AI and the World's Hard Problems · Economics & social sciences · arXiv · 2026-10-19
Emergent Discrimination and Polarization in Multi-Agent Systems
This paper theoretically studies how discrimination and polarization can emerge among learning agents. Each agent maps a language-model embedding of an issue to a binary opinion and updates its trust in other agents through an online Bayesian rule. In-context-learning experiments with large language models qualitatively reproduce the predicted updates, and simulations show four collective states depending on the strength and prevalence of group bias.