Human choice behavior, including route choice, exhibits systematic biases that deviate from full rationality assumptions. Cumulative prospect theory (CPT) is widely recognized as an effective framework for characterizing such patterns. However, large-scale application in simulation and agent-based modeling critically depends on specifying individual-level CPT parameters, which remains a major bottleneck. Conventional approaches rely on surveys and controlled experiments to calibrate these parameters, but they often fail to generalize and capture the full diversity of human decision-making.
To address this, the paper investigates whether large language models (LLMs) can reproduce human behavioral biases without explicit specification of prospect-theoretic parameters. Using route choice as a representative scenario, a behavioral evaluation framework is designed to systematically compare LLM-generated decisions with established human behavioral patterns predicted by CPT. Experimental results demonstrate that LLMs can reproduce non-rational choice biases and exhibit decision behaviors consistent with prospect-theoretic effects under uncertainty. These findings suggest that generative AI models may offer a scalable alternative for modeling human decision processes and provide a promising foundation for next-generation large-scale agent-based simulations and AI-driven behavioral research.
Blogger's Review: This paper innovatively applies large language models to human behavior modeling, showcasing their potential in understanding and predicting irrational decision-making. This research not only provides new tools for behavioral science but also theoretical support for future intelligent agent systems. Its results may inspire new thoughts on leveraging AI for simulating human decision-making.