Bioethicist Ezekiel Emanuel and VC Vinod Khosla Say AI Already Beats Doctors at Some Tasks, AMA Pushes Back

Bioethicist Ezekiel Emanuel and venture capitalist Vinod Khosla published a commentary in the Journal of the American Medical Association last month making a claim that has doctors furious: AI-only medical care is likely to beat both physician-only and physician-AI hybrid care at core cognitive tasks, and soon.

The paper reviewed hundreds of studies and concluded AI already surpasses doctors at five specific jobs, according to Futurism: pulling relevant information out of patients, diagnosing conditions, ordering the right tests, managing chronic disease, and reducing errors. Emanuel and Khosla wrote separately in The Atlantic that up to 15 percent of American patients are misdiagnosed every year, a failure rate they say costs tens of thousands of lives.

The AMA says the evidence doesn't hold up

American Medical Association CEO John Whyte pushed back hard, telling WIRED that many of the studies Emanuel and Khosla cited are simulations, not blinded clinical trials, which remain the gold standard in medicine. Whyte also pointed to a Nature study from February that found most patients struggled to get useful answers out of AI chatbots in the first place, a finding that cuts directly against the paper's conclusion.

Whyte's position: AI belongs inside a care plan a physician actually runs, not replacing the physician. The AMA hasn't budged from that stance, calling doctor oversight non-negotiable even as it acknowledges AI's usefulness as a tool.

Emanuel and Khosla aren't backing down either. "We're making a prediction for four years from now," Emanuel told WIRED. "You don't think that [AI is] going to be way more advanced than a doctor?" Their paper goes further than most AI boosters, arguing that human-AI hybrids can actually perform worse than AI alone, meaning doctors staying in the loop might soon make care worse, not better.

Both sides have points worth considering. Google's Articulate Medical Intelligence Explorer, tested against real doctors, got actors playing patients to open up more thoroughly than human physicians did in a study published last year, according to The Atlantic. But Whyte's concern about simulation-based evidence versus blinded trials is a legitimate methodological gap. Nobody has run a large blinded trial putting AI-only care head to head against physicians on real patients with real outcomes.

Deskilling is already happening, regardless of who wins the argument

Senior doctors are already warning that medical students lean too heavily on AI tools, according to Futurism, and internal data from OpenEvidence, a chatbot built specifically for clinicians, shows roughly two-thirds of U.S. physicians already use it. Whether or not AI ever runs care solo, doctors are outsourcing cognitive work to it right now.

Research out of Yale School of Medicine adds a separate warning that gets less attention in this fight. Qingyu Chen and his team, publishing in Nature Communications, found that fine-tuning AI models to improve their diagnostic accuracy on real hospital records also made those models more likely to leak sensitive patient information they'd memorized during training. Making medical AI smarter can make it less safe on privacy grounds, a tradeoff neither Emanuel and Khosla's commentary nor the AMA's rebuttal addresses.

Washington is moving on this at the same time

HHS Secretary Robert F. Kennedy Jr. announced four new FDA leadership appointments on September 8, and one of the four portfolios is explicitly AI innovation, according to the Epoch Times. Karim Mikhail, a 20-year Merck veteran, will run the FDA's vaccine and biologics center. Dr. Michael Davis takes over drug evaluation and research. Bret Koplow was confirmed as tobacco products director after serving in the role on an acting basis. HHS's announcement did not name who will fill the AI innovation seat in the material reviewed here.

Kennedy framed the moves as building "an FDA that moves faster, demands excellence, and delivers results." That's the stated goal. Whether faster FDA review of AI diagnostic tools means better outcomes for patients, or just faster approval with less scrutiny, is an open question nobody can answer yet because no product has gone through the process.

No clinical trial putting AI-only care against physicians on real patients has been announced. No FDA pathway for autonomous AI diagnosis exists yet. The debate between Emanuel, Khosla and the AMA is happening ahead of the regulatory framework that would actually let anyone test the claim at scale.