I think this is one of the most interesting limitations of AI in mathematics. An AI can potentially find a correct path through a huge amount of mathematical machinery, but that doesn't necessarily mean it has found the explanation humans actually want.
For me, the real breakthrough would be an AI that can take a complicated proof and recognize which parts are essential, then reduce it to the simplest argument that preserves the underlying idea. A proof that is correct but impossible for most mathematicians to digest feels more like a result to investigate than the final result itself.
I think this is one of the most interesting limitations of AI in mathematics. An AI can potentially find a correct path through a huge amount of mathematical machinery, but that doesn't necessarily mean it has found the explanation humans actually want.
For me, the real breakthrough would be an AI that can take a complicated proof and recognize which parts are essential, then reduce it to the simplest argument that preserves the underlying idea. A proof that is correct but impossible for most mathematicians to digest feels more like a result to investigate than the final result itself.