Opinion: AMA CEO: AI won’t replace doctors — it will work alongside them
The essay is part of an ongoing STAT First Opinion debate between American Medical Association CEO John Whyte and oncologist/bioethicist Ezekiel J. Emanuel and Abe Baker-Butler over AI's proper place in medicine. Whyte frames the core issue as a choice not just about which parts of medicine can be automated, but which parts we want to automate—two different questions. He asks whether patients would want AI, on its own, to tell them they have cancer or guide them through what comes next, rather than helping a physician identify the cancer, analyze scans, synthesize medical history, or recommend treatment.
Whyte acknowledges Emanuel's argument that AI will eventually perform many physician functions, sometimes better, and concedes AI can already interpret images, summarize medical records, suggest diagnoses, and answer health questions with surprising fluency. But he argues there is a leap in logic from “AI can do many things physicians do” to “AI can replace physicians,” and that this leap rests on an imprecise view of practicing medicine. He says he is optimistic about augmented intelligence, believes physicians should lead in shaping and adopting these technologies, but profoundly disagrees that increasingly capable AI makes the physician expendable.
He rejects reducing medicine to a collection of tasks such as reading an X-ray, identifying a diagnosis, renewing a prescription, or selecting a treatment. Medicine, he writes, is recognizing when a diagnosis does not quite fit, understanding why an evidence-supported treatment may not be right for the person in front of you, helping a patient choose among imperfect options, delivering life-changing news, and knowing when to push and when to listen—the art and science of medicine.
Whyte says too much of the AI debate asks whether AI can outperform physicians. Studies comparing AI and physicians on diagnostic questions, image interpretation, and other clinical tasks are important, and rigorous benchmarks are needed to understand where AI performs well and where it does not. But passing a benchmark is not the same as practicing medicine: patients do not arrive as carefully constructed clinical vignettes, they describe symptoms imperfectly, and their medical records contain gaps and contradictions. The provided excerpt ends mid-sentence with “The decisions the…”