Nature review casts LLMs in four roles across social science

A new review in Nature Human Behaviour argues that large language models have taken up four different seats at the social-science table at once.

Fengli Xu, James Evans and co-authors frame it this way: LLMs now serve as 'subjects that simulate human behaviour, assistants that automate analysis, collaborators that generate hypotheses and evaluators that audit and improve research quality.'

The review covers sociology, political science, psychology, anthropology, economics and related fields. The authors are explicit that the picture is not uniformly rosy. 'LLMs struggle with causal inference and raise urgent ethical concerns about privacy and misinformation,' they write, and their ability to reproduce social complexity 'remains uneven and unproven.'

Running through the paper is a point the authors keep returning to: the object of study and the instrument have begun to overlap. The review describes LLMs as 'both our research tools and authentic research subjects.' Or, as the authors put it: 'As artificial intelligence reshapes human society, understanding this transformation requires the very tools driving it.'

The piece is dated 26 October 2026 and tagged as a Review Article. Two researchers on our Who's Who radar had already pushed the link into their networks within days.