Social Networks of LLM Agents

Summary

Asks whether an interacting population of LLM agents aggregates genuine knowledge or collapses into false consensus, and argues classical social-network belief-aggregation models overstate pooling because LLM agents attend to only part of what they are exposed to. Their SNLA framework models how much each agent actually influences others rather than how the network merely connects them, and proves that narrow attention causes herding with effective sample size bounded regardless of population size, while wide attention recovers wisdom-of-crowds behaviour only when the exposure graph is undirected and degree-regular. That undirected/degree-regular condition is a precise structural statement of when a polyarchy-like arrangement delivers crowd wisdom and when it does not.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a high-priority candidate for the Architecture as Epistemology (ISR) paper (angle: MAS topology / structure vs. aggregation rule; correlated errors / homogeneity across LLM agents; aggregating LLM judgments). Verified against: Opened https://arxiv.org/abs/2607.03695 (arXiv abstract page, v1 4 Jul 2026)

Critique / open questions

Queued — not yet assessed.

Key quotes

Large language model (LLM) agents are increasingly deployed in interacting populations, raising the question of what such populations come to believe collectively. (abstract)

Whether a population aggregates genuine knowledge or collapses into a false consensus directly affects how much such systems can be trusted. (abstract)