However, art isn't solely interpolation. The critical part is that art styles shift around due to innovations or new viewpoints, often caused by societal development. AI might be able to make a new Mondriaan when trained on pre-existing Mondriaans but it won't suddenly generate a Mondriaan out of a Van Gogh training set - and yet that's still roughly what happened historically.
This is giving ML models, more credit than they are due. They are unable to be imagine, they might convincingly seem to produce novel outputs, but their outputs are ultimately proscribed by their inputs and datasets and programming. They're machines. Humans can learn like machines, but humans are also able to imagine as agents. "AI" "art" is just neither of its namesakes. That doesn't mean it isn't impressive, but implying they are the same is granting ML more powers and abilities than it is capable of.