Cooperation in AI agent networks changes with language and peers
The Butterfly Defects: Tracing Cooperation in LLM Agent Networks
Computation and Language
Summary
When large language models act as agents working together, their willingness to cooperate is not fixed. The authors studied a game that tests cooperation and found that small changes—like the other models they interact with or how choices are described—can make models switch from cooperating to competing. Even subtle cues about uncertainty or reversing labels of choices affected behavior dramatically. This shows that cooperation depends on the network and language, not just on the models themselves.
What this means in practice
- •For multi-agent system developers: Tailor interaction protocols and language framing to promote cooperation among diverse AI agents in networked environments.
- •For ai behavior analysts: Analyze how subtle changes in communication and agent mix impact cooperation for better prediction of collective agent outcomes.
Authors
Maria Sampani, Maria Lymperaiou, Angeliki Dimitriou, Giorgos Filandrianos, Chrysoula Zerva, Giorgos Stamou
Abstract
As large language models (LLMs) increasingly operate as interconnected agents, a fundamental question emerges: does individual cooperation translate into collective cooperation? We investigate this question through the Iterated Prisoner's Dilemma (IPD), moving from dyadic interactions to homogeneous and heterogeneous networks of open-weight LLMs. Our findings reveal that cooperation is highly sensitive to model composition and seemingly minor linguistic changes. Models that cooperate with identical peers can shift toward defection in mixed populations. A simple uncertainty cue can increase or decrease cooperation, redistribute costs, and affect unexposed agents several hops away. Even reversing action-label order can dramatically alter behavior, with uncertainty framing sometimes amplifying this sensitivity. Together, these findings challenge the notion of cooperation as a stable model-level property, revealing how strategic behavior depends on interaction partners, network positions, and the language used to frame decisions.