Challenge-Aware Knowledge Recruitment with Meta-Agent Supervision for Medical Multi-Agent Reasoning

Computer Science > Artificial Intelligence arXiv:2611.05393 (cs) [Submitted on 14 Sep 2026] Title: Challenge-Aware Knowledge Recruitment with Meta-Agent Supervision for Medical Multi-Agent Reasoning Authors: Riccardo Riccio, Haitz Sáez de Ocáriz Borde, Pietro Liò.

计算机科学 > 人工智能 arXiv:2611.05393 (cs) [提交于 2026 年 9 月 14 日] 标题：用于医学多智能体推理的挑战感知知识招募与元智能体监督 作者：Riccardo Riccio, Haitz Sáez de Ocáriz Borde, Pietro Liò。

Abstract: Large language models perform strongly on standard medical QA but degrade on harder reasoning tasks, where multi-agent debate often fails to beat strong single-model prompting. We argue that a key limitation is that existing systems select experts and retrieve knowledge from the question alone, rather than from gaps that emerge during reasoning.

摘要：大型语言模型在标准医学问答任务中表现出色，但在更难的推理任务中性能会下降，此时多智能体辩论往往无法超越强大的单模型提示。我们认为，一个关键的局限性在于现有系统仅根据问题本身来选择专家和检索知识，而不是根据推理过程中出现的知识缺口。

We propose challenge-aware adaptive recruitment: an external CHALLENGER inspects the evolving reasoning trace and identifies specific factual or expertise gaps. These signals trigger targeted knowledge retrieval or mid-discussion specialist recruitment.

我们提出了挑战感知自适应招募：一个外部“挑战者”（CHALLENGER）会检查不断演变的推理轨迹，并识别特定的事实或专业知识缺口。这些信号会触发有针对性的知识检索或讨论过程中的专家招募。

On the MEDAGENTSBENCH hard subset (862 questions across nine biomedical datasets), upfront retrieval provides little benefit, while challenge-aware in-between retrieval performs better; on a multi-specialist backbone it improves accuracy on 8 of 9 datasets and adds 6 percentage points over Challenger Only.

在 MEDAGENTSBENCH 的困难子集（涵盖九个生物医学数据集的 862 个问题）上，预先检索几乎没有益处，而挑战感知的中途检索表现更好；在多专家骨干模型上，它在 9 个数据集中的 8 个上提高了准确率，并比仅使用“挑战者”的方法提升了 6 个百分点。

Ablations indicate benefits from both retrieval timing and gap-specific query signals, and a live web-search experiment shows that the timing advantage persists across retrieval backends. For specialist recruitment, self-triggered adaptation is unreliable, whereas Challenger-triggered recruitment raises GPT-4.1-mini MedQA accuracy from 46% to 54%, with positive direction on all tested generalization datasets.

消融实验表明，检索时机和针对缺口的查询信号均有益处，且实时网络搜索实验显示，这种时机优势在不同的检索后端中依然存在。在专家招募方面，自触发式自适应是不可靠的，而由“挑战者”触发的招募将 GPT-4.1-mini 在 MedQA 上的准确率从 46% 提高到了 54%，并在所有测试的泛化数据集上表现出积极趋势。

Naively combining retrieval and specialist recruitment degrades performance, highlighting the need to coordinate adaptive intervention channels. These results show that when and why external resources are recruited is a key design choice in multi-agent medical reasoning.

简单地结合检索和专家招募会降低性能，这凸显了协调自适应干预渠道的必要性。这些结果表明，何时以及为何招募外部资源是多智能体医学推理中的一个关键设计选择。