How AI amplifies, mitigates, and creates social biases: A tripartite framework

Artificial intelligence (AI) is increasingly embedded in social life, shaping how individuals perceive themselves and others and influencing social dynamics that bear directly on equality and collective welfare. Yet its relationship to social bias is paradoxical: the very systems that can entrench and amplify long-standing prejudices may also be leveraged to challenge them, and may simultaneously generate entirely novel forms of bias. This review synthesizes recent empirical literature within a tripartite framework organized around the core pathways along which AI shapes social biases: AI design, lay beliefs about how AI operates, and processes of social evaluation and attribution. Drawing on these pathways, I outline three corresponding avenues for intervention: more equitable AI design, targeted adjustment of lay beliefs about AI capabilities, and redesign of social evaluation processes to account for the attribution biases that AI introduces. Together, these findings underscore the urgency of a socially informed approach to AI development and governance.