Ai in medicine is overhyped — and doctors are calling it out
The pitch sounds compelling: artificial intelligence that reads X-rays, flags diagnoses, answers patient questions, and helps overwhelmed clinicians cut through mountains of data. Microsoft, Google, and OpenAI are all racing to plant their flags in healthcare. But Dr. Mieses Malchuk isn't buying it — and her criticism, published in ZDNET, lands harder than most industry pushback tends to.
The gap between the demo and the exam room
Malchuk's argument isn't that AI is useless. It's that the gap between what these tools actually do and how they're being marketed to the public is dangerously wide. Platforms are being promoted as systems capable of interpreting symptoms and supporting clinical diagnoses with high precision. That framing, she argues, is where things go wrong.
A real diagnosis isn't a pattern-matching exercise. It involves reading a patient's history, picking up on physical signs that don't appear in any dataset, understanding the social and emotional context behind a complaint, and applying years of clinical judgment that no training corpus can fully replicate. Compressing all of that into a chatbot response doesn't just oversimplify medicine — it distorts what medicine fundamentally is.
Her sharpest concern: framing these tools as potential substitutes for human medical judgment. Not assistants. Substitutes. That distinction matters more than most headlines let on.
Why users trust ai answers they probably shouldn't
Research has repeatedly shown that people tend to treat AI-generated responses as authoritative, even when accuracy isn't guaranteed. In most contexts, that's an inconvenience. In a medical context, it can mean a patient delays seeing a doctor because a chatbot gave them a plausible-sounding explanation for their symptoms.
Then there's the hallucination problem. Current large language models still produce responses that sound clinically coherent but contain factual errors or invented details. That's a known limitation that the industry acknowledges in technical documentation and conveniently downplays in product announcements. When hallucinations happen in a legal brief or a marketing email, someone catches the mistake. When they happen in a medical recommendation, the margin for error is a different category entirely.
Malchuk put it directly: