Rethinking Multi-Agent Intelligence Through the Lens of Small-World Networks

2025-12-19

AI Summary

Existing LLM-based multi-agent systems (MAS) employ communication topologies—typically fully connected or manually designed sparse graphs—without principled theoretical guidance, overlooking the robustness and efficiency benefits of small-world (SW) networks in collaborative reasoning. Method: This work pioneers SW topology as a structural prior for MAS design, integrating insights from neuroscience and complex network theory. We propose a dynamic rewiring mechanism guided by LLM uncertainty signals—particularly semantic entropy—to adaptively configure SW topologies per task. Our approach unifies SW network modeling, multi-agent debate (MAD), semantic-entropy-based uncertainty quantification, and a dynamic graph rewiring algorithm. Results: Experiments demonstrate significantly stabilized consensus evolution trajectories—without sacrificing accuracy or token efficiency—while enabling scalable, heterogeneous, and difficulty-adaptive collaborative inference.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Large language models (LLMs) have enabled multi-agent systems (MAS) in which multiple agents argue, critique, and coordinate to solve complex tasks, making communication topology a first-class design choice. Yet most existing LLM-based MAS either adopt fully connected graphs, simple sparse rings, or ad-hoc dynamic selection, with little structural guidance. In this work, we revisit classic theory on small-world (SW) networks and ask: what changes if we treat SW connectivity as a design prior for MAS? We first bridge insights from neuroscience and complex networks to MAS, highlighting how SW structures balance local clustering and long-range integration. Using multi-agent debate (MAD) as a controlled testbed, experiment results show that SW connectivity yields nearly the same accuracy and token cost, while substantially stabilizing consensus trajectories. Building on this, we introduce an uncertainty-guided rewiring scheme for scaling MAS, where long-range shortcuts are added between epistemically divergent agents using LLM-oriented uncertainty signals (e.g., semantic entropy). This yields controllable SW structures that adapt to task difficulty and agent heterogeneity. Finally, we discuss broader implications of SW priors for MAS design, framing them as stabilizers of reasoning, enhancers of robustness, scalable coordinators, and inductive biases for emergent cognitive roles.

Problem

• Design optimal communication topologies for multi-agent systems
• Apply small-world network theory to stabilize agent consensus
• Introduce uncertainty-guided rewiring to scale multi-agent coordination

Innovation

• Small-world networks balance clustering and integration
• Uncertainty-guided rewiring adapts to task difficulty
• SW structures stabilize consensus and enhance robustness