AI Alone Will Be a Better Doctor Than Your Doctor, a New Paper Argues

Key points

• A paper published in the Journal of the American Medical Association in 2025 argues that AI alone will surpass both human doctors and doctor-AI teams at core medical tasks by around 2030.
• The authors reviewed all published studies on AI in medicine from January 2024 to the present before reaching their conclusion.
• A February 2026 Nature study found that most real patients struggled to hold effective conversations with large language models, the technology behind AI chatbots, in clinical settings.
• The American Medical Association's CEO warns that many studies cited are simulations, not real-world blind trials.
• Experts on both sides of the debate agree on one risk: doctors who rely too heavily on AI may gradually lose the skills they trained years to build.

A paper published this month in the Journal of the American Medical Association lands like a grenade in a hospital staffroom. Its argument, stripped bare: AI, on its own, will soon deliver better medical care than any human doctor, and better care than a doctor and AI working together.

The authors are not fringe voices. Lead author Ezekiel Emanuel is chair of the Department of Medical Ethics and Health Policy at the University of Pennsylvania and a respected cancer specialist. His co-authors include venture capitalist Vinod Khosla and Khosla's son Neal, who runs Curai Health, an online service that already uses AI to treat patients. Their conflicts of interest are declared in the paper.

What exactly does the paper claim?

The authors reviewed every published study on AI in medicine from January 2024 to now and concluded that AI is closing in fast on what doctors do best. Five tasks sit at the centre of their argument: taking a patient's medical history, reaching a diagnosis, ordering the right tests, prescribing treatment, and managing long-term conditions like diabetes or heart disease.

For all five, they write, a "superior autonomous AI" will likely outperform humans by around 2030. Crucially, they go further than most: they argue that keeping a doctor in the loop actually makes outcomes worse. "Humans in the loop degrade AI performance," the paper states.

Emanuel told Wired AI he spent years dismissing this idea. He thought doctoring was simply too complex for machines. He changed his mind roughly a year ago after reading an advance copy of a book by Robert Wachter, head of medicine at the University of California, San Francisco, who described a future where AI handles routine care and doctors handle the complicated cases.

Who is pushing back?

The loudest critic so far is John Whyte, chief executive of the American Medical Association, which represents the country's doctors. He points out that several studies cited in the paper are computer simulations, not blind trials with real patients. He also flags a February 2026 study in Nature which found that, in real clinical situations, most patients could not hold a productive conversation with a large language model well enough to get useful medical guidance.

Whyte's bottom line: AI tools belong inside a care plan that a physician controls, not running the show alone.

Even Wachter, the researcher the paper singles out as too cautious, now concedes the authors have a point. "There will be times when humans will muck up the performance," he said. Yet he insists trained doctors bring something an algorithm cannot easily copy: the human skills needed to deliver a devastating diagnosis with care, or to coax a scared patient toward the right choice.

What does this mean for doctors, and for patients?

Patients stand to gain faster, cheaper access to diagnosis if AI proves as accurate as the paper predicts. But two risks sit alongside that promise.

First, doctors who lean on AI to make decisions may gradually lose the clinical skills they spent years developing. Medical schools are already debating whether students should be allowed to use AI tools during training. If a junior doctor has never learned to take a medical history without AI help, what happens when the system fails?

Second, Vinod Khosla predicts that within a few years regulators will allow AI to prescribe medication directly, without a doctor signing off. That is a significant shift in who, or what, is legally responsible when something goes wrong.

Wachter offers a more measured picture. He calls it the "doorman fallacy": when automatic doors arrived, people feared doormen would vanish. They didn't. They took on new roles. Doctors, he argues, will find their own version of that shift, doing things AI cannot, like human connection, procedural work, and ethical judgment.

The debate is no longer about whether AI can help medicine. It is about how much human expertise we are willing to set aside, and how fast.

Common questions

Does this mean AI will replace my doctor soon?

Not tomorrow. The paper's own prediction points to around 2030, and that depends on regulatory approval, safety testing, and solving the real-world communication problems that the 2026 Nature study identified.

Should I trust AI for a medical question right now?

AI tools can be a useful first step for general information, but they are not a substitute for a consultation with a qualified clinician, especially for symptoms that are new, serious, or getting worse.