arXiv cs.MA· Arnesh Batra, Aniket Khandelwal, Jashn Khemani, Krish Thukral, Arush Gumber, Denis Federico Lim, Ujwal Kumar, Phan Xuan Tan·· 11 小时前AI 评分34
谁说服了谁？追踪 LLM 在多智能体系统中如何改变想法
Who Convinced Whom? Tracing How LLMs Change Their Minds in Multi-Agent Systems
AI 导读
研究提出 MAD Bench，一个覆盖知识推理、价值观与道德判断、安全与拒答的 2400 题基准，用轮次匹配的独立反思作为对照评估三个开源权重模型。在 Qwen3.5-9B 上，辩论中生成理由前的强制选择读数与最终答案一致率为 65.4%，反思仅 22.3%。辩论相对反思带来小幅准确率提升，但全票错误率高出 3.1-7.4 个百分点，更长辩论中共识更持久且智能体正确性相关性更高。
正文
Abstract:Multi-agent debate (MAD) aims to improve language-model decisions through the exchange of arguments, yet agreement after debate need not reflect independent corroboration. We investigate how peer communication influences individual decisions and collective outcomes through behavioral and internal interventions. We introduce MAD Bench, a 2,400-question benchmark spanning knowledge and reasoning, values and moral judgment, and safety and refusal, and evaluate three open-weight models using a round-matched independent-reflection control. Among answer-changing responses, a forced-choice readout before rationale generation matches the eventual answer more often in debate than in reflection (65.4% versus 22.3% on Qwen3.5-9B). Message ablations reveal model-dependent effects of peer answers and reasoning, while attributing identical text to the receiver rather than another agent substantially changes its influence. For the tested source descriptions, stated track record affects answer preferences more strongly than professional-status cues across all three models. Attention masking produces different changes in decision margins across the three checkpoints. On fully parsed, reference-labelled non-numeric questions, debate yields small estimated accuracy improvements over initial majority voting, while increasing the rate of unanimous wrong answers by 3.1-7.4 percentage points relative to reflection. Longer debates show persistent consensus and greater correlation between agents' correctness than reflection. These findings show that stronger consensus can coexist with more shared errors, motivating evaluations that assess dependence and unanimous errors alongside accuracy.
| Subjects: | Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.35930 [cs.MA] |
| (or arXiv:2610.35930v1 [cs.MA] for this version) | |
| https://doi.org/10.48550/arXiv.2610.35930 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Aniket Khandelwal [view email]
[v1]
Sun, 18 Oct 2026 16:04:25 UTC (293 KB)
来源：arXiv cs.MA · arxiv.org