[cosignal]

Some people wonder why the use of ‘AI’ in pathology is not more effective, and this is part of the reason.

[rowanG077]

Hmm that is the opposite what I would have expected. Because there is so much variation an AI that has trained on an insane breadth of data should easily be able to outcompete a human on specialized tasks.

[lanstin]

Unless that training on exceptions isn't quite enough to convey the principles needed to reason about the domain.

I am sure for some problems it will amaze and for some it will not.

[eagerpace]

It seems like exactly the kind of thing AI will be better at than humans. If just needs to be trained on the variations instead of the textbook.

[bitnasty]

My thought is the exact opposite. AI can handle predictable situations fine, but the 20% of situations which are unpredictable require a general intelligence that can understand context that current AI technology (or possibly any AI technology) just doesn’t possess. There will always be situations that training doesn’t cover.

[mejutoco]

Sounds similar to symbolic reasoning (for example algebra) vs finding the most likely completion.