Why I didn’t sign the Fields medallists’ letter
Concerns over how AI is being used to solve high-profile math problems are exposing a rift within the mathematical community between valuing human understanding and valuing rapid problem‑solving. Commenters debate whether corporate AI efforts that “scoop” open problems risk hollowing out the social structures, incentives, and training pathways that sustain human mathematics, or whether mathematicians should instead adapt and refocus on interpreting, theorizing, and setting directions for AI. The thread also touches on broader themes of technological disruption, the future of expert labor, and whether society will continue to fund human inquiry when machines can generate proofs at scale.
Role of AI in Mathematics
- Many see AI proofs as “helicopters” to mountain peaks: they may solve problems but risk bypassing the human journey that builds conceptual understanding, taste, and new theory.
- Others argue that once a result is known, humans can still “climb the mountain” by digesting and rederiving proofs, and AI just raises both the floor and the ceiling of what’s possible.
- A recurring worry: instant “true/false” answers hollow out the difficult, friction-filled work that produces new abstractions and future tools.
The Fields Medallists’ Letter and This Blog Post
- Commenters debate whether the letter is anti‑AI or anti‑specific AI practices (corporate benchmark races, scooping, prestige farming).
- Some think the blog’s refusal to sign is overly concerned with optics and false “pro vs anti AI” binaries; others agree that criticizing “too many solutions too fast” is futile and adaptation is the only option.
- There is tension between asking AI companies to change behavior vs. expecting the mathematical community to reorganize its own norms and incentives.
Institutional and Incentive Concerns
- Fear that AI results will:
- Undermine existing prestige structures (open problems as curated communal resource).
- Reduce funding and career prospects (“why pay humans if AI can do it?”).
- Push students and policymakers to conclude human mathematicians are obsolete.
- Counterpoint: math is already minimally funded; the real issue is how to justify paying for “understanding” rather than just new theorems.
Human Understanding vs. Problem-Solving
- Thread revisits the split between:
- Problem‑solvers who value understanding as a tool.
- Conceptualists who value understanding as the primary goal.
- Some say future math will shift from proving to:
- Inventing concepts and conjectures.
- Digesting, formalizing, and explaining AI‑generated work.
- Others reply that exactly these activities are at risk of being automated or devalued.
Broader Societal and Political Context
- Several comments generalize to AI in all fields: potential mass displacement, erosion of training ladders, and concentration of power in well‑funded labs.
- Visions diverge between:
- Post‑scarcity “everything is a hobby/game” scenarios.
- Darker futures of techno‑feudalism or dependence on opaque machine knowledge.
- Some propose taxing AI windfalls to reinvest in human science and education; others doubt coordination will succeed.