of seeing that makes violence easier for someone else. The same goes for mechanistic language: describing opponents as "programmed," "brainwashed" or "drones" denies that they have minds at all, which is a convenient way to avoid engaging with their reasons.

Worked example: the comment thread

A local news site posts a story about a new shelter for asylum seekers. Within an hour, the comments include: "These people breed like rats." "Anyone who opposes the shelter is a Nazi." "The council just processes these cases like cattle." "Supporters are bots running a script."

Apply the level's tools. The first comment is animalistic dehumanization of asylum seekers. The second treats opponents as moral monsters, which shuts down any chance of hearing their concerns about cost or location. The third is a complaint about mechanistic dehumanization by the council, and may be a fair one. The fourth mechanistically dehumanizes supporters by denying they have minds. A citizen who wants to oppose or defend the shelter can do so in full, but can do it without any of these moves, and can point out, calmly, when others use them.