As AI agents increasingly operate autonomously in a multi-agent world, they need to learn to cooperate with other agents and with humans to generate mutual benefits. However, cooperation is challenging because the costs are often incurred early on while the benefits are realized later, creating an incentive to defect. How can AI agents cooperate with commitment?
Here, we draw inspiration from legal institutions and contracts that human societies have used to solve principal-agent problems. Contracts provide observable representations of agreements that enable credible commitments through the enforcement of terms. We study the role of contract-based cooperation using LLM-based agents in ext{CT}, a spatial-temporal game that combines bargaining with navigation towards a goal.
We examine a suite of contract representations ranging from formal contracts that compile to code to natural contracts that require reinterpretation. We evaluate agents with different LLM backbones using various sizes and providers. Our findings indicate that self-negotiated contracts can improve cooperative outcomes beyond what is possible with regular trading.
Blogger's Review: This article explores how AI agents can achieve cooperation through self-negotiated contracts, showcasing the potential of contracts in multi-agent environments. The findings provide new insights into future cooperative mechanisms for AI agents, particularly in scenarios requiring trust and commitment. Effective contract design will be a crucial factor in facilitating AI cooperation.