As AI tools get smarter, they’re growing more covertly racist, experts find

AI models were more likely to assign AAVE speakers to lower-paying jobs compared with speakers of ‘standard American English’. Photograph: Bloomberg/Getty Images

Popular artificial intelligence tools are becoming more covertly racist as they advance, says an alarming new report.

A team of technology and linguistics researchers revealed this week that large language models like OpenAI’s ChatGPT and Google’s Gemini hold racist stereotypes about speakers of African American Vernacular English, or AAVE, an English dialect created and spoken by Black Americans.