Cansu Malak¹, Christian Wallraven¹; ¹Korea University

When we perceive a face, we rapidly judge both ethnicity and social traits like attractiveness and trustworthiness. But how do these two processes interact - especially for visually similar groups? We investigated whether judgments of attractiveness and trustworthiness enhance or distort the categorization of East Asian (Korean, Japanese, Chinese) faces by Korean observers. For this, we presented 600 full-face images of Korean, Japanese, and Chinese males (200ms exposure) to 104 Korean participants. The participants were assigned to two tasks: one group (N=52) rated each face for attractiveness and trustworthiness, while a second group (N=52) categorized the ethnicity of the same faces. We then correlated the social ratings from the first group with the categorization accuracy from the second. Results revealed a strong in-group bias: using the dataset labels, Korean faces were rated as significantly more attractive and trustworthy than Japanese or Chinese faces: for in-group Korean faces, higher attractiveness ratings (r =.62) and trustworthiness ratings (r =.65) strongly predicted greater categorization accuracy. For out-group Chinese faces, the opposite pattern emerged: higher ratings predicted lower accuracy (attractiveness r = -.44; trustworthiness r =-.35), revealing a misattribution bias where socially positive faces were incorrectly classified as Korean. No significant link was found for Japanese faces. Across all stimuli, more attractive or trustworthy faces were also more likely to be labeled as Korean, reflected in positive correlations with “Korean” responses (attractiveness r =.61; trustworthiness r =.65) and negative correlations with “Japanese” (r =–.23; –.37) and “Chinese” classifications (r =–.45; –.32). Overall, rapid social evaluations are not neutral; they are deeply intertwined with ethnic categorization. Our results are compatible with a strong in-group projection bias, where socially positive faces are claimed as "one of us," demonstrating that face categorization is a social-affective process, not just a visual one.

Acknowledgements: Supported by the National Research Foundation of Korea (BK21 FOUR, RS-2025-00555141), by IITP grants funded by the Korea government (No. RS-2019-II190079, Department of Artificial Intelligence, Korea University; No. RS-2021-II212068, Artificial Intelligence Innovation Hub).