A debate that once sounded speculative is now moving into the center of medicine: whether AI should assist doctors, or whether it may eventually outperform them without a physician in the loop.
The latest flashpoint is an article in the Journal of the American Medical Association by authors including Ezekiel Emanuel and Vinod Khosla. Their position is direct: autonomous AI is rapidly approaching the point where it could provide better care than physicians or physician-AI teams in several fundamental areas of medical work.
The Claim Behind Autonomous AI in Medicine
The JAMA article asks, “Will Autonomous AI Exceed AI-Physicians as the Best Medical Care?” According to the source article, the question is rhetorical. The authors argue that AI alone can deliver the best outcomes in medicine.
Their case centers on five core medical tasks: taking medical histories, establishing a diagnosis, identifying what tests are needed, prescribing treatment, and managing chronic diseases. In plain terms, these are not side functions. They are central parts of what patients understand as doctoring.
The authors write that a “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 also argue that “humans in the loop degrade AI performance.”
That is the sharp edge of the debate. Many people in medicine have begun to accept a future in which doctors work with well-trained bots. This article goes further by suggesting that the best clinical result may come when doctors step back.
Why Ezekiel Emanuel Changed His Mind
Emanuel had not always accepted that premise. He told the source article’s author that for years he rejected the idea that AI could handle core medical functions. When Vinod Khosla argued in the 2010s that AI would be doing 85 percent of what doctors did by 2035, Emanuel’s response was blunt: “I said bullshit, there’s no way.”
Khosla had been making the argument for years. The source article notes his 2012 piece in TechCrunch asking, “Do We Need Doctors or Algorithms?” and a 101-page treatise in 2016 on the same broad idea.
The shift for Emanuel came after Robert Wachter sent him galleys of his book A Giant Leap. Wachter, head of medicine at UCSF, described a future in which high-end care would combine doctors and AI, while many others would largely receive AI-only care. Emanuel then began to reconsider whether Khosla might have been right.
That led to the central anxiety running through the discussion: “If AI takes over, what’s left for doctors to do?”
The Evidence Is Still Contested
To examine the premise, Emanuel asked Khosla to collaborate. Neal Khosla joined the effort, along with Emanuel’s researcher. Neal runs Curai Health, an online company that uses AI to treat patients, with doctors available to prescribe medicine and address more complicated cases. The source article notes that this conflict of interest is disclosed in the paper, along with Vinod Khosla’s relevant investments and Emanuel’s grants and consultancies.
The group reviewed studies on AI medicine from January 2024 to the present and concluded that, in many cases, a “superior autonomous AI” will likely surpass humans using AI by 2030. They describe that prediction as “unsettling but seems probable.”
John Whyte, CEO of the American Medical Association, strongly objects to removing physicians from the clinical loop. He points to limits in the evidence base, including that some studies are simulations rather than blind experiments and that not all of the surveyed work supports the article’s conclusion.
The source article also cites a February 2026 Nature article that found most patients in real-life cases were unable to converse effectively with large language models in order to access their expertise. For Whyte, that matters because a medical answer is only useful if patients can actually reach, understand, and act on it.
His position is that AI tools may have value, but within a physician-governed care plan. As he puts it: “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.”
What Doctors May Still Do Better
Wachter offers a more complicated view. He accepts that the paper raises an important argument. He says AI is already good, and that today AI and humans together are better. But he also warns that this assumption may not always hold: “There will be times when humans will muck up the performance.”
Still, Wachter argues that trained physicians bring human abilities that may remain vital. The source article points to moments such as delivering a grim prognosis or helping a patient choose a treatment path. These are not only information problems. They involve trust, timing, judgment, and the relationship between clinician and patient.
Wachter describes this through what he calls the doorman fallacy. The fear was that doormen would lose their purpose once doors could open automatically. Yet doormen continued to perform many other functions that residents valued.
His analogy suggests a possible future for physicians: they may accept an AI diagnosis while still performing other roles around care. He says, “I can imagine a world where a doctor accepts the AI diagnosis but also performs a lot of other functions that have value.”
The Training Problem
The hardest long-term issue may be what happens to medical expertise itself. If doctors rely heavily on AI for histories, diagnosis, tests, and treatment decisions, their own judgment could become less central over time.
The source article calls this process “de-skilling.” The concern is that new generations of doctors may not spend years building knowledge in the same way if an always-available AI system can supply answers on demand.
Whyte says medical schools and residency programs are already debating whether trainees should use these tools. His concern is practical: “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?”
At the same time, the source article notes the counterargument: ignoring such a powerful tool could itself be seen as malpractice. That tension will shape how medicine teaches future doctors, not just how it treats future patients.
Vinod Khosla says doctors will still be needed, at least in the short term, for surgery and emergency rooms. But he sees much of their expertise and judgment as expendable, except perhaps when doctors debate with AI to improve the systems. Neal Khosla expects that in coming years AI will receive regulatory permission to prescribe medicine.
That leaves medicine with a difficult choice. AI may become too capable to ignore, but the human role in care cannot be reduced to a simple checklist. The future of doctors may depend less on whether AI can answer medical questions and more on which parts of healing society still wants humans to own.