Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generation

Summary

Directly attributes diversity loss in open-ended ideation to the interaction structure rather than model insufficiency, which is this paper's core claim stated in the negative. It reports three findings the paper can build on: authority-driven (hierarchical) dynamics suppress semantic diversity relative to junior-dominated groups, dense communication topologies accelerate premature convergence, and group-size scaling shows diminishing returns; it also names a compute-efficiency paradox in which stronger aligned models yield less marginal diversity. This is the closest empirical precedent for structure-moves-novelty and must be cited and distinguished, since it measures idea diversity but not justification soundness or traceability.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-09-07 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; epistemic quality of LLM outputs). Verified against: Opened https://arxiv.org/abs/2604.18005 (arXiv abstract page; confirmed title, 8 authors, cs.MA, v1 20 Apr 2026 / v2 21 Apr 2026, Findings of ACL 2026, full abstract read)

Critique / open questions

Queued — not yet assessed.

Key quotes

Multi-agent systems (MAS) are increasingly used for open-ended idea generation, driven by the expectation that collective interaction will broaden the exploration diversity. (abstract)

However, when and why such collaboration truly expands the solution space remains unclear. (abstract)