Comments and Cautions About Generative AI

Generative artificial intelligence (GAI) already has disparate impacts on different communities, and are likely to further exacerbate existing disparities and biases. Recent research demonstrates that ChatGPT and other GAI tools are most capable when responding to prompts in English, meaning they may “exacerbate bias for English and English speakers.” Namely, GAI systems struggle when translating English into other languages, and particularly languages that do not use the Latin alphabet; an especially alarming related research finding from academics at the University of Oregon note that ChatGPT is “more likely to fabricate information” in non-English languages. For the 31% of Asian Americans who have limited English proficiency (LEP), many of whom primarily speak character-based languages like Chinese or Korean, GAI in its current state is likely not a reliable or useful tool.