Artificial intelligence is reshaping how we ask—and answer—questions about human behavior and society. Our new AI Methods Series showcases powerful AI approaches that open new possibilities for social and behavioral science research. Each session will feature leading experts who share how they have applied these methods in their research. We hope the sessions will spark ideas for your own projects.
This session features Stefan Uddenberg, Assistant Professor in the Department of Psychology, whose research explores the intersection of social and visual cognition, combining simple behavioral methods with cutting-edge generative AI techniques.
Registration is required. Questions may be directed to csbscience@illinois.edu. This event is in person only and will not be recorded.
Wednesday, November 12 | 12:00 – 1:00 pm
NCSA Room 1030 | 1205 W. Clark St., Urbana
Hyper-realistic reverse correlation reveals gender bias in leadership representations
Abstract: Appearance influences election outcomes via leadership stereotypes. We used hyper-realistic reverse correlation to elucidate such visual stereotypes. Participants (N=300) viewed generated faces and judged their leadership quality. Aggregating these choices yielded visually compelling and interpretable mental representations at both individual and group levels. While political group-averaged representations were similar along many subjective attributes (e.g., “trustworthiness”), they revealed a gender bias: conservatives’ “good leaders” were more masculine than those of liberals. We directly replicated this result using richer latent face representations (N=300). We validated individual participant models on other observers (N=150), probing their willingness to vote for different faces. Participants were not only more willing to vote for “good” leader faces, but were most willing for faces generated by participants sharing their political orientation. Our results demonstrate how political orientation is linked to a novel gender bias in leadership representations, showcasing the utility of our reverse correlation technique.