Skip to content
Technology & Innovation

AI can now predict internet comment fights

A team from Cornell University looks to shut down the ongoing flamewar that has taken over, well, pretty much every public area of the internet.
Mildly famous WoW internet troll Leroy Jenkins, c/o South Park (Comedy Central)

If you’ve been on the internet for more than 25 minutes, you’ve most likely run into someone saying something you disagree with. As most of us realize, this is part of adult life. The world was not built explicitly for us. Most (read: good) people might move on,  but many — about 30% of Americans — take the internet comment section as their own ideological battleground. 

A team at Cornell University has created an AI that could in the future root out online conversations going awry and nips them in the bud. After studying 1,200 Wikipedia Talk pages, they came upon a few surprising conclusions.

The most useful? Many negative internet aggressions often start when one person accuses another… which usually starts with the word “you.” Also, direct questions seem to throw people through a loop because they’re framed like the truth. You can read the whole paper here

Humans still proved better at rooting out trolls, by a 72% margin over the AI’s 61.6%. But as we’ve reported very recently, AI has a startling capacity to learn. Hopefully, in just a few years this could salvage social media from becoming a flame-war. Companies like Twitter, Facebook, and YouTube (just to name a few) seem to be caught in a horrible ideological loop: fight “free speech” by trying to weed out trolls, or allow them to run rampant and ruin any and all conversations. Editorializing here, haven’t we all reached a point where can agree that nobody will ever “win” an internet fight? 


Up Next