Why AI Struggles to Replace Radiologists: Long Tail, Sample Size, and Hinton's Miss
gregmushen · x · 2026-09-26
Responding to a radiologist's quip that "the first 10,000 chest X-rays are easy to read," the author argues the opposite of the AI-hype takeaway: experienced radiologists know how hard and nuanced the job is, which is why AI can't fully replace them.
- The ML problem itself: can you train something better than a human?
- The long tail: can it generalize to weird cases almost never seen in training sets?
- The sample-size problem: can you collect enough observations to establish safety without humans?
Combined, these make the problem nasty enough to fool some of AI's greats — like Hinton, who famously said in 2015 we should stop training radiologists.
Related event: Radiologists push back: AI replacing them is far harder than VCs think(4 posts)→
More from AGI Musings
- AI's Real Value Is Accelerating Scientific Discovery, Not Generating Games and Memes — Dr_Singularity · 2026-09-26
- Richard Socher's new book 'The Eureka Machine' argues AI unlocks a new era of science — RichardSocher · 2026-09-26
- AI-drafted 166-page Navier-Stokes proof is correct but nearly unreadable for humans — Pascallisch · 2026-09-26
- Debate: LLMs are made of human writing but experience time, identity and death differently — qualadder · 2026-09-26
- Leo Gao on AI views: boosting productivity and doomerism aren't mutually exclusive — nabla_theta · 2026-09-26
- Two years in, this practitioner says truly unattended AI content is a myth — Cold_Hall_5384 · 2026-09-26