[论文] The Interaction Tax: When Communication Erases Diversity in Multi-Agen...

论文概要

研究领域: ML 作者: Summer Eunhyung Ann, Haokun Liu, Chenhao Tan 发布时间: 2025-08-26 arXiv: 2508.17620

中文摘要

多智能体LLM交互究竟是有益还是有害？一些研究报道了辩论、批判循环和混合智能体综合的收益，而其他研究发现在同等预算下交互增加成本却不提升质量，或独立采样已能捕获多智能体收益。本文认为这一矛盾部分反映了缺失的区分：并非所有多智能体通信都是等价的。不同模型家族找到结构不同的解决方案，但当智能体读取彼此的完整输出时，其提议在一轮内就会收敛，抹除了使用多个模型的多样性动机。我们称此为交互税（interaction tax）。在11个验证器评分的优化任务上的测试表明，在匹配预算下，完整方案交互是一个弱默认设置。独立提议生成避免了这种坍缩。完整方案交互主要使智能体停留在它们看到的第一个解决方案附近，而非尝试不同方法；而批判仅在违反的规则容易被LLM发现和修复时才有帮助。结果表明，多智能体性能更依赖于交换的信息而非智能体数量，且交互仅在智能体在正确时间共享正确信息时才有效。

原文摘要

Does multi-agent LLM interaction help or hurt? Some work reports gains from debate, critique loops, and mixture-of-agents synthesis, while other work finds that interaction adds cost without improving quality under equal budgets, or that independent sampling already captures multi-agent gains. We argue this contradiction partly reflects a missing distinction, because not all multi-agent communication is equal. Different model families find structurally different solutions, but when agents read each other's complete outputs, their proposals converge within one round, erasing the diversity that motivates using multiple models. We call this the interaction tax. We test 11 verifier-scored optimization tasks under matched budgets and find that full-solution interaction is a weak default. Indepe...

--- *自动采集于 2026-08-26*

#论文 #arXiv #ML #小凯