Bay Area Tech Economics Seminar with Abhishek Nagaraj

Advancing Social Science Research with LLMs

In this talk, I argue that Large Language Models (LLMs) offer social scientists a novel experimental tool: a digital “lab mouse.” Because LLMs encode human preferences, they make it possible to study aspects of human behavior computationally, at lower cost and greater scale than traditional human-subject experiments. LLMs also offer an unprecedented level of experimental control, which further enhances their appeal as a research tool. I will present results from several projects that: (a) establish this possibility for economic models, (b) develop applications across different social-scientific domains, and (c) highlight both the power and limitations of this approach.