The artificial intelligence was created by researchers at the University of Southern California (USC). They taught the AI to read Twitter posts and pick up on language that indicated hostility. The AI then cross-referenced these posts with real-world protests to predict violent altercations, reports Digital Trends:
[The researchers] turned their gaze toward some 18 million tweets about the 2015 Baltimore protests against police brutality, following the death of Freddie Gray. The system scanned arrest rates, a statistic that’s often used as a proxy for violence, and found that arrests increased as “moralized” tweets increased, nearly doubling on days of violent clashes between police and protestors.
As Morteza Dehghani, one of the researchers in the USC study noted:
“By tracking moralized tweets posted during the 2015 Baltimore protests, we were able to observe that not only did their volume increase on days with violent protests, but also that their volume predicted hourly arrest rates, which we used as a proxy for violence, during the protests. To further unpack these effects, we conducted a series of controlled behavioral experiments and we consistently observed the same effect of moral convergence.”
It is hoped that USC’s research could be used to create a tool that could be used to predict when protests are about to turn violent. If it can do that, it’s possible the tool could then be used to send out messages on social media for people on both sides to calm down and take a step back.