When AI Can Do Research Math, What Is Mathematics Actually For?

In an essay adapted from a 2026 International Congress of Mathematicians lecture, the author sidesteps the usual argument over whether AI will ever match human mathematicians. Instead, the piece takes machine capability at research level as a given and asks a different question: if tools can already perform the work, what is the point of the discipline? The framing treats capability and purpose as orthogonal — you can grant the former without settling the latter.

Problem-solving serves as the central case study. It’s the part of mathematics most obviously automatable and most easily mistaken for the whole enterprise, which makes it a useful lens for separating what the field produces from what it values. The implicit argument is that if the answers themselves become cheap, the community’s attention has to shift to the things machines don’t supply — understanding, judgment about which problems matter, and the human motivations behind the work.

The significance is less technical than cultural. Rather than another capability forecast, it’s a prompt for the mathematical community to define its goals before those goals are defined for it by the tools, a debate that maps directly onto pressures other knowledge-work fields are beginning to feel as AI moves up the difficulty curve.