Had a very optimistic discussion with a math professor friend about the role of AI in research. His take:
1) Having a fast way to disprove conjectures is already very useful and will save mathematicians lots of time.
2) Humans are very good at moving to higher levels of abstraction and will likely integrate AI proofs as a new abstract level that they operate over.
3) Out of necessity, math has operated on claims that can be explained by/to humans but surely there is an even bigger swath of math that cannot be explained to humans but is useful for other human goals. Having AI that can understand and verify those components and turn them into tools for human use could be revolutionary.
The less optimistic discussion was over whether the games that humans set up for ourselves to stay motivated (challenge problems, medals/awards, etc) will be maxxed out by AI in a way that is ultimately detrimental.