Will mathematicians lose relevance if other fields bypass them for AI?

The essay argues that the public debate over "AI for mathematics" has collapsed into a narrow question — can neural theorem provers settle famous open conjectures — and that this framing "is sucking the air out of the room" by crowding out formalization and proof assistants, symbolic AI (SAT solvers, SMT solvers, constraint solvers), and non-LLM machine learning used for pattern detection and discovery. Its sharper claim is about where the real danger to mathematicians actually sits: not in AI out-proving them, but downstream, in other fields. "This is the real sense in which we are in danger of being replaced," the essay writes: "disciplines that currently look to mathematics for guidance will turn instead to AI."

The reasoning is about legitimacy rather than capability. The essay traces mathematics's authority to the fact that other disciplines — science, engineering, business, government — have historically needed mathematical understanding, not just correct answers, to do their work; even the most abstract post-Grothendieck questions trace a path back to "counting, measuring, predicting, building things." If practitioners in those fields stop relying on mathematics to understand what they're doing, and instead take answers straight from AI, "the subject will lose a substantial part of its legitimacy and appeal." The essay's proposed response is for mathematicians to settle "what we want mathematics to be" before asking how to get there, and to become contributors of new AI-era reasoning procedures rather than mere consumers of them — partly because LLMs are, in its view, an inefficient way to mechanize mathematical reasoning in the first place (it imagines "teaching an LLM to multiply big numbers by talking through the process, 'put down the seven, carry the three...'").

This sits alongside Can LLM theorem provers tackle genuinely open-ended research problems?, which makes a related but narrower complaint about the same crowding-out: that the field conflates solving well-defined problems with doing open-ended research. This essay broadens that complaint past theorem-proving to the whole AI-for-math toolkit, and relocates the stakes from "what counts as research" to "who still needs mathematicians at all." It also gives a concrete, discipline-level instance of the mechanism described in Does AI separate intellectual form from the thinking behind it?: if AI can hand other fields usable answers decoupled from mathematical understanding, those fields may simply stop asking mathematicians for the understanding.

The excerpt is a single mathematician's reflective position piece, not an empirical study — it offers no measurement of how many practitioners in science, business, or government have actually turned to AI instead of consulting mathematicians, and no survey of mathematicians' own views beyond anecdote (a workshop closing remark, hallway conversations, conference panels). Its central warning is a prediction grounded in an analogy to history (the rise of abstract "conceptual" mathematics), not an observed trend. The implication it supports is accordingly modest: a reasoned argument for broadening what counts as "AI for mathematics" and for mathematicians to act as contributors rather than consumers, not a demonstrated case that other fields are already bypassing mathematics for AI.