UC San Diego researchers teach a 7B AI agent to read its partner’s mind

Researchers at the University of California, San Diego have trained a relatively small AI agent to guess what its conversation partner believes and wants. The agent can then act on those guesses without help from any outside model. The work arrives in a paper titled “Mental-Models for Multi-Agent Systems” (arXiv:2610.12453), posted on October 8, 2026. It has been accepted at NeurIPS 2026. The authors are Hanan Gani, Lulu Shao, and Manmohan Chandraker, all affiliated with UC San Diego. What the agent actually learns The core idea borrows from psychology. “Theory of mind” is the ability to recognize that other people hold beliefs, goals, and knowledge that differ from your own. The paper builds in both layers. First-order theory of mind covers what a partner believes or wants. Second-order theory of mind covers what that partner believes about the agent itself. The researchers call their approach “mental-model-enabled agents.” The agent maintains an explicit internal representation of a partner’s hidden mental state, stored as what the authors describe as an amortized recursive latent state. Unpacked, “amortized” means the expensive reasoning work is folded into training. The agent l...