Xu, Evans map four LLM roles for social science research

Large language models now play four distinct roles in computational social science research: subjects, assistants, collaborators and evaluators. A new Nature Human Behaviour review, published 26 October 2026 and led by Fengli Xu with the University of Chicago's James Evans as senior author, defines them as "subjects that simulate human behaviour, assistants that automate analysis, collaborators that generate hypotheses and evaluators that audit and improve research quality."

The review spans economics, psychology, political science, sociology and anthropology. When prompted with demographic or ideological profiles, the authors write, models "reproduce survey distributions, economic games and polarization dynamics." Applications documented include social network simulations with millions of agents, political polarization forecasting, and market simulation.

The caveats are blunt. LLMs "still struggle with causal inference," and their ability to reproduce social complexity "remains uneven and unproven." The authors call validation "the central challenge for generative social simulation," and point to persistent anti-Muslim bias in reviewed models, noting separately that "LLMs can exaggerate gender differences in opinion related to harms of misinformation."

The closing framing is reflexive: as AI saturates society, the same models being used to study humans become part of what is being studied. Two researchers on our tracker circulated the paper the day it posted.