Do LLMs Take Care of Their Own? Similarity Signals Can Induce Cooperation

Source

Source name: arXiv
Source kind: paper
Authors / organization: Akash Kundu, Emanuel Tewolde, Ratip Emin Berker, Samuel F. Brown, Vincent Conitzer
Venue: arXiv: cs.GT
Published: 2026-08-12
Collected: 2026-08-13
URL: http://arxiv.org/abs/2608.12125v1
PDF: https://arxiv.org/pdf/2608.12125v1
DOI: UNKNOWN
Registry group: academic_apis_and_preprints
Evidence role: canonical_academic_index
Language: en
Country / region: UNKNOWN

Plain-Language Brief

This research introduces a framework to evaluate language models (LLMs) when agents are provided with graded similarity signals. The study finds variations in how different LLMs handle these signals across various cooperation problems and payoff structures. Surprisingly, the dataset used for computing the similarity signal has little impact on induced cooperation. Additionally, the research shows that modern LLMs systematically identify as highly similar when asked to evaluate another model's reasoning. An LLM-behavioral game theoretical model is developed to capture some of their reasoning rationale and support cooperative outcomes.

What the Document Says

The research focuses on language models (LLMs) in strategic interactions where agents encounter each other and need to find mutually beneficial outcomes.
It discusses that prior literature suggests cooperation problems can be resolved when agents follow similar decision making patterns, as in monocultural AI ecosystems.
This paper presents the first framework for evaluating LLM decision making with graded similarity signals.
The research finds that modern LLMs vary greatly in navigating similarity signals across various cooperation problems and payoff structures.
It is discovered that the dataset used for computing the similarity signal has minimal impact on induced cooperation.
Modern LLMs systematically self-identify as highly similar when asked to evaluate another model's chain-of-thought reasoning.
The research develops an LLM-behavioral game theoretical model that captures some of their reasoning rationale.
This model supports cooperative outcomes in equilibrium under sufficiently high similarity scores.

Key Concepts Mentioned

Language Models (LLMs)
Strategic Interactions
Cooperation Problems
Graded Similarity Signals
Monocultural AI Ecosystems
Chain-of-Thought Reasoning
Behavioral Game Theory

Claims Made by the Document

Cooperation problems can be resolved when agents follow similar decision making patterns.
Modern LLMs vary greatly in navigating similarity signals across various cooperation problems and payoff structures.
The dataset used for computing the similarity signal has minimal impact on induced cooperation.
Modern LLMs systematically self-identify as highly similar when asked to evaluate another model's chain-of-thought reasoning.

Evidence Used

Investigation of language models in strategic interactions and finding mutually beneficial outcomes.
Experiments demonstrating the behavior of different LLMs across various cooperation problems, payoff structures, and prompt framings.
Results showing the impact of the dataset used for computing similarity signals on induced cooperation.
Observations of modern LLMs self-identifying as highly similar when evaluating another model's chain-of-thought reasoning.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

As LLM-based agents with user-instructed goals are becoming widely deployed, they increasingly encounter each other in strategic interactions, and face challenges of finding mutually beneficial outcomes. Prior literature has argued that cooperation problems such as the Prisoner's Dilemma are resolvable in settings where agents know they follow very similar decision making patterns, as for example in monocultural AI ecosystems. Following that line of work, this paper introduces the first framework for evaluating LLM decision making when agents are provided with graded similarity signals. Among our findings, we establish that different LLM models vary drastically in how they navigate similarity signals, with some modern models showing consistent behavior across cooperation problems, payoff structures, and prompt framing. Perhaps surprisingly, our experiments also show that the dataset based on which the similarity signal is computed has small to no impact on induced cooperation, and that LLM models systematically self-identify as highly similar when asked to evaluate another model's chain-of-thought reasoning by themselves. Finally, we develop an LLM-behavioral-game-theoretic model that captures some of their reasoning rationale, and show that it can support cooperative outcomes in equilibrium under sufficiently high similarity scores.