by Alexander Gelfand

McArthur designs online behavior systems for Riot Games, the company behind League of Legends (LoL), one of the most popular video games on the planet. LoL boasted more than 80 million monthly active users in 2017 and is a pillar of e-sports, with professional gamers competing in North American, European, and Asian leagues. But some of its players can get a little salty—or worse. So McArthur and the rest of Riot’s Player Behavior Team use a combination of data science and social science to maintain order in an unruly virtual world.

LoL belongs to a class of online games that pit teams of players against each other in pitched battle. While inspiring intense loyalty, such games also breed toxicity, with the most egregious offenders spewing racist, sexist, and homophobic slurs. And LoL, a fantasy game in which two five-player teams of mages, marksmen, and other champions struggle to destroy one another’s castle-like base, is no exception, with aggrieved gamers complaining of harassment by opponents and teammates alike. It’s the kind of thing that can tarnish a game’s reputation and drive away users. (Riot has already banned several professional LoL players for in-game abuse.)

Some of this negativity, McArthur says, is fostered by the absence of normal curbs on nastiness. And some is due to the nature of the game itself.

Their work makes for interesting bedfellows. McArthur, who analyzed neuroscientific data using statistical methods while studying computation and neural systems at Caltech, worked for several years with a psychologist specializing in deviant behavior, and she often collaborates with academic researchers to reap behavioral insights whose relevance extends well beyond LoL.

McArthur recently co-authored a paper with a group of researchers at MIT that used game data to investigate “collective intelligence,” or the ability of a team to perform a wide variety of tasks. Among their findings: Lasting teams perform better than temporary ones (see above re: playing nicely). So do teams with female members, who tend to be more socially perceptive than men. And the highly virtual, fast-paced nature of the game means that tacit coordination among teammates matters more than explicit communication.

These takeaways could help improve the performance of virtual teams operating in business, or even groups operating in real-world environments that share the game’s intensity and reliance on rapid decision-making—such as emergency response teams and military combat units.

McArthur herself has come to appreciate the value of such research to everyday life. “It fundamentally helps you understand the interactions you have in your day-to-day environment with your peers and family, especially when it comes to cooperative behavior,” she says.

But she’s primarily focused on applying those lessons at Riot, where she has developed machine-learning models to analyze the reports that players submit on disruptive gamers and to match that data with particular kinds of in-game activity, creating—and enforcing—a set of data-driven community standards for acceptable behavior.

The system can identify an infraction and issue feedback to offending players minutes after a game has ended, explaining precisely how they misbehaved and handing out penalties ranging from temporary suspensions to permanent bans. A new name-detection feature can even identify vulgar or offensive player names and force users to change them—a costly process that requires spending either real money or hard-earned game currency.