How AI Sees Us: Psychological and Societal Implications of AI, Nuria Oliver

Edward Thorndike identified the “halo effect” in 1920: A single positive trait, like physical attractiveness, biases broader judgments. A century of research confirms that attractive individuals are perceived as more intelligent, trustworthy, and capable. Despite this awareness, the bias persists, and it is now embedded in AI systems at scale. Recent work shows that AI beauty filters amplify, rather than reduce, attractiveness bias by aligning faces with dominant beauty norms. Multimodal AI models also exhibit the halo effect, rating more attractive faces as more competent and sociable. Generative AI goes further, producing more attractive faces for positive traits and less attractive ones for negative traits, embedding bias in content creation itself. These distortions propagate into downstream systems, worsening performance for underrepresented groups. The issue is systemic: AI inherits human biases and amplifies them. Psychology’s extensive bias research remains largely unused in AI governance, raising urgent questions about accountability.