Collaboration and Conflict between Humans and Language Models through the Lens of Game Theory

2025-09-05

AI Summary

This study investigates how large language models (LLMs) cooperate with humans and evolve strategies in multi-round repeated prisoner’s dilemma games. We adopt an Axelrod-style tournament framework to systematically evaluate LLMs’ long-term interactions with 240 canonical game-theoretic strategies and conduct controlled strategy-switching experiments to assess their dynamic adaptability. Our work provides the first empirical evidence that LLMs exhibit human-like cooperative traits—namely, initial friendliness, provocability, and forgiveness—and rapidly adjust to abrupt strategy shifts. Notably, their cooperation stability and environmental adaptability match or surpass those of optimal classical strategies, including Tit-for-Tat and Win-Stay-Lose-Shift. These findings establish a testable theoretical foundation and empirical basis for understanding long-term cooperation mechanisms in human–AI hybrid societies.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Game Theory and Economic Paradigms: Cooperative Game Theory
• 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
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Language models are increasingly deployed in interactive online environments, from personal chat assistants to domain-specific agents, raising questions about their cooperative and competitive behavior in multi-party settings. While prior work has examined language model decision-making in isolated or short-term game-theoretic contexts, these studies often neglect long-horizon interactions, human-model collaboration, and the evolution of behavioral patterns over time. In this paper, we investigate the dynamics of language model behavior in the iterated prisoner's dilemma (IPD), a classical framework for studying cooperation and conflict. We pit model-based agents against a suite of 240 well-established classical strategies in an Axelrod-style tournament and find that language models achieve performance on par with, and in some cases exceeding, the best-known classical strategies. Behavioral analysis reveals that language models exhibit key properties associated with strong cooperative strategies - niceness, provocability, and generosity while also demonstrating rapid adaptability to changes in opponent strategy mid-game. In controlled "strategy switch" experiments, language models detect and respond to shifts within only a few rounds, rivaling or surpassing human adaptability. These results provide the first systematic characterization of long-term cooperative behaviors in language model agents, offering a foundation for future research into their role in more complex, mixed human-AI social environments.

Problem

• Investigating cooperative and competitive behavior of language models in multi-party settings
• Examining long-horizon interactions and human-model collaboration through game theory
• Analyzing language model adaptability and strategy evolution in repeated prisoner's dilemma

Innovation

• Iterated prisoner's dilemma framework
• Axelrod-style tournament evaluation
• Strategy switch adaptability analysis