Super-additive Cooperation in Language Model Agents
This research addresses the challenge of designing multi-agent AI systems that align with human values, though it is incremental in applying existing theories to language models.
The study tackled the problem of fostering cooperative behavior in language model agents by simulating a Prisoner's Dilemma tournament with team dynamics and inter-group competition, resulting in a substantial boost in both overall and initial one-shot cooperation levels.
With the prospect of autonomous artificial intelligence (AI) agents, studying their tendency for cooperative behavior becomes an increasingly relevant topic. This study is inspired by the super-additive cooperation theory, where the combined effects of repeated interactions and inter-group rivalry have been argued to be the cause for cooperative tendencies found in humans. We devised a virtual tournament where language model agents, grouped into teams, face each other in a Prisoner's Dilemma game. By simulating both internal team dynamics and external competition, we discovered that this blend substantially boosts both overall and initial, one-shot cooperation levels (the tendency to cooperate in one-off interactions). This research provides a novel framework for large language models to strategize and act in complex social scenarios and offers evidence for how intergroup competition can, counter-intuitively, result in more cooperative behavior. These insights are crucial for designing future multi-agent AI systems that can effectively work together and better align with human values. Source code is available at https://github.com/pippot/Superadditive-cooperation-LLMs.