Copies or Sources? Measuring How LLM Aggregators Count Restated Evidence in Multi-Agent Systems
作者: Jianxin Gao, Runze Li, Tianyi Yu, Liangwei Ren, Bohan Chen, Zining Wang
分类: cs.AI
发布日期: 2026-10-05
💡 一句话要点
提出一种新方法以优化多智能体系统中LLM聚合器的证据计数
🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 多智能体系统 大型语言模型 信息聚合 复制权重 决策优化
📋 核心要点
- 现有的多智能体系统在处理信息时,往往会重复计数相同的证据,导致信息的冗余和误导。
- 本文提出了一种新的聚合器设计,通过引入复制权重来区分信息来源与信息陈述,从而优化信息计数。
- 实验结果表明,采用新方法后,早期承诺的比例显著降低,且信息的真实证实能力得以提升。
📝 摘要(中文)
基于大型语言模型(LLMs)的多智能体系统在转述观察时常常会重复信息。本文提出了一种聚合器应计数来源而非陈述的观点。通过将报告的概率转换为独立阅读的单位,赋予每个转述一个复制权重,0表示计数来源,1表示计数每个陈述。研究在三种测试环境中验证了这一方法,结果显示,转发的复制在新阅读中仅占0.06到0.42,且某些回复会计数每个陈述。通过调整代理的沟通方式,显著降低了早期承诺的比例,提升了信息的真实证实能力。
🔬 方法详解
问题定义:本文旨在解决多智能体系统中信息转述导致的冗余计数问题。现有方法未能有效区分信息来源与重复陈述,造成信息处理的低效与误导。
核心思路:论文提出通过将报告的概率转化为独立阅读的单位,赋予每个转述一个复制权重,以此来优化聚合器的计数方式,确保其计数来源而非重复陈述。
技术框架:研究设计了三种测试环境,分别为精确的贝叶斯预言机的消息日志、附加复制的网页文档,以及在四种沟通协议下由LLM代理团队撰写的日志。通过这些环境,验证了不同计数策略的有效性。
关键创新:最重要的创新在于引入了复制权重的概念,使得聚合器能够在不同的计数策略下进行有效的信息处理。这一方法与传统的简单计数方式有本质区别。
关键设计:在实验中,设定了不同的复制权重参数,并通过调整代理的沟通方式,显著降低了早期承诺的比例,从而提升了信息的真实证实能力。
🖼️ 关键图片
📊 实验亮点
实验结果显示,在控制日志中,转发的复制权重仅为0.08或更低,且通过调整代理的沟通方式,早期承诺的比例从11.2%降至1.1%。这些结果表明新方法在信息处理上的显著提升。
🎯 应用场景
该研究的潜在应用领域包括智能客服、信息检索和多智能体协作系统等。通过优化信息计数方式,可以提升系统的决策效率和信息处理能力,具有重要的实际价值和广泛的应用前景。
📄 摘要(原文)
Multi-agent systems built on large language models (LLMs) restate observations as a matter of course: relays forward them, shared boards repeat them and discussion rounds echo them. An aggregator that pools such messages should count sources, not statements. We convert a reported probability into units of independent readings, which assigns every restatement a copy weight, 0 for an aggregator that counts sources and 1 for one that counts every statement, and yields the implied decision under any cost structure. Three testbeds hold the evidence fixed and vary how it is restated: message logs with an exact Bayesian oracle, web documents with appended copies, and logs written by LLM agent teams under four communication protocols. Across four models from three providers, a forwarded copy counts for 0.06 to 0.42 of a new reading, mostly because some replies count every statement. On 5% to 40% of logs that state one reading three times, the reported belief implies an early commitment that the oracle never makes. The models that count copies least and most on controlled logs do so on web copies and agent-written logs as well. A one-paragraph declaration of what a copy contributes brings the copy weight on controlled logs to 0.08 or less. A rule that has agents refer to readings instead of restating them cuts belief-implied early commitment from 11.2% to 1.1% and preserves genuine corroboration.