GENERATIVE AI AND THE SOCIAL SCIENCES: RETHINKING METHODOLOGY WITH LARGE LANGUAGE MODELS AND SYNTHETIC POPULATIONS

The advent of large language models (LLMs) promises a new paradigm shift in social science methodology with unprecedented powers to simulate human populations, create synthetic survey data, and model complex social dynamics.This paper explores the epistemological underpinnings and methodological considerations of generative AI in social research, engaging with this topic critically.We believe that the transformative promise of LLM is that they can be used to tackle persistent data issues such as missing observations, privacy concerns and counterfactual inaccessibility, but that requires radical rethinking of traditional statistical assumptions.