Understanding the Information Propagation Effects of Communication Topologies in LLM-based Multi-Agent Systems

Summary

Provides a causal framework for how agent outputs, correct or erroneous, propagate under communication topologies of varying density, and finds moderately sparse structures best balance error suppression against useful information sharing. The mechanism-level counterpart to the Sah-Stiglitz hierarchy/polyarchy trade-off (type I versus type II error filtering) expressed in LLM-agent terms. Their EIB-learner mixes dense and sparse patterns, speaking directly to why an intermediate structure may dominate pure hierarchy or pure polyarchy.

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). Verified against: Opened https://arxiv.org/abs/2505.23352 (arXiv abstract page, v1 29 May 2025)

Critique / open questions

Queued — not yet assessed.

Key quotes

The communication topology in large language model-based multi-agent systems fundamentally governs inter-agent collaboration patterns, critically shaping both the efficiency and effectiveness of collective decision-making. (abstract)

While recent studies for communication topology automated design tend to construct sparse structures for efficiency, they often overlook why and when sparse and dense topologies help or hinder collaboration. (abstract)