Munger’s study comes amid a growing appreciation of Twitter’s serious problems with harassment. Earlier this year, comedian Leslie Jones, a star of the recent Ghostbusters remake, was inundated with horrifying, racist tweets. She was the year’s most prominent victim of mass abuse, but far from its only one. As Charlie Warzel wrote on Buzzfeed, “Today, Twitter is a well-known hunting ground for women and people of color, who are targeted by neo-Nazis, racists, misogynists, and trolls, often just for showing up.” The problem has been acknowledged by Twitter’s CEO, and the company has today launched new tools designed to address it, including the ability to mute certain conversations and to filter our chosen words or phrases.
Munger tried to do so by creating several bots. He gave them all the same profile information and male cartoon avatar, but he varied their skin color and names to make them identifiably white or black. He also gave them followers—either fewer than 10, or between 500 and 550, which he “bought from a sketchy website.” And he wrote fake tweets from their accounts so that no one would suspect that they weren’t real.
Next, he compiled a list of white men on Twitter who tweeted the n-word at other users, regularly and offensively. He then targeted each of these people with one of the various bots, who admonished them for their slurs. And he carefully chose words that were not aggressive, but would emphasize common humanity: “Hey man, just remember that there are real people who are hurt when you harass them with that kind of language.”
In the following months, Munger found that people reduced their use of racist language if they were sanctioned by the white bot with lots of followers—and only that bot. This wasn’t just a drive-by effect, either: It dwindled over time, but lasted for at least a month.
“It tracks with previous research,” says filmmaker and futurist David Dylan Thomas. For example, in 2014, 14-year-old Trisha Prabhu created an anti-cyberbullying app called Rethink which detects when people are writing hurtful comments and asks them if they’re sure they want to post it; most pull back. As Thomas says, “If you can make someone aware of the fact that what they’re doing has an impact, to disrupt the process of going ‘I have this emotion, I’m just going to post it,’ in some circumstances, it can have an effect on the frequency of posting hateful commentary.”
The approach is clearly scalable. You could imagine an army of bots, crawling through Twitter and Facebook and speaking out against hate speech.
“It seems weird to advocate for the use of bots to corral behavior, but it doesn’t have to be bots,” says Damien Williams, a philosopher at Kennesaw State University who has studied futurism and AI. “Someone like William Gibson with hundreds and thousands of followers taking the time to say, ‘Hey this isn’t acceptable, you’re hurting real people,’ that would have a major impact on a lot of people who see him as an important powerful figure in their group.”