Stochastic Self-Organization in Multi-Agent Systems
Nurbek Tastan, Samuel Horvath, Karthik Nandakumar
TL;DR
SelfOrg tackles the orchestration challenge in LLM-based multi-agent systems by forming a per-instance directed acyclic graph (DAG) from agent responses and using a Shapley-inspired, embedding-based contribution estimate to route information from high- to low-contributing agents. It avoids pretrained topology generators and external judges, enabling a lightweight, self-organizing collaboration that adapts to stochastic agent outputs. The paper provides theoretical bounds on the contribution approximation and demonstrates that multiple agents amplify correct signals, especially in weak-backend regimes, with strong empirical gains across a range of benchmarks and backbones. Overall, SelfOrg offers a practical, scalable approach to MAS coordination that improves robustness and performance without additional supervision or training.
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.