I think there are certainly issues with ML in radiology (i.e. lots of publications of low power studies or overfit models validated incorrectly), but I agree wi

I think there are certainly issues with ML in radiology (i.e. lots of publications of low power studies or overfit models validated incorrectly), but I agree with you that Marcus was being unfair. It is principle a straightforward problem to solve with deep learning. The reason biggest reasons "no actual radiologists have been replaced" are probably just political/social. Hinton cannot be blamed for those hurdles.