The stability of multi-agent systems consisting of Autonomous Large AI Models (AutoLAMAs)

This project will explore the stability of agentic AI-driven systems by modeling interactions between heterogeneous agents in a game-theoretic framework. Initially, we will construct systems where agents interact strategically, with some having the ability to modify the interaction graph at a cost. Building on these foundations, we will extend our analysis to more complex interaction mechanisms, such as natural language and dialogue, to deepen our understanding of stability in multi-agent environments. We can then potentially explore intervention mechanisms to make AI agents less exploitable on these interaction graphs.