Evaluating LLMs in Open-Source Games

2025-11-29

AI Summary

This study investigates the capacity of large language models (LLMs) to evolve cooperative and deceptive strategies in open-source, programmable game-theoretic environments. Method: We construct a multi-agent repeated-games platform built upon an open-source game framework, integrating formal game-theoretic modeling, strategy-prediction classification, and evolutionary algorithms to enable verifiable analysis of LLM agent behavior. We introduce the novel concept of “programmatic equilibrium” to characterize the convergence properties of adaptive cooperation among agents under code transparency. Results: Experiments demonstrate that LLMs spontaneously generate highly adaptive cooperative strategies in both binary and evolutionary games, with behavior significantly modulated by environmental incentive structures. Crucially, open-source games are validated as a controllable, reproducible, and interpretable experimental paradigm for guiding and analyzing the emergence of multi-agent cooperation.

Technology Category

• Game Theory and Economic Paradigms: Cooperative Game Theory
• Multiagent Systems: Mechanism Design
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models' (LLMs) programming capabilities enable their participation in open-source games: a game-theoretic setting in which players submit computer programs in lieu of actions. These programs offer numerous advantages, including interpretability, inter-agent transparency, and formal verifiability; additionally, they enable program equilibria, solutions that leverage the transparency of code and are inaccessible within normal-form settings. We evaluate the capabilities of leading open- and closed-weight LLMs to predict and classify program strategies and evaluate features of the approximate program equilibria reached by LLM agents in dyadic and evolutionary settings. We identify the emergence of payoff-maximizing, cooperative, and deceptive strategies, characterize the adaptation of mechanisms within these programs over repeated open-source games, and analyze their comparative evolutionary fitness. We find that open-source games serve as a viable environment to study and steer the emergence of cooperative strategy in multi-agent dilemmas.

Problem

• Evaluating LLMs' strategy prediction and classification in open-source games
• Analyzing cooperative and deceptive strategies in evolutionary game settings
• Studying program equilibria emergence in multi-agent dilemmas

Innovation

• LLMs program in open-source games for transparency
• Evaluate LLM strategies and program equilibria evolution
• Study cooperative strategy emergence in multi-agent dilemmas