A paper published in the Journal of the American Medical Association is making a claim that would have seemed far-fetched just a few years ago: that AI alone, without any physician involvement, will deliver better medical care than either doctors or human-AI teams. The authors are not fringe voices. They include Ezekiel Emanuel, chair of the Department of Medical Ethics and Health Policy at the University of Pennsylvania and a renowned oncologist, alongside venture capitalist Vinod Khosla. Their conclusion is that a “superior autonomous AI” will likely surpass humans using AI by 2030, across five core medical tasks: taking patient histories, establishing diagnoses, identifying necessary tests, prescribing treatment, and managing chronic diseases.
That is not a narrow technical claim. That is a claim about the entire practice of medicine.
From Skeptic to Co-Author: How the Argument Shifted
What makes this paper worth examining is not just its conclusion but its origin. Emanuel was not always a believer. For years, he dismissed Khosla’s long-standing argument that AI would handle the vast majority of what doctors do. When Khosla first made that case publicly, in a 2012 piece and a 2016 treatise, Emanuel’s reaction was blunt rejection. The complexity of clinical medicine, he believed, was simply beyond what any algorithm could replicate.
His position began to shift after reading a book by Robert Wachter, head of medicine at UCSF, which described a future where AI-assisted care would be first-class medicine for some, while the broader population would rely on AI alone. That framing prompted a different question in Emanuel’s mind: if AI takes over, what remains for doctors?
To answer it, he and his co-authors reviewed published studies on AI in medicine from January 2024 onward. Their reading of that body of work led them to a conclusion that even they describe as “unsettling but seems probable”: autonomous AI will exceed both physicians and physician-AI hybrids within this decade.
The Case Against, and the Concern Nobody Wants to Name
Not everyone accepts the argument. John Whyte, CEO of the American Medical Association, raised methodological objections: some of the studies the paper draws on are simulations rather than controlled experiments, and not all of them support the paper’s conclusions. A February 2026 article in Nature found that in real-world conditions, most patients struggled to interact effectively with large language models to access their medical expertise. That is a significant practical gap between what AI can do in a test environment and what it can do in an actual clinical encounter.
Wachter, who is cited in the paper as representing the opposing view, concedes more than one might expect. He agrees that AI is already capable, and that the moment when humans in the loop consistently improve outcomes may not last. His concern is different: he argues that trained physicians bring qualities that AI cannot replicate, particularly in delivering difficult news and guiding patients through emotionally complex decisions.
But the more uncomfortable issue sits beneath the surface of this debate. It is what the paper’s authors call “de-skilling.” If physicians increasingly rely on AI for diagnosis, history-taking, and treatment planning, the skills developed through years of training will atrophy. Medical schools and residency programs are already debating whether trainees should use these tools at all, precisely because a physician who has never learned to take a medical history without AI assistance may never develop that capacity. The AMA’s Whyte acknowledged this directly. The irony is sharp: the tool that could make medicine better may simultaneously make the next generation of doctors less capable of practicing without it.
What This Debate Is Really About
The medical question is real and urgent. The broader question is larger. Every profession built on accumulated expertise, on years of training, on the irreplaceable value of hard-won judgment, is watching this debate and recognizing something familiar.
Wachter offers a framework he calls the “doorman fallacy”: the fear that automated doors would eliminate doormen proved wrong, because doormen perform many functions beyond opening doors. The same logic, he argues, will apply to physicians. Doctors will find value in the tasks AI cannot perform: empathy, communication, procedural skill, the human presence that shapes whether a patient follows through on a treatment plan.
That is a reasonable argument. It is also, potentially, a consolation argument. The tasks being displaced here are not peripheral. They are the core of what medical training produces. Vinod Khosla’s position is starker: physicians will remain necessary in the short term for surgery and emergency care, but as sources of expertise and judgment, they are largely expendable. Neal Khosla, who runs Curai Health, an online company that uses AI to treat patients with a physician staff handling prescriptions and complex cases, expects that regulatory permission for AI to prescribe medicine will follow in the coming years.
In Short
A peer-reviewed paper by prominent medical and technology figures argues that autonomous AI will outperform both doctors and human-AI teams across the core tasks of medicine by 2030. The claim is contested, and the methodological objections are legitimate. What is harder to contest is the underlying dynamic: as AI handles more of what physicians do, the expertise required to practice without it will erode. That is not a distant risk. It is a process already underway, and medicine is simply the field where it is most visible right now.
Based on reporting from Wired.