If there was one application where deep learning was supposed to succeed, it was radiology

"people should stop training radiologists now" – Hinton, 2016

>Some were the sorts of teething issues that one might expect to get better over time, such as trouble integrating AI with existing IT infrastructure. Others were more fundamental. AI tools create new tasks and responsibilities, such as “post-deployment monitoring”, which involves “auditing to make sure [the tool] is still performing at the level of accuracy [that was] on the tin,” as she put it.

This is the kind of process that happens with any new technology. Hinton probably just didn't know because he's never worked outside of academia. A common problem with people commenting about "the future of work", AI-related or otherwise.