Large language models take on four new roles in social science research

Large language models are reshaping how social scientists study human behaviour, according to a review published in Nature Human Behaviour. The article, led by Fengli Xu, Chenyang Shao and colleagues including James Evans, examines how these AI systems now serve four distinct research roles in computational social science: as simulated human subjects, as assistants that automate analysis, as collaborators that generate hypotheses, and as evaluators that audit and improve the quality of research itself. The review surveys recent advances across sociology, political science, psychology, anthropology, economics and related fields.

As collaborators, they propose hypotheses that human researchers can refine and test, and as evaluators they can review methods and flag weaknesses in study design. The authors argue that this transformation arrives at a notable moment: society itself is becoming saturated with artificial intelligence agents. Understanding that change, the review notes, may require using the very systems driving it—a recursive challenge the authors say will define the future of the field.

The assessment is not uncritical. The review finds that LLMs struggle with causal inference, a core requirement for much social science, where distinguishing correlation from causation is essential. Their ability to reproduce social complexity also remains uneven and unproven, meaning simulated populations may not faithfully reflect real human diversity and dynamics.

The authors additionally highlight urgent ethical concerns, including privacy risks and the potential for these systems to generate or spread misinformation. To move the field forward, the review identifies critical research directions for harnessing LLMs' potential while addressing their limitations. As a review article, it synthesises existing work rather than presenting new experimental results, and its conclusions describe the state of a rapidly developing methodology.

Subject of Research: Roles and limitations of large language models in computational social science

Article Title: Large language models for computational social science

Article References: Xu, F., Shao, C., Chen, L., Zeng, Q., Chen, Z., Sukiennik, N., Wang, J., Gao, C., Wang, H., Lian, J., Meng, Q., Xie, X., Li, Y., & Evans, J. (2026). Large language models for computational social science. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02539-9

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02539-9

Keywords: large language models, computational social science, Nature Human Behaviour, AI agents, causal inference, research ethics, simulation, misinformation, hypothesis generation, sociology, review article, Large