Geoffrey Hinton said in 2016 radiologists would be finished. 🤯

This is Geoffrey Hinton in October 2016, on stage at a machine learning conference in Toronto. His line was that radiologists are like the cartoon coyote who has already run off the cliff and hasn't looked down yet. His advice was to stop training them. Deep learning would read scans better within five years, ten at the outside. Both deadlines have passed. The average radiologist made $571,000 last year, up 9% in twelve months. As of May there were 7,469 open radiology jobs in the US, and 1,470 of those had sat unfilled for more than two months. The number of working radiologists grew about 10% over the decade, while their caseload grew 25% between 2018 and early 2025. He wasn't wrong about the technology. The FDA has cleared roughly 1,400 AI medical devices and about three quarters of them are for radiology. The machines really do read scans. One tool writes the conclusion section of reports and saves close to an hour a shift. He was wrong about the job. Reading the image was one piece of it. Radiologists also consult with other doctors, monitor patients, run procedures, and sign the final read, which Medicare won't reimburse without a licensed physician. Automating one task moved the time to the others. That's the pattern most of these predictions miss. The capability arrives roughly on schedule. The job turns out to have been bigger than the task. Follow @hyperbolic.ai for more content on AI and technology. Media: Creative Destruction Lab #ai #radiology #geoffreyhinton #healthcare #technology