Terence Tao warns AI-driven scooping could erode mathematics and open science

Following OpenAI's reported progress on the Navier-Stokes equations, mathematician Terence Tao posted a long warning that mathematical automation may be eroding the field itself, and that AI-driven scooping could reverse centuries of open science. Several authors extended the discussion to transparency, knowledge boundaries, depth of understanding, and serendipitous discovery.

Confirmed

• Tao used the global regularity problem for the incompressible Navier-Stokes equations (and the corresponding Euler problem) to argue that the intermediate theory and methods a field depends on are produced by humans working through hard problems; if AI skips straight to answers, future research loses its foundation.
• As relayed by @GaryMarcus and @DynamicWebPaige, Tao noted that today a mere rumor that someone is working on a problem can trigger massive AI-driven efforts that "flatten" it before the original project matures; these incentives may lead researchers to stop sharing promising directions, reversing centuries of open-science tradition across mathematics and all sciences and engineering.
• Per @GaryMarcus, Tao's stance on AI has shifted notably: in fall 2024 he was skeptical but optimistic about o1 and human-AI coexistence, but has since issued a string of warnings, including three posts on mathstodon in two days.
• Tao also said ChatGPT's release ended openness in ML/AI research, though that field was always industry-linked; he fears a civilizational-level tragedy if such closure spreads to pure mathematics (relayed by @KordingLab).
• @DynamicWebPaige emphasized that Tao is regarded as one of the most moderate, AI-friendly mathematicians, making his concern especially alarming; if sharing conjectures means being scooped uncredited by systems with vastly more compute, disciplines slide into a "dark forest."
• @ValerioCapraro relayed Tao's view that pure mathematics poses hard problems not for the answers alone but because the human solving process advances the field; opaque premature AI solutions could pollute this process and even net-harm mathematical progress.
• @jjvincent (Engineering for AI) highlighted the lack of transparency: black-box AI solutions erase the clues toward future research directions, and automation may be quietly weakening the discipline itself.
• @qberthet offered a counter-model: mathematical knowledge is an "island of knowledge"—the larger it grows, the longer its boundary with the unknown—so mathematics is not exhaustible; every result breeds new questions.
• @tak3sh8 argued that those who solve problems with AI rarely understand them as deeply as those who grapple for months or years, and thus struggle to pose the right next question—perhaps why AI accelerates solving but not mathematical progress. He also asked: if AI solved Navier-Stokes and left piles of Lean-formalized proofs, what would the next interesting question be?
• @jjvincent generalized the concern to other sciences: the history of science is largely one of accidental breakthroughs—CMB discovered while cleaning antenna noise, CRISPR found in yogurt-fermenting bacteria—raising the question of whether such serendipity survives AI-driven automation.

Why it matters

• The debate touches a core question of the AI era: is the value of mathematics and science in the answers, or in the theories and methods accumulated along the way? Large-scale AI "grade-skipping" could force a reorganization of disciplines traditionally driven by hard open problems.
• Tao's warnings extend the discussion from solving capability to research incentives and scholarly communication: whether open science can survive high-speed AI scooping is a more systemic question than any single breakthrough.
• Sparked by OpenAI's Navier-Stokes news, the discussion spread from academia to tech media; the impact of black-box solutions on research-path continuity and serendipitous discovery merits continued attention.

2026-09-08 ~ 2026-09-10 · 18 related posts

• Terence Tao: AI solving key open problems could stunt a math field's growth — tak3sh8 · 2026-09-08
• Math as an island of knowledge: why new results keep generating new problems — qberthet · 2026-09-08
• Why AI-solved math problems rarely lead to the next good question — tak3sh8 · 2026-09-08
• If AI cracks Navier-Stokes and leaves a Lean formalization, what's next? — tak3sh8 · 2026-09-08
• [source] Terence Tao: premature AI solutions to math problems could harm mathematics itself — ValerioCapraro · 2026-09-08
• [source] Terence Tao: AI scooping may end open science and close off research sharing — GaryMarcus · 2026-09-09
• Terry Tao warns AI-era scooping could push math and all sciences 'dark' — DynamicWebPaige · 2026-09-09
• Terence Tao's AI views shift sharply as Marcus sounds alarm on field's direction — GaryMarcus · 2026-09-09
• Terry Tao's scoop fear spreads: sharing hunches means GPU clusters outrun your plans — DynamicWebPaige · 2026-09-09
• [source] Terence Tao: AI math automation may hurt mathematics by hiding hints and chilling research sharing — jjvincent · 2026-09-09
• Terence Tao on OpenAI's Navier-Stokes result: math automation may erode the field itself — jjvincent · 2026-09-09
• Does AI Automation Kill Accidental Breakthroughs? Lessons From CMB and CRISPR — jjvincent · 2026-09-09
• Terence Tao warns AI-powered rumor-chasing could reverse centuries of open science in math — FlorianGallwitz · 2026-09-09
• Terry Tao: AI's closure of ML was bad; in pure math it'd be a civilizational tragedy — KordingLab · 2026-09-09
• Terry Tao Warns AI-Era Scooping Could Push Math and Science Into Silence — windx0303 · 2026-09-09
• Terence Tao warns AI-powered problem-flattening may reverse centuries of open science — erikphoel · 2026-09-10

2 near-duplicate retellings: GaryMarcus · GaryMarcus