Although tools that spot stereotypes in AI models already exist, the vast majority of them work only on models trained in English. To get around these problematic generalizations, SHADES was built using 16 languages from 37 geopolitical regions. The researchers exposed the models to each stereotype within the data set, including through automated prompts, which generated a bias score. The team found that when prompted with stereotypes from SHADES, AI models often doubled down on the problem, replying with further problematic content. “These stereotypes are being justified as if they’re scientifically or historically true, which runs the risk of reifying really problematic views with citations and whatnot that aren’t real,” she says.