A JAMA Essay Says the Doctor Makes Your Care Worse. Three of Its Four Authors Sell the AI.

I have typed my own symptoms into a chatbot at one in the morning more than once, and walked away impressed. So I am not the reader who recoils on reflex when someone says software can diagnose. I think it often can. Which is why the new JAMA essay caught me off guard: I opened it expecting to nod along, and I could not get there.

The essay ran August 14 in JAMA under the title Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care?, and it argues the whole field has the hierarchy backwards. The American Medical Association and the American College of Physicians keep insisting AI should stay in a supportive role, a co-pilot to the human doctor. The authors say no. They predict autonomous AI will be ready to run “some, maybe many” clinical workflows on its own by 2030. Then comes the line that made me sit up. They suggest that bolting a human physician onto the AI may make the care worse than letting the machine work alone. The doctor, in this telling, is the weak link.

That is a bold thing to put in the most trusted masthead in American medicine. So I did the thing I always do when a claim runs this hot, which is read the author line, then read it again. Ezekiel Emanuel, the University of Pennsylvania bioethicist, is the recognizable name on top. The other three are Abe Baker-Butler and Neal Khosla of Curai Health, and Vinod Khosla of Khosla Ventures. Neal Khosla is Curai’s CEO; the company sells text-based AI primary care and raised $27 million to build it. Vinod Khosla is his father, the venture capitalist whose firm holds stakes in OpenAI and in Curai itself. Three of the four people telling you the human doctor is expendable sell or bankroll the software meant to stand in for him.

So what is the evidence underneath the prediction? The strongest piece is real, and I want to give it its due. Google’s AMIE, its Articulate Medical Intelligence Explorer, published in Nature, went head to head with 20 board-certified primary care physicians and came out rated superior on 30 of 32 evaluation axes by the specialist reviewers, with diagnostic accuracy at least as good and often better. The essay also leans on a newer number: ChatGPT o3 named the correct diagnosis first in 60 percent of 377 complex cases, where a group of physicians managed about 16 percent. Those are not nothing. AI diagnostic reasoning has genuinely gotten good, and I am not going to pretend otherwise to score a point.

But then I looked at how AMIE won, and this is where my read turned over. The contest happened in a text-chat window. The “patients” were trained actors reading scripted cases in an OSCE-style setup. The physicians were asked to work through that same unfamiliar chat interface, typing at simulated people, on the machine’s home turf. The study’s own authors say the plain thing: synchronous text chat is “unfamiliar in clinical practice,” and more work is needed before any of this moves to a real clinic. Wait, then what did we actually measure? We measured which contestant is better at a typed conversation with an actor. Not which one is better in an exam room with a frightened human being.

That gap is not a technicality, because so much of diagnosis lives in the body rather than the transcript. A real patient’s history does not arrive as clean sentences. It arrives as the thing they mention on the way out the door, the flinch when you press the wrong spot, the way someone downplays the pain that turns out to matter, the smell on the breath, the color of the skin. The study rated the machine better at eliciting complaints over text, and I believe it. But eliciting a complaint someone is willing to type is a different act from noticing the one they are trying not to say. That second act is where a lot of medicine actually happens, and none of these studies could test it. There was no room, no body, no silence to read.

Here is what nags at me. Even Vinay Prasad, no defender of the medical establishment, grants the core premise is plausible; he says of course AI will beat many physicians at many discrete tasks, and he is right. But he asks the sharper question. Why publish a prediction about what AI might do in 2030 as a JAMA article at all? This is a Viewpoint, an opinion essay, not a study with a finding. You could post the same forecast on a blog and reach more people. Running it through JAMA does one thing a blog cannot. It borrows the journal’s credibility and staples it to a sales pitch.

That is what this is really about, and it is not fundamentally about AI. It is about capture. We are used to watching pharma launder its interests through prestige journals, and this is the same move with a different buyer: a venture firm and an AI-company founder get medicine’s flagship publication to declare their product superior and the human obstacle expendable, with an ethicist’s name on top for ballast. The conflict here is not a disclosure line at the bottom of the page. It is the page. When the people who profit from the conclusion also write it, the burden of proof does not sit on the doctor they want to delete. It sits on them, and a text-chat contest with actors does not carry it.

So here is my honest landing. I will keep typing my symptoms into a chatbot at one in the morning, because it is genuinely useful for narrowing down what to worry about before I can see a person, and the evidence earns that much. But I will not let a paper written by the people who sell the machine tell me the human across the exam table is the part we can safely cut. When it is my body on that table, I want the model and the clinician who can read what I am not saying, and more than that, I want to know who is paying the author telling me I no longer need her.

Sources

1. JAMA – Emanuel, Baker-Butler, Khosla & Khosla, “Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care?” (Aug 2026)
2. Dr. Vinay Prasad – critique of the JAMA AI essay
3. Forbes – “AI Will ‘Likely’ Beat Doctors At Key Medical Tasks By 2030, Billionaire’s Son Argues” (Roeloffs, Aug 17 2026)
4. Endpoints News – “AI will be better than doctors at some medical tasks, healthcare and VC experts predict”
5. Nature – “Towards conversational diagnostic artificial intelligence” (Google AMIE)
6. Fierce Healthcare – “Curai Health nabs $27M to build out text-based primary care services”