ZDoggMD on AI in Medicine: What It Can Do, What It Can't, and What Physicians Should Understand

This article discusses ideas from ZDoggMD — Dr. Zubin Damania, MD , Stanford/UCSF-trained internist, founder of the Health 3.0 movement. Full archive: zdoggmd.com | YouTube @zdoggmd

AI in medicine is both overhyped and underestimated. The hype focuses on AI replacing physicians — which it won't do soon, if ever, for reasons that go beyond regulatory barriers. The underestimation ignores the genuine value AI is already delivering in specific, narrow tasks. ZDoggMD has engaged with AI in medicine seriously, neither dismissing it as a threat nor embracing it uncritically.

What AI Is Actually Good at in Medicine

Pattern recognition in high-volume, well-defined data. AI radiology tools that detect diabetic retinopathy, certain cancers on imaging, and specific ECG findings perform at or above specialist-level accuracy in controlled research settings. AI-powered ambient documentation — capturing what was said in a clinical encounter and generating structured notes — is delivering genuine value right now, reducing documentation time without reducing note quality. Drug-drug interaction alerts, sepsis early warning systems, and clinical decision support tools all rely on AI and have measurable patient safety impact.

ZDoggMD's distinction: AI is excellent at closed-domain pattern recognition on well-labeled data in settings where the ground truth is known. It is unreliable in open-domain reasoning, novel clinical presentations, complex ethical decisions, and any situation that requires the kind of contextual understanding that humans develop through lived experience.

The Automation Bias Risk

Automation bias — the tendency to trust algorithmic output over clinical judgment — is a real and documented phenomenon. When an AI tool gives a confident recommendation, physicians tend to follow it, even when their own clinical assessment would have led to a different and better decision. Understanding that AI tools fail in specific, non-random ways is essential to using them safely. The physician who knows what their AI tool was trained on and where it fails is safer than one who trusts it as an oracle.

AI That Does What It Says It Does

NoPajamaTime uses ambient AI to capture clinical encounters and produce structured notes. That's it — no diagnostic overreach, no algorithmic decision-making on clinical questions. Documentation, done well. Start with a free 24-hour proof to create and review one note.