临床医生兼 AI 研究员 Jonathan Chen 博士探讨 AI 如何支持医学中的人类判断力，与魔术进行类比，并分享真实临床实践中的见解。
Dr. Jonathan Chen, a clinician and AI researcher, discusses how AI can support human judgment in medicine, drawing parallels with magic and sharing insights from real clinical practice.
要点 · TL;DR
医疗 AI 需要人类判断来避免幻觉并确保安全。 AI in medicine requires human judgment to avoid hallucinations and ensure safety.
医生使用 AI 可能因锚定和谄媚而表现不如单独 AI。 Doctors using AI may underperform AI alone due to anchoring and sycophancy.
检索增强生成将 AI 回答锚定在来源中，提高可靠性。 Retrieval augmented generation grounds AI responses in sources, improving reliability.
核心观点 · Key points
AI 模型可能产生幻觉和虚构，因此人类判断比以往任何时候都更重要。 AI models can hallucinate and confabulate, so human judgment is more critical than ever.
医生使用 AI 工具时可能因锚定和谄媚效应而表现不如单独使用 AI。 Doctors using AI tools may underperform AI alone due to anchoring and sycophancy.
检索增强生成将 AI 回答锚定在来源上，提高了可靠性。 Retrieval augmented generation grounds AI responses in sources, improving reliability.
医学中的 AI 必须优先考虑安全性而非准确性，因为作为和不作为都可能造成伤害。 AI in medicine must prioritize safety over accuracy, as harm can occur from both commission and omission.
AI 可通过总结病历革新临床工作流程，节省时间并捕捉遗漏细节。 AI can revolutionize clinical workflow by summarizing patient records, saving time and catching missed details.
反共识 · Contrarian takes
AI 工具可以成为最好的老师，但用它来逃避困难会破坏学习。 AI tools can be the best teacher, but using them to avoid struggle undermines learning.
默认禁止在作业中使用 AI 的政策无法执行，并将学生置于道德困境。 Default policies disallowing AI on homework are unenforceable and put students in ethical dilemmas.
AI 生成的文本可能使写作风格同质化，但也可能让我们专注于更深层的思考。 AI-generated text may homogenize writing styles, but could also free us to focus on deeper thinking.
魔术和 AI 都挑战对现实的感知，需要判断力来区分真假。 Magic tricks and AI both challenge perception of reality, requiring judgment to distinguish truth.
AI 在医学中的最大风险可能是遗漏——未能推荐重要的行动。 The biggest risk from AI in medicine may be omission—failing to recommend important actions.
本期章节 · Chapters（共 10）
引言与背景Introduction and Background
申请MD-PhD的决定Decision to Apply for MD-PhD
早期研究的惊人结果Surprising results from early study
AI中的判断与安全Judgment and Safety in AI
RAG与基于事实的回应RAG and Grounded Responses
直接链接来源与医疗决策Direct Links to Sources and Medical Decision-Making
AI捕捉人类遗漏AI catching what humans miss
医学教育中的AI：作业困境AI in Medical Education: Homework Dilemma