Large language models (LLMs) – artificial intelligence (AI) systems designed to understand, process, and produce human-like text – generate outputs about people. These outputs are stories that can reinforce representational harms and stereotypes for diverse youth.

Faye Marie Vassel, a post-doctoral fellow at Stanford University’s Institute for Human-Centered AI, describes research examining bias in large language models used in educational contexts to generate stories about students. Vassel highlights the value of an intersectional lens in understanding how these AI depictions can impact young people.

Vassel describes three key forms of potential representational harm in AI outputs – erasure, subordination, and stereotypes – with implications for education.