AI will never be able to dumb down a proof to a level simple enough for someone to understand who has never studied math and put a lot of effort into it. Some concepts just need time and effort to absorb no matter how relatively simply they are phrased.
Sure, but some proofs are so complex that even experts can't follow them. As AI progresses, this may constitute an increasing fraction of proofs. I meant that AI could work to find the simplest possible proof.