内容来源：https://www.wired.com/story/ai-has-human-doctors-asking-whats-left-for-us/

内容总结：

AI取代医生？医学界激辩：机器能否独挑大梁

本月《美国医学会杂志》刊发的一篇论文引发医学界震动。由知名医学伦理学家伊齐基尔·伊曼纽尔与风险投资家维诺德·科斯拉联合撰写的研究提出一个大胆论断：到2030年，人工智能在提供最佳医疗服务方面将全面超越人类医生，甚至胜过“人机协作”模式。

文章标题以问句形式呈现，但作者态度明确——答案是肯定的。研究团队回顾了2024年1月以来所有AI医学相关文献，认为AI在病史采集、诊断、检查建议、治疗方案制定及慢病管理五大核心医疗任务上即将实现全面超越。作者甚至警告，人类医生的介入反而会“降低AI表现”，建议医生不要干预AI诊疗过程。

这一观点令医学界哗然。作为论文第一作者，伊曼纽尔坦承自己曾是AI医疗的怀疑者，但近一年态度发生转变。他援引2026年2月《自然》杂志的研究指出，多数患者已能有效与AI对话获取专业建议，技术迭代速度远超预期。

美国医学会CEO约翰·怀特则持审慎态度，强调AI工具必须置于医生主导的诊疗方案中。“AI有潜力，但不能脱离临床监督。”他同时担忧过度依赖AI会导致医生“技能退化”——若年轻医生不再需要苦练问诊查体基本功，未来医学教育将面临根本性挑战。

加州大学旧金山分校医学系主任罗伯特·瓦赫特虽被认为是“AI经济舱模式”的反对者，但他也承认“人机协作优于纯AI”的结论已非铁板钉钉。“人类有时确实会拖累AI表现。”不过他认为医生的共情能力和复杂沟通技巧难以被机器替代，就像门卫虽不用手动开门，仍承担着收发快递、照看宠物等多元服务。

科斯拉父子则更为激进。维诺德·科斯拉提出，除手术和急诊外，医生的专业判断功能将大幅弱化。其子尼尔·科斯拉预测，未来几年监管机构将批准AI独立开具处方。这场辩论的深层追问或许正如比尔·盖茨所言：当AI接管专业领域，人类还剩什么价值？他建议为部分职业设定“人类专属”领域，但这在追求效率的资本逻辑面前，恐怕难以实现。

中文翻译：

别被本月《美国医学会杂志》上发表的一篇文章标题中的问号给骗了。当文章作者——包括医学界巨星伊齐基尔·伊曼纽尔和风险投资人维诺德·科斯拉——发出疑问：“自主人工智能会超越AI辅助医生，成为最佳医疗服务提供者吗？”时，他们不过是反诘罢了。他们的回答是斩钉截铁的“会的”。医学界的许多人正不情不愿地接受这样一个前提：患者能通过一种混合方式获得最佳治疗，即医生与训练有素的机器人协同会诊。但这篇文章的核心论点却是，仅靠人工智能就能提供最好的疗效。

“对自2024年1月1日以来发表的有关医学人工智能的所有文章的回顾显示，医学正迅速接近一个转折点，届时仅靠人工智能就将在提供最佳护理方面超越医生和‘医生+人工智能’的混合模式，这体现在五项基本医疗任务上，”他们写道。这些任务包括采集病史、做出诊断、确定所需检查、制定治疗方案以及管理慢性疾病。你懂的，就是行医本身。他们认为，现在的人工智能已经如此强大，临床医生应该避免干预，因为“人类介入反而会降低人工智能的性能”。靠边站吧，医生，让人工智能大显身手吧，这曾是你的活儿。

该文章的第一作者伊曼纽尔——他是阿里和拉姆的兄弟，宾夕法尼亚大学医学伦理与卫生政策系主任，也是一位著名的肿瘤学家——告诉我，多年来，他一直对人工智能能够执行核心医疗功能的说法嗤之以鼻。在2010年代，他经常在会议上碰到科斯拉，这位风险投资人总会在他耳边喋喋不休地讲，到2035年，人工智能将完成医生工作内容的85%。这是科斯拉多年来一直热衷的话题，他在2012年就在TechCrunch上发文问道：“我们需要医生还是算法？”（同样是反诘），并在2016年写了一篇101页的论文来阐述这一观点。“我当时说，胡说八道，绝不可能，”伊曼努尔说。“医生做的事情太复杂了。”

但大约一年前，伊曼纽尔的朋友罗伯特·瓦赫特把他的新书《巨大飞跃》的样书寄给了他。瓦赫特是加州大学旧金山分校的医学负责人，他在书中描绘了一个世界：一流的医疗将是人工智能和医生之间的协作过程，而“经济舱”的大众则大多只能靠纯人工智能凑合。伊曼纽尔开始觉得，也许科斯拉是对的。他突然闪过一个令人不安的问题：“如果人工智能接管了一切，医生还能干什么？”

他首先需要验证这个前提。他邀请科斯拉合作，这位风险投资人建议他的儿子尼尔·科斯拉也加入这项努力，伊曼纽尔的研究人员也参与其中。尼尔经营着Curai Health，这是一家利用人工智能治疗患者的在线公司，并配备医生团队来开具处方和处理更复杂的病例。（这一利益冲突在论文中有所披露，包括维诺德·科斯拉的相关投资以及伊曼纽尔的各种资助和咨询职务。）合著者们开始梳理他们能找到的从2024年1月至今的所有关于人工智能医学的研究，并得出结论：在许多情况下，“卓越的自主人工智能”很可能在2030年之前超越使用人工智能的人类。他们承认，这一预测“令人不安，但似乎是可能的”。

“令人不安”这个说法都算轻了。反对声音中最大的一方是美国医学会首席执行官约翰·怀特，该协会代表着所有那些将担心自己还能干什么的医生。我和他交谈时，怀特首先指出了他认为这篇论文的不足之处，包括其中调查的一些研究是模拟实验，而非盲法试验，而且并非所有研究都支持文章的结论。值得注意的是，2026年2月发表在《自然》杂志上的一项研究发现，在真实病例中，大多数患者无法有效地与大型语言模型对话，从而无法利用其专业知识。不出所料，怀特对将人类完全赶出临床环节的想法感到愤怒。“美国医学会确实看到了这些工具的潜力，但它们必须在由医生主导的护理计划框架内使用，”他说。

伊曼纽尔反驳说，他和合著者在得出结论时是谨慎的。“距离ChatGPT问世还不到四年，”他对我说。“我们是在预测四年后的情况。你难道不觉得（人工智能）会比医生先进得多吗？”

为了寻求第二意见，我咨询了瓦赫特。在这篇论文的第一段中，他被特别点名，作为错误地认为纯人工智能治疗是“经济舱”服务的代表。（我本人认识瓦赫特。）令人惊讶的是，瓦赫特承认论文作者们的观点有道理。“他们提出的论点很重要，”他说。他说，人工智能很强大，目前人工智能和人类结合会更好。“但这不能再被奉为金科玉律了。有时候人类反而会搞砸事情。”

然而，瓦赫特坚持认为，训练有素的医生所具备的独特人文特质至关重要、不可替代。在传达严峻预后的坏消息方面，人工智能可能永远无法像人类那样有效，而患者也更有可能被一位现实中的临床医生引导至最佳治疗路径。

瓦赫特引用了所谓的“门卫谬误”，这个名称源于一种恐惧：公寓楼的开门门卫会担心，一旦门可以自动打开，他们就会失业。实际上，门卫承担着各种各样的任务，在提供高端服务的公寓楼里仍然不可或缺，比如接收亚马逊快递、喂宠物，有时还会倾听住户的抱怨。“我们会在医生身上看到门卫谬误的翻版，”瓦赫特说。“我可以想象这样一个世界：医生接受人工智能的诊断，同时还履行许多其他有价值的职能。”

然而，这里被取代的，不是像开门那么简单的东西，而是经年累月训练得来的、来之不易的专业技能。你知道《皮特医生》里那些场景吗？罗比医生总是逼问住院医生，要求他们当场给出判断？当人工智能随时都能给出答案时，这种场景就不必要了。医生正走上一条过度依赖人工智能的道路，以至于他们自身的判断和知识将变得不那么重要。这种“去技能化”过程可能会加速人工智能对医疗领域的接管，尤其是当新一代医生认为没有必要花多年时间学习那些只需向白大褂口袋里那个随时在线的AI专家提问就能获取的知识时。美国医学会的怀特承认，这是他担心的问题。“很多医学院和住院医师培训项目都在争论，受训中的医生是否应该使用这些工具，因为如果你从未学过如何采集病史和进行体格检查，你怎么可能真正学会？”另一方面，也有人认为，忽视这样一个强大的工具本身就是一种失职。

那么，医生的未来到底在哪里？维诺德·科斯拉说，至少在短期内，外科手术和急诊室还需要医生。但就提供专业知识和判断而言，他说他们基本上是可以被取代的，也许除了与人工智能辩论以改进这些系统之外。尼尔·科斯拉告诉我，这种转变已经在发生，他预计未来几年内，我们将给予人工智能开具处方的监管许可。

还记得电视剧《豪斯医生》吗？那个脾气暴躁、粗鲁无礼的诊断天才，一边解码疑难杂症，一边侮辱病人、让倒霉的学徒们难堪。在人工智能医学的世界里，豪斯医生式的人物将无处容身，因为人工智能将常规地、自如地将其庞大的知识库应用于最罕见的疾病。人类临床医生的价值将体现在豪斯最缺乏的那些方面，比如同理心和高情商的沟通。当然，那些从事动手操作的人不会受到人工智能的侵扰——除非机器人也学会了那些。

医疗保健是我们生活中非常重要的一部分，但这场辩论远不止于该领域。“医生还能干什么？”这个问题，很可能就是“人类还能干什么？”。比尔·盖茨在他最近关于人工智能的悲观论调中建议，社会可以将某些任务——比如医生为患者提供如何应对人工智能预后的咨询——“划归人类专属”，以避免经济崩溃和民怨沸腾。（试着把这个解决方案兜售给那些质疑搞些虚职岗位有何意义的公司股东们吧。）至少，失业大众的痛苦会因一个额外的好处而得到缓解：第一流的医疗建议触手可及。

这是史蒂文·利维“后通道”时事通讯的一期。点击此处阅读往期通讯。

评论
返回顶部

英文来源：

Don’t be fooled by the question mark in the title of an article published this month in the Journal of the American Medical Association. When the authors, including medical superstar Ezekiel Emanuel and venture capitalist Vinod Khosla, asked, “Will Autonomous AI Exceed AI-Physicians as the Best Medical Care?” they were being rhetorical. Their answer is an emphatic YES. Many in the medical world are begrudgingly coming to terms with the premise that patients can get the best treatment by a hybrid approach where doctors work in consultation with well-trained bots. But this article’s thesis is that AI alone can deliver the best outcomes.
“Review of all published articles on AI in medicine since January 1, 2024, shows that medicine is rapidly approaching the transition point at which AI alone will exceed physicians and physician AI-hybrids in providing the best care at five fundamental medical tasks,” they write. Those tasks include taking medical histories, establishing a diagnosis, identifying what tests are needed, prescribing treatment, and managing chronic diseases. You know, doctoring. AI is now so good, they argue, that clinicians should refrain from meddling, because “humans in the loop degrade AI performance.” Stand back, doc, and let AI do its thing. Which used to be your thing.
Lead author Emanuel—brother of Ari and Rahm, chair of the Department of Medical Ethics and Health Policy at Penn and a renowned oncologist—told me that for many years, he’d dismissed the idea that AI could perform core medical functions. He would run into Khosla at conferences in the 2010s, and the VC would bend his ear about how AI would be doing 85 percent of what doctors did by 2035. It was a hobbyhorse Khosla had been riding for years, writing a 2012 piece in TechCrunch asking, “Do We Need Doctors or Algorithms?”(again, rhetorical) and penning a 101-page treatise in 2016 to that effect. “I said bullshit, there’s no way,” says Emanuel. “What doctors do is too complicated.”
But about a year ago, Emanuel’s friend Robert Wachter sent him galleys of his book A Giant Leap, where Wachter, who is head of medicine at UCSF, outlined a world where first-class medicine would be a collaborative process between AI and doctors, but the masses in “economy class” would mostly make do with AI alone. Emanuel began to think that Khosla might be right. He suddenly flashed on a frightening question: “If AI takes over, what’s left for doctors to do?”
First he needed to validate the premise. He asked Khosla to collaborate, and the VC suggested that his son Neal Khosla join the effort, along with Emanuel’s researcher. Neal runs Curai Health, an online company that uses AI to treat patients, with a staff of doctors to prescribe medicine and handle more complicated issues. (This conflict of interest is disclosed in the paper, along with Vinod Khosla’s relevant investments and Emanuel’s various grants and consultancies.) The coauthors set out to look at all the studies on AI medicine that they could find from January 2024 to the present and concluded that, in many cases, a “superior autonomous AI” will likely surpass humans using AI by 2030. That prediction, they admit, is “unsettling but seems probable.”
Unsettling is an understatement. One of the loudest dissenting voices is John Whyte, CEO of the American Medical Association, which represents all those doctors who will be worrying about what’s left. When I talked to him, Whyte began by noting what he saw as the paper’s shortcomings, including that some of the studies it surveyed are simulations, not blind experiments, and they don’t all support the article’s conclusion. Notably, a February 2026 Nature article found that in real-life cases, most patients were unable to effectively converse with large language models to access their expertise. Whyte unsurprisingly bristled at the idea of removing humans from the clinical loop. “The AMA does see the potential in these tools, but they have to be utilized in the context of a care plan that’s governed by a physician,” he says.
Emanuel countered that he and his coauthors were careful in their conclusions. “It’s been less than four years since ChatGPT was introduced,” he said to me. “We’re making a prediction for four years from now. You don’t think that [AI is] going to be way more advanced than a doctor?”
Seeking a second opinion, I consulted Wachter. He’s singled out in the paper’s first paragraph as the guy who wrongly argues that AI-only treatment will be “economy class.” (I know Wachter personally.) Surprisingly, Wachter concedes that the authors of the paper have a point. “The argument they’re making is important,” he says. AI is good, he says, and right now AI and humans together are better. “But that can no longer be chiseled into a tablet. There will be times when humans will muck up the performance.”
Wachter is insistent, however, that the unique human attributes of trained physicians are vital and irreplaceable. AI may never be as effective in breaking the news of a grim prognosis than a human, and a patient may well be more likely to be guided to the best path of treatment by a real-life clinician.
Wachter invokes what he calls the doorman fallacy, named after the fear that people who opened doors for tenants in apartment buildings would be out of work once doors could open automatically. In fact doormen perform myriad tasks and are still an invaluable presence in white-glove buildings. Like accepting Amazon packages, feeding pets, and sometimes lending a sympathetic ear to a tenant sharing a gripe. “We’re going to find a version of the doorman fallacy for doctors,” says Wachter. “I can imagine a world where a doctor accepts the AI diagnosis but also performs a lot of other functions that have value.”
What’s being displaced here, though, isn’t something as straightforward as opening a door. It's the hard-won expertise that comes from years of training. You know those scenes in The Pitt where Dr. Robby relentlessly puts his residents on the spot by asking them to come up with instant assessments? That won’t be necessary when AI has the answer on demand. Physicians are on a path to relying so much on AI that their own judgment and knowledge will become less relevant. This “de-skilling” process could accelerate the takeover of medicine by AI, especially as new generations of doctors won’t find it necessary to spend years learning stuff that they can find out by querying an always-on AI expert in the pocket of their lab coat. The AMA’s Whyte admits that this is a concern for him. “A lot of medical schools and residency programs are debating whether physicians in their training can utilize these tools, because if you’ve never learned how to take medical history and do a physical, how will you ever learn?” On the other hand, there’s an argument that ignoring such a compelling tool is a form of malpractice.
So what’s the future for doctors? Vinod Khosla says that we’ll need them, at least in the short term, for surgery and in the emergency room. But in terms of providing expertise and judgment, he says they’re largely expendable, except maybe as debaters with AI to improve those systems. Neal Khosla tells me that the shift is already underway and that he expects that in coming years we will give AI the regulatory permission to prescribe medicine.
Remember the TV series House, where a dyspeptic and rude diagnostic genius decoded baffling conditions while insulting patients and humiliating his hapless acolytes? In the world of AI medicine, a Dr. House would be rendered homeless, because AI will routinely and freely apply its vast knowledge base to the most obscure ailments. The value of human clinicians will be the elements where House was most lacking, like empathy and high-EQ communication. Of course those who perform hands-on procedures are safe from AI encroachment—until the robots figure that out.
Health care is a huge part of our lives, but the debate here goes beyond that sector. The question, “What’s left for doctors to do?” may well be “What’s left for humans to do?” Bill Gates suggests in his recent jeremiad on AI that society might designate some tasks—like doctors counseling patients on how to deal with AI’s prognoses—“human-reserved,” to stave off economic ruin and a run on pitchforks. (Try to sell that solution to corporate shareholders who question the funding of make-work jobs.) At least the misery of the unemployed masses will be mitigated by one bonus: first-class medical advice that’s one click away.
This is an edition of Steven Levy’s Backchannel newsletter. Read previous newsletters here.
Comments
Back to top