Evaluating Cooperation in LLM Social Groups through Self-Organizing Leadership

Ryan Faulkner, Anushka Deshpande, David Guzman Piedrahita, Joel Z Leibo, Zhijing Jin

Keywords: Multi-Agent, Language Modelling, Computational Social Science

Abstract: Governing common-pool resources requires agents to develop enduring strategies through cooperation and self-governance to avoid collective failure. While foundation models have shown potential for cooperation in these settings, existing multi-agent research lacks insight into how agents internally organize and the mechanisms through which they achieve collective action. In this work, we propose a governance framework that simulates leadership through elected personas and candidate-driven agendas to coordinate group behavior. Our experiments demonstrate that structured leadership significantly improves social welfare and survival rates, while our analysis reveals how specific rhetorical strategies and social influence drive group cooperation. This work lays the foundation for developing prosocial, self-governing multi-agent systems capable of navigating complex resource dilemmas.

Submission Number: 57