[2A1-GS-10] AI application:

[2A1-GS-10-01] Can Large Language Models Reproduce Human-like Temporal Preferences?

〇Katsumasa Yoshikawa¹, Masaya Ohagi¹, Junya Takayama¹ (1. SB Intuitions Corp.)

Keywords:LLM, Behavioral Economics, Temporal Discounting

This study explores the application of large language models (LLMs) to validate the hyperbolic temporal discounting model from behavioral economics. We focus on replicating time preference experiments using LLMs as a case study for their potential in social science research.

Our methodology involves inputting standard time preference survey questions into multiple LLMs and analyzing their responses. The results indicate that some LLMs, particularly GPT-4, produce a reasonable replication of hyperbolic temporal discounting experiments, although there are significant variations across different models. We also demonstrate that providing persona information to LLMs can systematically alter their temporal preferences.

This research showcases the potential of LLMs in simulating social science experiments and highlights the importance of model selection and persona specification. Our findings open new avenues for interdisciplinary research at the intersection of natural language processing, artificial intelligence, and behavioral economics.