Debate Erupts Over Whether AI Solving Navier-Stokes Would Surpass Human Mathematicians

A hypothetical article about AI solving the Navier-Stokes problem sparked debate among mathematicians and AI researchers over whether "being good at solving problems" equals "surpassing humans in mathematics." Most concluded that posing good questions and inventing new concepts remain human strengths, though the uniqueness of mathematical results makes math harder to shield from AI disruption than art.

Confirmed

• Yann LeCun rejected the claim that AI has surpassed humans in mathematics, arguing that searching for and writing formal proofs of theorems is an automatable mechanical task—just as arithmetic or integration is not all of mathematics—and that inventing new concepts, intuition, and creativity remain AI's weak spots.
• Tony Feng endorsed Christian Szegedy's (littmath) essay on how mathematics should adapt to strong AI, but stressed its premise is hypothetical: even if AI solved Navier-Stokes, current models could not "robustly exceed humans," since posing good questions is itself the core of the mathematical profession.
• The original post cited by @dlouapre argued mathematicians' anger at AI solving problems should not be dismissed as gatekeeping: part of the joy of mathematics lies in exploring the unknown territory around a solution, not merely obtaining a proof. Comments added that for practical tasks, a "straight-line" AI solution is usually still optimal.
• moultano argued mathematics is harder than art to keep "human-only" by tacit agreement: mathematical results are unique—a theorem's priority belongs to a single prover—whereas many artworks can exist, so mathematicians cannot preserve the field as a human activity the way the art world might.
• Igor Carron, responding to a Fields medalist's skepticism about AI's mathematical ability, said whether AI can prove Navier-Stokes is only the math world's concern; for engineers the real battleground is theoretical computer science (TCS), which he predicts is "about to collapse," with separations or non-separations between complexity classes possibly first revealed by AI.
• Keller Jordan raised governance concerns: if mathematicians were a truly self-governing community, they could refuse to produce and simply ignore low-quality AI proofs; he lamented that seeing the community reputed as "the smartest" behave this way is disappointing.

Why it matters

• The debate clarified the distinction between "problem-solving ability" and "mathematical ability," offering a finer framework for assessing AI's impact on fundamental science.
• The uniqueness of mathematical results means AI disruption may hit professional identity and priority systems in mathematics more directly than in art, calling for early adaptation and governance thinking.
• Carron's perspective suggests AI's impact on "can-it-be-solved" questions may first show up in fields like theoretical computer science centered on complexity-class separations and reductions.

2026-09-16 ~ 2026-09-17 · 7 related posts

• [source] Mathematician Tony Feng: current AI is not robustly superhuman yet — littmath · 2026-09-16
• Mathematicians debate AI superhuman reasoning: even solving Navier-Stokes wouldn't end human expertise — littmath · 2026-09-16
• [source] LeCun: Proving Theorems Doesn't Mean AI Is Beyond Human Level at Math — ylecun · 2026-09-16
• [source] Why math is more vulnerable to AI than art: only one theorem gets priority — moultano · 2026-09-16
• Why mathematicians have a point about AI proofs: models hunt trophies, not exploration — _lewtun · 2026-09-17
• The real AI battlefield is TCS, not Navier-Stokes: complexity classes and new algorithms — IgorCarron · 2026-09-17
• Researcher faults mathematicians for 'giving up' self-governance on AI-generated proofs — kellerjordan0 · 2026-09-17