Nature review maps four LLM roles in social science

Large language models are being pressed into "four distinct research roles" across computational social science, according to a review published 26 October 2026 in Nature Human Behaviour by Fengli Xu, Chenyang Shao, James Evans and colleagues: "subjects that simulate human behaviour, assistants that automate analysis, collaborators that generate hypotheses and evaluators that audit and improve research quality."

The authors survey work across sociology, political science, psychology, anthropology and economics. In the subjects role, they write, models "reproduce survey distributions, economic games and polarization dynamics," and the documented applications include social-network simulations running at millions of agents, political polarization forecasting and market simulation.

The review is blunt about the limits. It calls validation "the central challenge for generative social simulation," and reports that the reviewed models display persistent anti-Muslim bias and gender-related opinion exaggeration when they stand in for diverse populations. The authors also flag an uneven ability to "reproduce social complexity" and ongoing struggles with causal inference, alongside privacy and misinformation concerns they treat as part of the field's "fundamental challenges."

Their framing is recursive: "As artificial intelligence reshapes human society, understanding this transformation requires the very tools driving it." In the authors' telling, LLMs are instrument and subject at the same time.

Two of the researchers we track posted the link the same week it appeared.