What does an internet troll look like? Just like you

Everybody knows what an internet troll is: a maladjusted keyboard warrior turning his frustration about the sad little life he has wasted into venomous spurts of Twitter abuse.

This picture, however, is wrong. An internet troll, according to American research, probably looks just like you.

In a 21st-century echo of the 1971 Stanford prison experiment, in which psychologists found that almost anybody could quickly become a vicious authoritarian if given the opportunity, scientists have discovered that pretty much the same holds true for online opprobrium today.

Far from being the preserve of straggly-haired conspiracy theorists operating under noms de guerre such as L1bLabConL1es31, trolling turns out to be a temptation that a solid majority of the population will succumb to given only mild provocation.

While reliable statistics on the phenomenon are difficult to pin down, about two fifths of internet users claim to have been harassed on the web, and one comment in five of the comments on CNN’s news site is taken down by the network’s moderators.

Researchers at Stanford University in California and Cornell University in New York designed a series of tests to work out what it took to turn a mild-mannered, run-of-the-mill person into a troll.

They recruited 667 Americans with an average age of 34 through an online platform called Amazon Mechanical Turk. The group had a distinct political bias to the left, with 54 per cent of participants declaring themselves to be Democrats compared with 21 per cent who said they were Republicans and 25 per cent who claimed to be “moderates”.

Half of these people had to answer 15 taxing questions, such as rearranging the letters “DEANYON” to make a meaningful word, within a tight time limit to put them in a foul mood. The other half were given easier questions.

The subjects were then asked to read an online op-ed article arguing that women should vote for Hillary Clinton rather than Bernie Sanders in the Democratic presidential primaries, before posting at least one comment under a pseudonym such as User9054.

Some participants saw vituperative remarks at the top of the thread, such as: “Oh yes. By all means, vote for a Wall Street sellout — a lying, abuse-enabling, soon-to-be felon as our next president. And do it for your daughter. You’re quite the role model.”

Others saw gentler comments, including: “I’m a woman, and I don’t think you should vote for a woman just because she is a woman. Vote for her because you believe she deserves it.”

The responses each participant wrote underneath were then judged by two researchers and scored for negativity with a tool known as LIWC, which is short for linguistic inquiry and word count.

The scientists found that about a third of the participants were dyed-in-the-wool “career” trolls, writing rude comments even though they were in a sunny mood and the previous contributions to the thread had been fairly clement.

When people were upset after a tough test, however, and saw that other commenters had already begun to abuse the author of the article, this proportion rose to 68 per cent. Further analysis of more than 16 million posts on CNN.com appeared to confirm that the most important factor in trolling was the tone of the comments above.

Cristian Danescu-Niculescu-Mizil, assistant professor of computer science at Cornell and one of the authors of the paper, compared the findings to the “broken window” effect in sociology, in which one person violating social norms can make others feel more comfortable about following suit.

“Is it really true what we tend to think when we think about trolls, that they are sociopaths who just want to harm?” he asked. “We found that there’s actually a troll in each of us.

“If you think about online discourse, it’s so vitriolic, and it’s easy for us to blame the others, but maybe it’s not just the others — it’s us, too. Everybody has their role to play and everybody contributes to the discourse.”