What advantage humans have over computers in mathematics?
There is an active research in machine learning world, more specifically in NLP (natural language processing) that tries to answer contrary, and even more general question:
What computers can do, that we previously thought that only humans can do?
In my opinion the most interesting approach is an ongoing collaborative benchmark for NLP models called BIG-bench; it is just a bunch of tasks which you can test your model on, or test yourself (or your friends, family or foes)! What is important, you can also add your own tasks.
Regarding the OP question, you can browse the tasks by the keywords here. For example you can test if your model is capable to "understand" Mathematical Induction. You may ask it if the following is true:
1 is an odd integer. Adding 4 to any odd integer creates another odd integer. Therefore, 9 is an odd integer.
You can even "check" if you model is self-aware. Moreover, regarding @logicute answer, you can "measure" your models creativity in various of tasks.
So, can computers do math already or not?... It haven't been tested carefully yet, but not yet perhaps. Will they?... To answer this question I believe we should keep track of the projects like BIG-bench.