Behavioral Economics of AI: LLM Biases and Corrections

Do generative AI models, as epitomized and popularized by large language models (LLMs), exhibit systematic behavioral biases in economic and financial decisions? If so, how can we mitigate these biases? Following the cognitive psychology literature and the experimental economics studies, we conduct the most comprehensive set of experiments to date---originally designed to document human biases---on prominent LLM families with variations in model version and scale. We document systematic patterns in the behavioral biases that LLMs exhibit. For experiments concerning the psychology of preferences, LLM responses become increasingly irrational and human-like as the models become more advanced or larger; however, for experiments concerning the psychology of beliefs, the most advanced large-scale models frequently generate rational responses. Further exploring various methods for correcting these behavioral biases reveals that prompting LLMs to make rational decisions according to the Expected Utility framework seems the most effective.