Personality Alignment of LLM Agents to Humans within Behavioral Games

In this work, the alignment of LLM agents' actions to those observed in human populations is studied in the context of single-term games used in behavioral economics research. Agents were prompted with a randomized profile of "Big-5" personality traits, along with the rules of each game, and then played several rounds of the game against another agent. Over 1,000 rounds of each game were played by unique pairs of AI agents.

Intern: Jeffrey Wang

Mentors: Cristopher Ratto and Rik Bose (REDD)