Xu and Evans map four LLM roles in social-science research

A Nature Human Behaviour review published 26 October argues that large language models now occupy four distinct jobs 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. The piece, led by Fengli Xu and James Evans with colleagues at Tsinghua, HKUST and Microsoft Research Asia, surveys applications across sociology, political science, psychology, anthropology and economics.

In the 'subjects' role, the authors catalogue models that 'reproduce survey distributions, economic games and polarization dynamics,' including social-network simulations running at millions of agents. The caveat lands inline: the review calls validation 'the central challenge for generative social simulation,' and reports that the surveyed models display persistent anti-Muslim bias and gender-related opinion exaggeration when they stand in for diverse populations.

The authors are blunt about where this breaks. LLMs 'still struggle with causal inference,' they write, and their ability to reproduce social complexity 'remains uneven and unproven.'

Their closing framing is recursive. Understanding AI's transformation of society, the review says, requires using the very tools driving it, 'a recursive challenge that will define the future of social science.' Two researchers we follow in our Who's Who tracker had posted the link within hours of publication.