Social Networks of LLM Agents

2026-07-03

AI Summary

Existing social network models struggle to capture belief aggregation in populations of large language model (LLM) agents, as they neglect information filtering driven by limited attention and fail to distinguish genuine consensus from conformity. This work proposes the SNLA framework, which, for the first time, integrates agent attention mechanisms into social influence modeling by characterizing actual influence exerted rather than relying solely on network structure. We reveal how attention breadth and network topology jointly govern collective belief dynamics. Theoretically, we prove that narrow attention induces herding under bounded effective sample sizes, whereas wide attention recovers the wisdom of crowds only in undirected degree-regular graphs. Combining graph-theoretic analysis, tractable agent-based models, and multi-agent LLM simulations, we empirically validate a phase transition from herding to collective intelligence across controlled experiments and operator-controlled variants of three benchmark tasks.

Abstract

Large language model (LLM) agents are increasingly deployed in interacting populations, raising the question of what such populations come to believe collectively. Whether a population aggregates genuine knowledge or collapses into a false consensus directly affects how much such systems can be trusted. Classical social-network models assume that the network itself determines how beliefs combine. This assumption breaks down for LLM agents, whose limited attention takes in only part of what they are exposed to, so these models overstate how much information a population actually pools and cannot tell genuine consensus from herding. We introduce SNLA, a framework that models how much each agent actually influences others, rather than merely how the network connects them. This influence depends on each agent's position in the network and on how sharply attention focuses. Theoretically, we show on a tractable proxy that narrow attention causes herding, where the effective sample size stays bounded regardless of population size, while wide attention recovers wisdom-of-crowds behavior only when the exposure graph is undirected and degree-regular. Empirically, a controlled testbed validates these predictions directly, and the herding-wisdom transition reproduces on operator-controlled variants of three multi-agent LLM benchmarks.

Problem

• LLM agents
• social networks
• herding
• collective belief
• attention

Innovation

• SNLA
• attention mechanism
• herding
• wisdom of crowds
• LLM agents