Hinton also said we should stop training radiologists 10 years ago because it would be solved by AI.
The only way to never make wrong predictions is to not make any.
That's not accurate, see https://www.lesswrong.com/posts/xE2HzcHWPFS9EsJqN/radiology-...
I was at a major radiology conference recently and the radiologists kinda seemed to agree. (The interventional radiologists are a little less nervous, for now.)
Apparently 80% of FDA-approved AI "medical devices" (which covers software, not just hardware) are for radiology.
(https://www.fda.gov/medical-devices/software-medical-device-...)