大语言模型智能体是否进行理性协商？面向A2A/MCP的可验证多智能体交互机制设计框架
Do LLM Agents Negotiate Rationally? A Mechanism-Design Framework for Verifiable Multi-Agent Interaction over A2A/MCP
浏览论文内容
中文总结 AI 辅助
本研究针对A2A/MCP协议提出可验证多智能体交互机制设计框架，实验发现机制激励兼容性无法自动迁移至大语言模型智能体行为，相关成果连接经典多智能体理论与现代LLM智能体基础设施。
中文摘要 AI 辅助
现代大语言模型智能体框架日益通过各类标准实现互操作，包括Anthropic的模型上下文协议（MCP）用于智能体对工具的访问，以及Google的智能体对智能体（A2A）协议用于智能体的委托与协商。然而，这些协议仅规定了传输与发现，未涉及策略正确性，也无法保证高效、个体理性或防策略的结果。本文提出一个框架，包含三部分：其一，将经典协商机制（包括交替报价讨价还价与维克里-克拉克-格罗夫斯式拍卖）编码为A2A消息模式的约束；其二，提供轻量级运行时验证与修复层，用于检查消息是否符合协议不变量；其三，构建协商与分配任务基准，该基准带有已知最优解，可用于衡量与博弈论预测的偏差。我们针对多种大语言模型主干，分别测试非结构化对话、结构化协议、带验证的结构化协议三种条件，每个条件的协商试验数量为30次。结果显示，验证可降低结果方差，结构化协议在两种模型中均实现100%成功率；修正解析器错误后，经审计的非结构化基线分别实现约97%与93.3%的成功率。在拍卖实验中，每个模型的试验数量为30次，两种模型均实现100%的有效分配，但在真实出价方面差异显著：一种模型在所有试验中均出价其真实估值，而另一种模型仅在3.3%的试验中如此。由此可见，机制层面的激励兼容性无法自动迁移至大语言模型智能体的行为。三方公平分配任务仅产生4.2%的可用结果，我们报告该负面结果并给出诊断。本研究连接经典多智能体系统理论与现代大语言模型智能体基础设施，在A2A协议层定义了可验证的交互。
英文摘要
Modern LLM-agent frameworks increasingly interoperate through standards such as Anthropic's Model Context Protocol (MCP) for agent-to-tool access and Google's Agent2Agent (A2A) protocol for agent delegation and negotiation. However, these protocols specify transport and discovery rather than strategic correctness and do not guarantee efficient, individually rational, or strategy-proof outcomes. We introduce a framework that (i) encodes classical negotiation mechanisms, including alternating-offers bargaining and Vickrey-Clarke-Groves-style auctions, as constraints over A2A message schemas; (ii) provides a lightweight runtime verification and repair layer that checks messages against protocol invariants; and (iii) offers a benchmark of negotiation and allocation tasks with known optimal solutions for measuring deviations from game-theoretic predictions. We evaluate multiple LLM backbones using unstructured dialogue, structured protocols, and structured protocols with verification. Across negotiation trials (N=30 per condition), verification reduces outcome variance, while structured protocols achieve 100 percent success for both models. After correcting parser artifacts, audited unstructured baselines achieve approximately 97 percent and 93.3 percent success. In auction experiments (N=30 per model), both models achieve 100 percent efficient allocation but differ sharply in truthful bidding: one bids its exact valuation in every trial, whereas the other does so in only 3.3 percent of trials. Thus, mechanism-level incentive compatibility does not automatically transfer to LLM-agent behavior. A three-party fair-allocation task produced only 4.2 percent usable outcomes; we report this negative result with a diagnosis. This work bridges classical multi-agent systems theory and modern LLM-agent infrastructure and defines verifiable interaction at the A2A protocol layer.
