A doctor spends twelve minutes with you, glances at your chart, and gives you a diagnosis. Now imagine an AI system doing the same thing — except it gets the answer right nearly four times more often. That is not a hypothetical anymore. It is the central claim of a new JAMA paper that has turned into one of the most contentious debates in American medicine this year, and it is why the phrase "AI beats doctors" is suddenly everywhere.
The paper comes from an unusual pairing: a world-renowned bioethicist who helped architect Obama-era healthcare policy, and a billionaire venture capital family with a financial stake in the very future they are predicting. Their argument is blunt — autonomous AI, working completely alone, is likely to soon outperform both unaided physicians and physicians who are using AI as a helper tool. Not eventually. Soon.
The American Medical Association is not happy. Physicians across the country are pushing back publicly. And a diagnostic accuracy gap of 60% versus roughly 16% has become the number everyone in health tech is arguing about. Here is everything you need to know about the study, the backlash, and what it actually means for the future of your next doctor's visit.
- What the JAMA Study Actually Claims?
- Who Wrote It - And Why Their Backgrounds Matter?
- The 5 Medical Tasks Where AI Is Already Winning
- The Chess Analogy Driving the Entire Argument
- Why Doctors Are Furious?
- The Conflict-of-Interest Problem Nobody Can Ignore
- What This Means for Patients Right Now
- The Bigger Picture: AI and the "Cognitive Professions"
- Should You Trust an AI Diagnosis Today?
- What Happens Next - Licensing, Liability, and 2030?
- Frequently Asked Questions
What the JAMA Study Actually Claims?
In mid-August 2026, the Journal of the American Medical Association published a Viewpoint article carrying a title that reads more like a challenge than a research summary: "Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care?" It was not a new clinical trial. It was a review and argument piece, built on evidence the authors pulled from studies published since January 2024 comparing autonomous AI against physicians, both with and without AI assistance.
Their conclusion was blunt. The authors argue that AI operating entirely on its own — with zero human oversight — is already matching or beating physicians at several core cognitive tasks of medicine, and that adding a doctor back into the loop can actually make outcomes worse once the AI is clearly the stronger performer.
The number driving most of the headlines comes from a comparison involving OpenAI's o3 reasoning model, first detailed in Becker's Hospital Review's breakdown of the JAMA Viewpoint. In a set of 377 complex, real-world diagnostic cases, o3 listed the correct final diagnosis as its top answer 60% of the time. Twenty internal medicine physicians, working through a 302-case subset of the same material, landed on the correct top diagnosis roughly 16% of the time. That is not a marginal edge. That is the kind of gap that ends debates in a boardroom — which is exactly why it has started one in the medical community instead.
| Study / Comparison | AI Performance | Physician Performance |
|---|---|---|
| Complex diagnosis accuracy (ChatGPT o3 vs. 20 internists, 377 cases) | 60% correct first diagnosis | ~16% correct first diagnosis |
| Patient complaint elicitation (Google's AMIE, 159 simulated cases) | 97% rated significantly better | 50% |
| Independent clinical judgement study cited by Dr. Robert Wachter | 90% accurate acting alone | 74% alone / 76% with AI assistance |
Notice the pattern in that last row — physicians working with AI barely improved on their solo performance, while AI working alone pulled far ahead of both. That single data point is essentially the entire thesis of the JAMA paper compressed into one table.
Who Wrote It - And Why Their Backgrounds Matter?
Understanding this story means understanding exactly who is making the argument, because it shapes how the paper has been received.
Ezekiel Emanuel: The Bioethicist Heavyweight
Lead author Ezekiel Emanuel is not a fringe voice. He is a University of Pennsylvania bioethicist, a former White House health policy adviser, and one of the architects behind Obama-era healthcare reform — a background Chief Healthcare Executive explored in a follow-up interview with Emanuel. When someone with that resume publishes a paper arguing doctors should step back and let AI run the show, medical institutions listen — even the ones who disagree with him.
Vinod and Neal Khosla: The Billionaire VC Family With Skin in the Game
The paper's other authors are where things get complicated. Neal Khosla is the CEO of Curai Health, an AI-powered virtual primary care company. His father, Vinod Khosla, is the founder of Khosla Ventures, an early investor in OpenAI, and also an investor in Curai Health itself, as Forbes' reporting on the paper's authors lays out in detail. In other words, two of the four authors stand to financially benefit if the argument they are making becomes the new standard of care.
Vinod Khosla did not soften his position when the paper dropped. He posted on X, asking bluntly whether the "game" was "mostly over for human doctors vs. AI" — a nine-word line that ended up quoted across nearly every outlet that covered the story.
The 5 Medical Tasks Where AI Is Already Winning
The JAMA authors did not make a vague claim about "AI getting better." They pointed to five specific, narrow, cognitive functions of medicine where they say the evidence already leans toward autonomous AI:
- Gathering patient information - taking a history, asking follow-up questions, and identifying symptoms without missing details a rushed human clinician might skip.
- Forming a differential diagnosis - generating and ranking the list of possible conditions that could explain a patient's symptoms.
- Ordering diagnostic tests - selecting the most cost-efficient and clinically relevant tests rather than over-testing or under-testing.
- Prescribing guideline-concordant treatment - matching a treatment plan to current clinical guidelines without the drift that comes from years of individual habit.
- Managing chronic disease - the ongoing, repetitive monitoring work of conditions like diabetes or hypertension, where consistency matters more than creativity.
Notice what is missing from that list: surgery, emergency trauma response, end-of-life conversations, and anything requiring physical presence or human touch. The authors are not arguing AI should replace doctors everywhere. They are arguing it should replace doctors specifically in the "thinking" parts of medicine — the pattern-recognition and decision-making work that a computer, in theory, never gets tired doing.
The Chess Analogy Driving the Entire Argument
To explain why they think human oversight will eventually hurt rather than help, the authors lean on a comparison from competitive chess. In 1997, IBM's Deep Blue defeated world champion Garry Kasparov. For roughly a decade after that, "centaur chess" — human grandmasters paired with AI engines — actually outperformed either humans or machines alone. It looked, for a while, like the hybrid model was the permanent answer.
Then, by 2017, pure AI systems blew past even the best human-AI teams — a shift The Decoder's analysis of the paper notes the authors lean on heavily. The humans had become the bottleneck, not the safety net.
The JAMA authors argue medicine is tracking the exact same curve, with one darker twist chess never had to deal with: the longer physicians lean on AI to do the cognitive heavy lifting, the more their own independent clinical skills may quietly erode — making the eventual handoff to full autonomy harder to reverse, not easier.
Why Doctors Are Furious?
The medical establishment's response has been swift and largely negative. Professional bodies including the American Medical Association and the American College of Physicians have long held the position that AI should support physicians, never replace their final judgement. The JAMA Viewpoint directly challenges that stance, and physicians have not taken it quietly.
The AMA's Response
AMA leadership, including CEO John Whyte, has pushed back publicly on the paper's conclusions, a dispute STAT News covered directly with the JAMA authors themselves. The core objections generally fall into a few buckets: current licensure and liability frameworks were never built for a world where an algorithm, not a licensed human, makes the final call; accountability becomes murky when something goes wrong; and there is concern that leaning too hard into full automation could erode the human trust and communication that patients rely on during a health crisis.
Robert Wachter's Pushback
One of the more pointed rebuttals came from Dr. Robert Wachter, a UCSF medicine professor and author who was directly referenced in the original JAMA piece. The authors framed his earlier writing as arguing that AI-only care is medicine's "economy class." Wachter responded publicly on his own Substack that this mischaracterized his actual position — he has never argued AI-only care is inherently second-rate, only that the transition needs far more nuance than a single Viewpoint article can capture.
The Conflict-of-Interest Problem Nobody Can Ignore
Here is the part of this story that keeps coming up in every physician forum and healthcare newsletter reacting to the paper: two of the four authors have a direct financial interest in the conclusion they are arguing for. Neal Khosla runs a company that sells AI-powered virtual care. His father's venture firm has invested in that same company, plus OpenAI, one of the AI labs whose models are cited as outperforming doctors in the paper itself.
That does not automatically make the underlying data wrong. Independent studies — including the Google AMIE research and multiple nephrology-focused comparisons published in peer-reviewed journals — point in a similar direction, with AI systems matching or beating physicians on narrow diagnostic tasks in study after study. But it does mean readers, and especially healthcare executives making purchasing decisions, are right to read the paper's more aggressive framing with a raised eyebrow.
What This Means for Patients Right Now?
Nothing changes at your next doctor's appointment tomorrow. This is a policy and research debate playing out in journals, on X, and inside hospital administration meetings — not a rollout announcement. No major health system has announced it is removing physicians from diagnostic decision-making based on this paper.
What is worth understanding is the direction the evidence is genuinely pointing. AI diagnostic tools are already being piloted inside hospitals as assistive technology, and the gap between "assistive" and "autonomous" is where this entire fight is happening. The JAMA authors themselves project that autonomous AI could be ready for real-world implementation on some, possibly all, of the five tasks they identified by 2030 — not this year, and not next year.
The Bigger Picture: AI and the "Cognitive Professions"
Medicine is not the only white-collar, high-trust profession having this exact argument right now. Law, accounting, financial analysis, and software engineering are all going through versions of the same "is the human still adding value, or just adding friction" debate. What makes medicine different — and why this particular paper has travelled so far outside the usual academic circles — is that the stakes are literally life and death, and the profession has an unusually strong culture of gate-keeping who gets to make clinical decisions.
Several tech executives have been remarkably candid about how fast this shift could move across industries generally, not just in healthcare.
At the same time, not every AI capability claim from 2026 has held up under scrutiny. Several bold announcements this year turned out to be overstated once independent researchers dug in.
That pattern of hype followed by pushback is worth keeping in mind here too. It does not mean the JAMA paper's data is wrong — but it is a useful reminder that "AI beats doctors" headlines deserve the same scrutiny as any other bold AI claim before anyone rewrites clinical policy around them.
Should You Trust an AI Diagnosis Today?
Short answer: not on its own, and not yet, no matter what any single Viewpoint article argues. Here is a simple way to think about where things actually stand right now:
- AI as a second opinion tool: Reasonable and increasingly well-supported by data — but pair it with a licensed physician's judgement, not instead of it.
- AI for symptom research before an appointment: Useful for organizing your thoughts and questions, not for self-diagnosis.
- AI replacing your doctor entirely: Not currently how any legitimate healthcare system operates, regardless of what a single research paper projects for 2030.
The honest reality is that the technology is moving faster than the legal, ethical, and licensing frameworks needed to responsibly deploy it at full autonomy. That gap — not the raw accuracy numbers — is the actual bottleneck right now.
What Happens Next - Licensing, Liability, and 2030?
The JAMA authors are not just making a provocative argument and walking away. They have stated they are working on a follow-up paper specifically proposing a licensing and liability framework for autonomous clinical AI — essentially, a legal pathway for an algorithm to hold something resembling medical authority. That paper, once published, will likely be just as contested as this one.
In the meantime, expect three things to keep happening through the rest of 2026 and into 2027: more peer-reviewed studies comparing AI-alone versus AI-assisted versus human-only care across different specialities; continued public friction between AI-optimistic researchers and professional medical bodies; and health systems quietly expanding pilot programs for AI-assisted diagnostics, even as the "fully autonomous" question stays unresolved.
Frequently Asked Questions
Q: Did a real, peer-reviewed study prove AI beats doctors at diagnosis?
The 60% vs. 16% figure comes from a specific comparison cited within a JAMA Viewpoint article — a review and argument piece, not a standalone new clinical trial. The underlying data points, including the ChatGPT o3 comparison and Google's AMIE study, do come from real published research, but the JAMA piece itself is the authors' interpretation and argument built on top of that evidence.
Q: Who wrote the JAMA paper?
Bioethicist Ezekiel Emanuel and researcher Abe Baker-Butler of the University of Pennsylvania, along with Neal Khosla (CEO of Curai Health) and his father, venture capitalist Vinod Khosla.
Q: Is the AMA replacing doctors with AI?
No. The American Medical Association's official position remains that AI should support physicians, not replace their final clinical judgement, and AMA leadership has publicly criticized the JAMA paper's conclusions.
Q: When could autonomous AI actually be used in real hospitals?
The paper's own authors project 2030 as a realistic window for autonomous AI to be ready for real-world implementation on some or all of the five cognitive tasks they identified - not sooner.
Q: Should I trust an AI diagnosis over my doctor right now?
No major health authority currently recommends this. AI tools are best used as a second opinion or research aid alongside a licensed physician, not as a replacement for one.
Read More: