Indeed, mathematics requires a big deal of creativity. You could argue it is a somewhat subjective discipline, as interesting questions receive infinitely much more attention than a random question. And how do you define what is interesting? There are an infinity of directions you can take -- which for humans is fine, because they know that you eventually have to end up with something your peers find interesting as well. How would a machine do this? I feel this could be akin to train a computer to enjoy music. While computers will continue to become more and more involved in theorem proving, it's not clear to me that they will be able to ask interesting questions within the next few years.

On the other hand, journal papers keep pouring every day at rates that are impossible to keep up. In the future this situation will be hopeless in the absence of some AI to guide us among that mess. But then again, papers are written in a combination of natural language and mathematics.