抄録
This research examines the feasibility of using ChatGPT as a subject in an auction experiment. The author will also test the idea of giving ChatGPT a “persona” to participate as a human. Without personas, the author finds overbidding in the FPA and slight underbidding in the SPA; the FPA results are consistent with prior studies, but the SPA results differ. Given the persona as an excellent economics student, ChatGPT bid close to the theoretical prediction in the FPA but significantly underbid in the SPA. These results raise questions about using LLM-based AI subjects in auctions.
