研究大模型评估者在多智能体系统中偏好传播机制，发现架构先验比提示词影响更大。
Contagion Networks: Evaluator Preference Propagation in Multi-Agent LLM Systems
- 构建传染网络框架，量化评估者偏好在智能体间传播程度
- 实验显示偏好传播率在0.157至0.352之间，架构先验主导传播（贡献-63.5%）
- 提出通过扩大评估委员会规模降低传染，适合多智能体系统设计者参考
当大型语言模型作为多智能体系统中的评估者时，其策略偏好——无论是由显式提示还是共享架构先验引发——会通过智能体网络传播。我们提出传染网络框架，用于测量评估者偏好在交互式LLM智能体间的传播情况。在使用DeepSeek-chat的三智能体受控实验中，三种不同评估偏好配置（结构化、平衡、基于证据）下测得跨智能体传染矩阵Gamma_3，发现偏好始终传播（gamma ∈ [0.157, 0.352]）。中性提示对照实验揭示反直觉结果：共享架构先验主导显式提示作为传播驱动因素（rho_neutral = 1.498 vs. rho_mixed = 1.299；提示贡献率 -63.5%）。我们识别出由谱半径rho(Gamma_N)决定的三种传播模式，并验证链式拓扑抑制传染（beta_3 = 0.0126 ± 0.0038，95% CI [0.0089, 0.0163]，n=4种子），而全连接拓扑则引发级联效应（Delta H_avg = -0.020）——该拓扑依赖性转变在同构与跨模型智能体池中均得到验证（rho^cross = 1.296 ± 0.016，n=4）。我们证明将评估委员会规模从k=1增至k=3可使有效传染降低68.9% ± 14.1%（n=4种子），提供可操作的缓解策略。代码已开源。
原文摘要 · Abstract (English)
When large language models serve as evaluators in multi-agent systems, their strategy preferences -- whether induced by explicit prompts or by shared architectural priors -- propagate through the agent network. We introduce Contagion Networks, a formal framework for measuring how evaluator preferences spread across interacting LLM agents. In a controlled 3-agent experiment using DeepSeek-chat with three distinct evaluator preference profiles (structured, balanced, evidence-based), we measure the Cross-Agent Contagion Matrix Gamma_3 and find that preferences consistently propagate between agents (gamma in [0.157, 0.352]). A neutral-prompt control experiment reveals a counter-intuitive result: shared architectural priors dominate explicit preference prompts as the driver of contagion (rho_neutral = 1.498 vs. rho_mixed = 1.299; prompt contribution: -63.5%). We identify three propagation regimes governed by the spectral radius rho(Gamma_N) and demonstrate that the same agents suppress preference contagion in chain topology (beta_3 = 0.0126 +/- 0.0038, 95% CI [0.0089, 0.0163], n=4 seeds) but cascade in fully-connected topology (Delta H_avg = -0.020) -- a topology-dependent regime transition validated both for homogeneous and cross-model agent pools (rho^cross = 1.296 +/- 0.016, n=4). We show that increasing evaluator committee size from k=1 to k=3 reduces effective contagion by 68.9% +/- 14.1% (n=4 seeds), providing an actionable mitigation strategy. We release the open-source Contagion Network experimental framework.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。