Healthcare AI
Essays, papers and books on how AI is changing clinical care and the training of the doctors who deliver it. Newest first within each section.
AI in Clinical Practice
On Straw Men and Doormen
A gentle rebuttal to Khosla and Emanuel's JAMA argument that a physician in the loop degrades AI diagnosis. Wachter grants that AI often wins on curated facts, but points to garbage-in patient data and the "doorman fallacy" (most clinical work isn't diagnosis), and predicts AI-first primary care alongside AI-enabled clinicians: "And, Not Or."
A Giant Leap: How AI Is Transforming Healthcare and What That Means for Our Future
Bob Wachter's book on how AI is reshaping care delivery, the clinician's role and the health system, with a special section on medical education.
Will AI Rescue Primary Care? Eventually, But Not Yet
Starting from the collapse of Forward, the AI-heavy primary care startup, Wachter argues that primary care is in crisis and concierge care is too expensive for most people. He predicts a tiered system: AI-first care with clinician oversight for basic plans, human-first care for those with more resources or complex needs. He compares it to economy and first-class seats on a plane.
The airline analogy is the heart of it. Economy, business and first class differ in legroom, meals and how much a human attends to you, but every seat flies under the same high minimum safety floor. AI-first care for the "economy" tier is acceptable only if that floor holds: judged against the care people can actually get (often very little, in much of India), and with a safe, well-routed handoff to a human doctor when the case needs one. The analogy stretches further. Autopilot flies most of the trip, yet pilots still go back to the simulator to practise rare failures. Doctors trained alongside agentic AI should likewise keep the skills to give safe care on the day the AI goes down.
The algorithm will see you now
Radiology looked like the easiest specialty for AI to automate, yet radiologists are busier than ever. Models that beat benchmarks stumble on real hospital data, regulation and malpractice insurance slow adoption, and diagnosis is only part of the job; many knowledge jobs may follow the same path.
Medical Education
Physician Training for the AI Era: Rethinking a Century-Old Wager
Medical training has long bet that mastering knowledge plus supervised experience yields clinical judgment. As AI takes over cognitive work, learners risk "never-skilling", so the authors propose strengthening the skills AI makes more valuable, building the ones it makes newly necessary, and de-emphasizing the rest.