On August 17, 2026, JAMA published a Viewpoint arguing that autonomous AI will likely beat physicians — with or without AI help — at five tasks: taking patient histories, forming a differential diagnosis, ordering tests, prescribing treatment, and managing chronic disease. The authors say it could be ready for real-world use in some workflows by 2030. The authors are University of Pennsylvania bioethicist Ezekiel Emanuel, Penn research fellow Abe Baker-Butler, Curai Health co-founder Neal Khosla, and Khosla Ventures founder Vinod Khosla. Khosla Ventures has backed Curai and other AI health startups. American Medical Association CEO John Whyte disagreed publicly, and on September 9 the two sides published dueling essays in STAT News. UCSF medicine chair Robert Wachter had already published his own rebuttal on August 27.

1. AI Is Ready Now (Ezekiel Emanuel, Neal Khosla, Vinod Khosla)

Emanuel and the Khoslas say the study data already prove autonomous AI is the safer choice for these tasks — right now, not eventually.

Emanuel says human oversight can backfire. "As AI improves, having humans in the loop will likely worsen patient care," he wrote, arguing the field should plan for autonomous deployment rather than treat AI as a permanent assistant.

Their paper cites study after study where AI won outright. ChatGPT beat physicians at differential diagnosis 92% to 74% in one cited study. Microsoft's AI Diagnostic Orchestrator reached the correct final diagnosis 80.4% of the time against physicians' 20%, and did it for less money. In a study of 461 real patient visits, physicians' recommendations were worse than AI's recommendations — even when the physicians could see what the AI had recommended.

2. Someone Has To Be Responsible (AMA CEO John Whyte)

Whyte says medicine isn't a checklist of tasks, and nobody's settled who's liable when autonomous AI gets it wrong.

Whyte doesn't think patients want an AI treating their chest pain. "As it is today, physicians have to be in the loop," the American Medical Association's CEO said. "We're talking about the delivery of health care to people. Do you really want to go to the ER and be treated by an LLM if you have chest pain? I don't think so."

He also asks who's liable when AI gets it wrong. Whyte argues physicians work inside licensure, ethics rules, and accountability structures that AI systems don't have.

Whyte picked apart the paper's evidence, too. He says some of the cited studies were simulations, not the blinded trials that count as the real gold standard. He also says one study the authors lean on found that patients actually struggled to get what they needed from AI chatbots.

Some jobs in medicine aren't about information at all, he says. His example: delivering life-altering news to a patient takes a human being in the room, not a chatbot.

3. Both Sides Are Testing The Wrong Thing (UCSF's Robert Wachter)

Wachter, who chairs UCSF's Department of Medicine, says the AI-beats-doctors studies measure a skill nobody is actually asking AI to replace.

Wachter says AI looks great because the tests hand it clean answers. AI performs well when researchers hand it a curated, tidy set of clinical facts. Real patients hand doctors raw, messy information instead. Figuring out what's wrong from that mess is a skill AI hasn't matched. When regular people talk to AI models directly, without a doctor curating the case first, the tools got it wrong roughly two-thirds of the time.

He calls the replace-the-doctor argument the "doorman fallacy." Beating a doctor at a head-to-head diagnosis test isn't the same as being able to do a doctor's whole job. Wachter estimates AI could handle only about 10% of what he actually does in a day — the rest is end-of-life conversations, sorting out a patient's insurance, and other explicitly human work.

Cutting doctors out breaks a rule of informatics, Wachter says. Doing so "really flies in the face of the classical fundamental theorem" that human-plus-computer should beat either one working alone.

His fix splits the work between AI and doctors. Let AI handle the straightforward cases, and keep physicians for the complicated and emotional ones.

Where This Lands

Emanuel, Baker-Butler, and the Khoslas say the data already points to autonomous AI beating doctors on specific tasks, and they want medicine to start planning for that by 2030. Whyte and the AMA say medicine is bigger than the tasks those studies measure, and nobody has answered who's liable when an autonomous system gets it wrong. Wachter says both sides are testing the wrong skill. AI wins the narrow test only because the test is narrow. AI hasn't touched the harder 90% of a doctor's job yet — end-of-life talks, insurance, judgment calls. The AMA has invited Emanuel to discuss a licensing framework for AI in medicine, but the FDA still hasn't cleared a single autonomous AI system to prescribe anything.

Sources