Nature review maps four research roles for LLMs in social science
TL;DR
- A Nature Human Behaviour review groups LLM use in computational social science into four roles: subjects, assistants, collaborators and evaluators.
- Spanning five disciplines, the authors say models reproduce survey distributions, economic games and polarization dynamics when prompted with profiles.
- They flag causal inference, validation, anti-Muslim bias and exaggerated gender differences in misinformation opinion as unresolved problems.
Large language models now serve four research roles in computational social science: subjects that simulate human behaviour, assistants that automate analysis, collaborators that generate hypotheses, and evaluators that audit and improve research quality. That is the framework laid out in a review published in Nature Human Behaviour on 26 October 2026, led by Fengli Xu with James Evans as senior author.
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," with applications including social network simulations involving millions of agents, political polarization forecasting, and market simulation.
The authors are more cautious about what the tools still cannot do. LLMs "still struggle with causal inference," and their ability to reproduce social complexity "remains uneven and unproven." Validation, they argue, is "the central challenge for generative social simulation." The paper also flags persistent anti-Muslim bias in the models surveyed, and notes that "LLMs can exaggerate gender differences in opinion related to harms of misinformation."
Two researchers we track circulated the link the same week the review went up.
Originally reported by nature.com
Read the original article →Original headline: Large language models for computational social science - Nature Human Behaviour