Do LLM Agents Cooperate - Insights from GovSim and SanctSim

Explore how large language model agents behave in cooperative scenarios through this research presentation examining two groundbreaking studies on AI cooperation in social dilemmas. Learn about GovSim, which investigates common-pool resource management where agents must balance immediate extraction with long-term sustainability, revealing that most LLMs struggle to maintain cooperation except for the most advanced models. Discover insights from SanctSim, which studies public goods games with institutional mechanisms and identifies four distinct behavioral patterns: cooperators, oscillators, deteriorators, and rigid strategists. Understand the counterintuitive finding that reasoning-optimized LLMs often underperform compared to traditional LLMs in sustaining cooperation, highlighting that enhanced reasoning capabilities don't automatically guarantee cooperative behavior. Gain valuable insights into how cooperation depends on communication strategies and moral reasoning, with important implications for designing collaborative multi-agent AI systems and ensuring safe deployment of AI in multi-agent environments.