A paper published this month in the Journal of the American Medical Association argues that autonomous AI will outperform both physicians and physician-AI hybrids on five fundamental medical tasks by 2030. The authors, led by Penn oncologist and medical ethicist Ezekiel Emanuel and venture capitalist Vinod Khosla, reviewed every study on AI in medicine published since January 1, 2024, and concluded that medicine is approaching a transition point where clinicians actively hurt outcomes. Their line, delivered without hedging: "humans in the loop degrade AI performance." The five tasks are the ones that define the job — taking histories, diagnosis, ordering tests, prescribing treatment, and managing chronic disease.
The prediction sits four years out, which is roughly the same interval that has passed since ChatGPT launched. Khosla has been pushing a version of this thesis since 2012, when he asked in a widely-read piece whether patients need doctors or algorithms, and he followed with a 101-page treatise in 2016 that projected AI handling 85% of what doctors do by 2035. The JAMA paper accelerates that timeline and moves it from op-ed territory into a peer-reviewed medical journal, co-signed by one of the most cited figures in American medical ethics.
Emanuel told Wired he had spent years dismissing Khosla's claim at conferences. His position shifted about a year ago after reading galleys of A Giant Leap by Robert Wachter, head of medicine at UCSF, which described a two-tier future in which first-class care would come from AI-doctor collaboration while most patients would rely on AI alone. Emanuel began asking a different question: if AI takes over, what's left for doctors to do.
Key facts
- 01A new JAMA paper by Ezekiel Emanuel and Vinod Khosla argues autonomous AI will surpass physicians and AI-hybrid care by 2030 on five core tasks.
- 02The five tasks: taking medical histories, diagnosis, ordering tests, prescribing treatment, and managing chronic disease.
- 03The paper reviewed AI-in-medicine literature published since January 1, 2024 and concludes 'humans in the loop degrade AI performance.'
- 04Khosla has predicted since 2012 that AI would handle 85% of what doctors do by 2035; he wrote a 101-page treatise on the thesis in 2016.
- 05A February 2026 Nature study found most patients could not effectively converse with LLMs to extract clinical expertise in real-life cases.
To answer it, he recruited Khosla, Khosla's son Neal — who runs Curai Health, an AI-driven online clinic staffed by doctors for prescribing and complex cases — and his own researcher. The team surveyed AI-in-medicine studies from January 2024 forward and concluded that a superior autonomous AI will likely surpass humans using AI by 2030. The authors call the prediction "unsettling but seems probable." Conflicts of interest, including Vinod Khosla's investments and Emanuel's grants and consultancies, are disclosed in the paper.
The American Medical Association is not persuaded. John Whyte, the AMA's CEO, told Wired that some of the studies cited are simulations rather than blind experiments and that they don't uniformly support the paper's conclusion. He pointed to a February 2026 Nature study that found most patients in real-life cases could not effectively converse with large language models to extract clinical expertise — a finding that undercuts the assumption that patients can drive an autonomous system without a clinician mediating.
Whyte's institutional position is that AI belongs inside a care plan governed by a physician, not outside one. Emanuel counters that in the four years since ChatGPT launched, model capability has moved faster than any credentialing body anticipated, and another four years of progress will change the comparison again.
“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”— John Whyte, CEO of the American Medical Association
Wachter, the UCSF chief singled out in the paper's opening paragraph for defending a hybrid model, conceded ground when Wired asked him. He called the argument important and acknowledged that the "AI plus humans is always better" line "can no longer be chiseled into a tablet." But he argued the unique attributes of trained physicians — delivering a grim prognosis, guiding a patient through treatment options, sitting with someone in a hard moment — are not replaceable by a model.
Wachter's framing is what he calls the doorman fallacy: the mistaken belief that automating one visible task eliminates the role. Doormen accept packages, feed pets, and hear out tenants; automatic doors did not put them out of work in white-glove buildings. He expects a version of that for physicians, where the diagnosis is deferred to AI but the human still performs functions that have value.
The subtler problem the paper surfaces is de-skilling. If new physicians defer to AI for history-taking and diagnosis during training, they never develop the underlying judgment that would let them override a bad model output. Whyte acknowledged that medical schools and residency programs are actively debating whether trainees should be allowed to use these tools, because a doctor who never learned to take a history has no baseline against which to check an AI's recommendation. That dynamic could accelerate the very takeover the paper predicts.
Vinod Khosla told Wired that surgery and emergency care will keep human clinicians in the short term. Neal Khosla expects regulators to eventually grant AI systems prescribing authority. Neither treats the timeline as speculative — Curai Health already operates on the premise that AI handles most of the clinical interaction and doctors handle the exceptions.
The JAMA paper matters less for its 2030 date, which is easy to argue about, than for who signed it. When a chair of medical ethics at Penn co-authors a piece in the flagship journal of American medicine arguing that clinicians degrade AI outcomes on the core tasks of doctoring, it moves the debate from venture-capital keynotes into the credentialing and liability conversations that actually shape care delivery. Autonomous-care startups now have a citation to hand regulators, insurers, and hospital systems. The AMA's counter — that AI belongs under a physician's care plan — is the incumbent position, and incumbency in medicine is durable. But the paper reframes the question the profession has to answer: not whether AI belongs in the exam room, but what a doctor is actually being paid to do once it is.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
