That seems like a straightforward one; if humans are incapable of original thought then the original part probably came from sources that are not human and 'innovation' in the sense of 'an AI couldn't do this' never existed. We just need to make the AI better at interfacing with the real source of novelty - reality - than the current method of trying to interpret existing pictures on the internet.

There will be a big leap forward when we get to teaching these things from video streams so they can get the concept of 3D properly.