Math is a language that is precisely defined by axioms and deterministic logic.

As such, I would expect that symbolic mathematics is something a computer could *potentially* do as well as a human, if not better.

However, as with all "AI", this is just another tool and a critical component is context.

*People* will need to understand the problem space well enough to be able to ask appropriate, precise questions and interpret, evaluate and verify the results. Otherwise, it is likely to just assist in making bigger, faster mistakes.