This article examines the extent to which ChatGPT can function as a proxy for human respondents in social science research. The authors explore the simulation of human attitudes and assess how large language model outputs compare with expected human patterns in attitude-related tasks. The paper contributes to methodological discussions on the use of LLMs for behavioral and attitudinal research, with implications for validity, bias, and the limits of using AI-generated responses as substitutes for human data.

Author of the paper: Antonina Rafikova, Anatoly Voronin

Publisher or journal of publication: Journal of Computational Social Science (Springer Nature)

The paper is available at the following link.