挑战感知知识招募与元智能体监督用于医疗多智能体推理
原标题：Challenge-Aware Knowledge Recruitment with Meta-Agent Supervision for Medical Multi-Agent Reasoning
AI 摘要
该论文提出一种“挑战感知的知识招募”方法，用于医疗多智能体推理：由外部元智能体 Challenger 检查推理轨迹并识别具体知识缺口，再触发针对性检索或中途招募专科智能体。在 MedAgentsBench 困难子集（862 题、9 个生物医学数据集）上，挑战感知的中途检索在 8/9 数据集上提升准确率，比仅用 Challenger 高 6 个百分点；Challenger 触发的专科招募将 GPT-4.1-mini 在 MedQA 上的准确率从 46% 提升到 54%。研究还发现，简单叠加检索与专科招募反而会降低性能，说明需协调不同自适应干预通道。
正文节选
Managing Agents that Manage Agents: Workshop on Responsible Use of Meta-Agents Challenge-Aware Knowledge Recruitment with Meta-Agent Supervision for Medical Multi-Agent Reasoning 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 fr