Since candidate theories already exist, the real challenge will be figuring out how to discriminate among them experimentally.
Despite the remarkable progress in mathematics, an AI-driven breakthrough here seems extremely unlikely to me, although I'd love to be proven wrong!
