Multi-agent discussion gains less when dissent is withheld

2026-09-29

AI Summary

This study addresses the problems of consensus distortion and unclear performance gains caused by dissent concealment in multi-agent discussions. We construct a parsimonious four-behavior model to provide the first unified analytical framework for conflict evolution mechanisms. Based on Bayesian estimation, we propose a theoretical framework centered on a critical concealment rate, quantifying the dynamic processes of dissent concealment, internalization, and revision, while identifying the boundary conditions under which discussion outperforms voting. Empirical validation using benchmarks such as HiddenBench and large language model-based multi-agent experiments demonstrates that discussion gains diminish monotonically as the concealment rate increases, whereas prohibiting concealment or disabling reasoning significantly improves accuracy. This work offers both theoretical foundations and empirical evidence for understanding and optimizing multi-agent collaboration.

Abstract

Multi-agent systems of LLMs add discussion to majority voting and are therefore expected to be more capable. However, empirical reports conflict on whether discussion improves accuracy or leads to an incorrect consensus. Here, we introduce a parsimonious model that explains when discussion improves accuracy and when it ends in an incorrect consensus, built from four behaviors repeatedly observed in LLM agents: (1) withholding dissent, (2) internalizing a stated answer, (3) reconsidering after seeing dissent, and (4) correcting toward the correct answer. The model shows that discussion can overturn an incorrect initial majority only when the withholding rate $c$ is below a critical rate $c^* = γ/(γ+ a)$, set by the net correction rate $γ$ and the internalization rate $a$. We estimate these rates from conversation logs with a Bayesian method and place LLM teams relative to $c^*$. As the model predicts, the gain from discussion shrinks as withholding rises, across LLMs and on a hidden profile benchmark, HiddenBench, and MedEInst. Instructing agents not to withhold dissent increases this gain. Turning reasoning off also increases the gain, because reasoning raises the internalization rate $a$ and keeps agents from reconsidering a minority answer. These findings reconcile the conflicting reports and identify when discussion outperforms majority voting.

Problem

• multi-agent systems
• large language models
• multi-agent discussion
• withholding dissent
• majority voting

Innovation

• Multi-agent discussion
• Withholding dissent
• Parsimonious model
• Bayesian estimation
• Majority voting