JAMA paper argues autonomous AI will beat doctors by 2030

An article published this month in the Journal of the American Medical Association, coauthored by oncologist and bioethicist Ezekiel Emanuel and venture capitalist Vinod Khosla, argues that autonomous AI alone, not a doctor working alongside AI, will soon deliver the best medical care. Reviewing published studies on AI in medicine since January 1, 2024, the authors write that medicine is 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": taking medical histories, establishing a diagnosis, identifying needed tests, prescribing treatment and managing chronic disease. Their conclusion is blunt: "humans in the loop degrade AI performance."

Emanuel had long dismissed Khosla's version of this argument. Khosla, who wrote a 2012 TechCrunch piece asking "Do We Need Doctors or Algorithms?" and a 101-page treatise on the topic in 2016, used to tell Emanuel that AI would be doing 85 percent of what doctors did by 2035; Emanuel's reply for years was "bullshit, there's no way, what doctors do is too complicated." He began to reconsider about a year ago after his friend Robert Wachter, head of medicine at UCSF, sent him galleys of his book A Giant Leap, which describes a future where AI-doctor collaboration remains "first class" medicine while an AI-only "economy class" serves the masses. That framing left Emanuel with a new worry: "If AI takes over, what's left for doctors to do?"

To test the idea, Emanuel asked Khosla to collaborate; Khosla brought in his son Neal, who runs Curai Health, an online service that uses AI to treat patients with a staff of doctors handling prescriptions and complex cases. The paper discloses the resulting conflicts of interest, including Vinod Khosla's relevant investments and Emanuel's grants and consultancies. The coauthors conclude that a "superior autonomous AI" will likely surpass humans-using-AI by 2030, a prediction they call "unsettling but seems probable."

John Whyte, CEO of the American Medical Association, is one of the most vocal critics. He points out that some of the studies the paper surveys are simulations rather than blind experiments and do not all support its conclusion, and cites a February 2026 Nature article finding that in real-life use, most patients could not effectively converse with large language models to get useful expertise from them. "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," Whyte says. Emanuel counters that the timeline itself argues for humility about doubting the pace: "It's been less than four years since ChatGPT was introduced. 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?"

Wachter, despite being singled out in the paper's opening paragraph as the one who wrongly expects AI-only care to stay second class, partly agrees. "The argument they're making is important," he says, adding that AI-and-human teams beat either alone today, "but that can no longer be chiseled into a tablet. There will be times when humans will muck up the performance." He still argues that some human functions, breaking bad news, guiding a patient through treatment choices, will stay valuable even once AI handles diagnosis, invoking what he calls the "doorman fallacy": doormen kept their jobs after automatic doors arrived by doing everything else a doorman does. "I can imagine a world where a doctor accepts the AI diagnosis but also performs a lot of other functions that have value," he says.

A further worry raised in the piece is de-skilling: if AI increasingly takes medical histories, diagnoses and prescribes, newer doctors may stop building the underlying judgment those tasks exist to teach. Whyte confirms the debate is already happening: "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?" The counterargument, also raised in the piece, is that refusing to use such a capable tool could itself be a form of malpractice.

On what remains for doctors, Vinod Khosla says they will still be needed in the short term for surgery and emergency-room work, but calls their expertise and judgment "largely expendable" outside that, except perhaps as debate partners to help improve the AI systems themselves. Neal Khosla says the shift is already underway and expects regulators to eventually grant AI permission to prescribe medicine. Bill Gates, in a separate recent essay on AI, has suggested some tasks, such as doctors counseling patients on how to interpret AI's prognoses, could be designated "human-reserved" to soften the economic disruption.

Key facts

A JAMA paper by Ezekiel Emanuel, Vinod Khosla and coauthors argues autonomous AI alone, without a doctor's oversight, will soon deliver better care than doctors or doctor-AI hybrids.
Reviewing AI-in-medicine studies published since January 2024, the paper says AI is approaching a transition point at five core tasks: taking histories, diagnosing, ordering tests, prescribing treatment and managing chronic disease.
The authors predict a "superior autonomous AI" will likely surpass humans-using-AI by 2030, calling that prediction "unsettling but seems probable."
AMA CEO John Whyte pushes back, noting some reviewed studies are simulations rather than blind trials and citing a February 2026 Nature article showing most patients could not effectively query large language models on their own.
UCSF's Robert Wachter, named in the paper's opening for expecting AI-only care to stay "economy class," partly concedes the argument but says human skills like delivering hard news remain valuable regardless.

Why it matters

A JAMA paper coauthored by a prominent bioethicist and a venture capitalist puts a specific, citable claim into print: autonomous AI, not physician oversight, will soon deliver the best medical care. That moves the debate past the settled compromise of doctor-AI collaboration and attaches a concrete date, 2030, to the more radical position that doctors become optional for diagnosis, testing, treatment and chronic disease management.

Who it affects

Practicing physicians, whose authority over five specific clinical tasks is being challenged directly; the American Medical Association, which now has a paper on record that its own CEO has to argue against; medical schools and residency programs currently debating whether trainees should be allowed to use these tools while still learning core skills; and patients, especially the wider group Wachter's book expects to receive AI-only rather than AI-plus-doctor care.

How to use it

The AMA's stated position gives clinicians and administrators a concrete line to hold for now: treat AI as a tool operating inside a physician-governed care plan, not a replacement for one, until claims like the paper's five-task threshold are tested in blind trials rather than simulations. Medical educators weighing whether to let trainees use these tools can point to the paper's own reliance on simulation data as a reason to keep hands-on history-taking and physical-exam training mandatory rather than optional.

How solid is it

The paper is a literature review of AI-in-medicine research from January 2024 onward, not new experimental data, and by the AMA CEO's own account some of the studies it surveys are simulations rather than blind trials and do not all support its conclusion. The author list carries disclosed conflicts of interest: Neal Khosla runs an AI-based online clinic, Vinod Khosla holds relevant investments, and Emanuel discloses grants and consultancies. A separate February 2026 Nature study, cited by Whyte, found that most patients could not effectively converse with large language models to extract useful expertise, a direct complication for the paper's autonomous-AI-alone scenario.

Risks and caveats

The authors and their critics agree on a de-skilling risk: if AI takes over medical histories, diagnosis and prescribing, newer doctors may never build the judgment those tasks were meant to teach, and Whyte says medical schools are already debating whether to restrict tool access during training for that reason. Several of the paper's more concrete follow-on predictions remain unresolved by their own proponents' admission: Neal Khosla expects regulators to permit AI prescribing only in unspecified "coming years," and Bill Gates's proposal to designate some tasks "human-reserved" carries no stated funding mechanism. The paper's central 2030 timeline is itself framed by its authors as "unsettling but seems probable," a forecast rather than a settled fact.

“Humans in the loop degrade AI performance.”

— the JAMA paper's authors