The Imagined Congress: What Generative AI Models Think U.S. Politicians Look Like

• In Event: LLMs and Generative AI

Brief Overview

We audit image generators’ portrayals of U.S. legislators. Using attribute-masked prompts and a large corpus of synthetic headshots, we quantify age, gender, and race deviations versus ground truth and discuss implications for representation.

Authors

• Ezgi Siir Kibris, Rochester Institute of Technology

• Cantay Caliskan, University of Rochester

• Da Gong, SUNY College at Geneseo

• Andong Yan, University of Hong Kong