Stochastic Self-Organization in Multi-Agent Systems

2025-10-01

AI Summary

Existing multi-agent systems (MAS) often rely on fixed communication topologies or external supervision, limiting adaptability and scalability. To address this, we propose SelfOrg—a fully decentralized, training-free framework enabling large language model (LLM) agents to self-organize via response-driven dynamic communication. At its core, SelfOrg employs real-time contribution estimation using approximate Shapley values to construct and iteratively update a directed acyclic graph (DAG), thereby optimizing information flow paths. Crucially, it requires no pre-trained graph generators, edge optimization, or external LLM evaluators. We theoretically prove that SelfOrg improves overall correctness and ensures high-quality responses dominate the information flow, significantly enhancing robustness—especially under weak LLM backends. Empirical results demonstrate consistent effectiveness across both strong and weak LLM settings, with substantial performance gains over state-of-the-art methods in low-capability regimes.

Technology Category

• Multiagent Systems: Agent Communication
• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Multi-agent systems (MAS) based on Large Language Models (LLMs) have the potential to solve tasks that are beyond the reach of any single LLM. However, this potential can only be realized when the collaboration mechanism between agents is optimized. Specifically, optimizing the communication structure between agents is critical for fruitful collaboration. Most existing approaches rely on fixed topologies, pretrained graph generators, optimization over edges, or employ external LLM judges, thereby adding to the complexity. In this work, we introduce a response-conditioned framework that adapts communication on-the-fly. Agents independently generate responses to the user query and assess peer contributions using an approximation of the Shapley value. A directed acyclic graph (DAG) is then constructed to regulate the propagation of the responses among agents, which ensures stable and efficient message transmission from high-contributing agents to others. This graph is dynamically updated based on the agent responses from the previous collaboration round. Since the proposed framework enables the self-organization of agents without additional supervision or training, we refer to it as SelfOrg. The SelfOrg framework goes beyond task- and query-level optimization and takes into account the stochastic nature of agent responses. Experiments with both strong and weak LLM backends demonstrate robust performance, with significant gains in the weak regime where prior methods collapse. We also theoretically show that multiple agents increase the chance of correctness and that the correct responses naturally dominate the information flow.

Problem

• Optimizing dynamic communication structures in multi-agent systems
• Enabling self-organization without external supervision or training
• Improving collaboration robustness for weaker LLM agents

Innovation

• Dynamic communication structure adapts on-the-fly
• Shapley value approximation assesses peer contributions
• Directed acyclic graph regulates response propagation