I agree. I would love to see AI come up with novel theories like quantum mechanics or the theory of relativity. They don't feel like a remix of previous things that happened in the past to me... He could test that theory himself by training an AI model with data that predates these theories and ask the AI to come up with a theory explaining the results observed during a double slit experiment.

If this weren’t true you wouldn’t see Newton and Leibniz invent calculus near-simultaneously. It may not be sufficient but it is necessary, in my view.