Strategic Communication and Language Bias in Multi-Agent LLM Coordination

2025-07-30

AI Summary

This study investigates how linguistic formulations amplify strategic coordination biases among agents in multi-agent systems through communication. Using the FAIRGAME framework, we conduct controlled experiments with GPT-4o and Llama 3 Maverick models in both one-shot and repeated games, systematically varying linguistic variants, agent personas, and game structures, while comparing conditions with and without inter-agent communication. Results demonstrate that communication—while enhancing cooperation—significantly exacerbates language-driven behavioral biases. Crucially, linguistic choice, agent persona, and task structure interact to jointly modulate coordination efficacy. This work provides the first empirical evidence of communication’s “double-edged sword” effect on linguistic bias: it facilitates alignment yet concurrently reinforces systematic deviations rooted in language design. We thereby establish linguistic engineering as a critical intervention dimension for strategic regulation in multi-agent systems.

Technology Category

• Multiagent Systems: Agent Communication
• Game Theory and Economic Paradigms: Coordination and Collaboration
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Large Language Model (LLM)-based agents are increasingly deployed in multi-agent scenarios where coordination is crucial but not always assured. Previous studies indicate that the language used to frame strategic scenarios can influence cooperative behavior. This paper explores whether allowing agents to communicate amplifies these language-driven effects. Leveraging the FAIRGAME framework, we simulate one-shot and repeated games across different languages and models, both with and without communication. Our experiments, conducted with two advanced LLMs, GPT-4o and Llama 4 Maverick, reveal that communication significantly influences agent behavior, though its impact varies by language, personality, and game structure. These findings underscore the dual role of communication in fostering coordination and reinforcing biases.

Problem

• Explores how communication affects language-driven cooperation in multi-agent LLMs
• Investigates impact of language, personality, and game structure on agent behavior
• Examines dual role of communication in coordination and bias reinforcement

Innovation

• FAIRGAME framework simulates strategic games
• Tests communication impact across languages models
• Analyzes bias coordination via agent interactions