Terence Tao Declares Mathematics 2.0 as AI Proofs Flood the Field

Fields Medalist Terence Tao has formally declared the arrival of a "Mathematics 2.0" era, arguing that AI is rapidly turning mathematical proof from a scarce resource into an abundant one. Since OpenAI's announcement of solving the Navier–Stokes millennium problem, Tao and other mathematicians have both criticized OpenAI and called for a pause, but Tao now poses a deeper question: how can the growing flood of AI proofs be channeled to increase human understanding, improve training, and yield more reliable real‑world contributions?

From "Proof Scarcity" to "Proof Abundance"

In the Mathematics 1.0 era, proofs were scarce; solving an open problem required deep expertise and immense effort, so "who solved what first" became the primary measure of contribution. AI is fundamentally altering that environment. OpenAI recently released 722 mathematical papers in one night, targeting the Riemann hypothesis, Hodge conjecture, BSD conjecture, and other major problems. It also extended the recent Fields Medal work of Wang Hong on the three‑dimensional singular set dimension conjecture to four dimensions. These feats rewrite the rhythm of research and shatter the old evaluation system, forcing mathematicians to ask: beyond proving, what remains a mathematician's true moat?

Tao states: "The barrier to obtaining answers has indeed lowered. With sufficient compute and access to frontier models, many previously hard open problems can be solved in relatively short time. But judging whether an answer is correct, how significant the contribution is, and how it connects with existing research still requires careful handling. Otherwise we face a worrying misalignment — problems are solved without increasing human understanding."

Tao identifies three overlapping conditions that make mathematics especially vulnerable to AI:

1. Proofs can be objectively verified.

2. Mathematical objects and reasoning are highly digitizable.

3. Vast amounts of high‑quality research literature already exist in digital form.

Combined with massive investment from frontier AI labs, these factors have turned mathematics into a core battlefield for model capability advances.

Tao also invokes the classic bullet‑hole diagram (survivor bias) to caution that most frontier problems remain unsolved; AI capabilities are uneven, and labs only showcase cherry‑picked successes. He urges the community to establish a transparent, scientific evaluation mechanism that reports failures and compute costs alongside successes.

Blindly Optimizing "Solving" May Harm Long‑Term Health

Tao uses a lighthouse analogy: open problems are lighthouses guiding exploration of the surrounding mathematical landscape; reaching the lighthouse does not equal completing the exploration. The process of attempting a problem — whether successful, failed, or partially advanced — often yields insights more valuable than the final answer. If automated tools deliver answers too early and the reward system only recognizes "solved", the unexplored paths lose motivation. Millennium prizes are finite, but the pathways, understanding, connections, and new questions growing from a single problem are nearly infinite. Tao warns: "Continuing to blindly optimize the 'solving' metric will damage the long‑term health of mathematics!"

This concern underpins his sharp critique at the SAIR conference: "It's simply crazy — AI companies are accelerating endlessly without knowing what happens next. We must slow down; there is absolutely no reason to go this fast, none at all!" In a September blog comment he wrote: "AI can be a tool for mathematical research, but chasing short‑term targets may sacrifice the long‑term sustainability of the mathematical community." He worries that when answers become instantly available, people may stop taking the time to understand them and explore the surrounding mathematical world.

After the Proof: Understanding, Training, and the Real World

To illustrate why human understanding remains essential, Tao presents a striking thought experiment — the cancer therapy trial :

Suppose you ask an advanced AI to find a cancer therapy that passes Phase III clinical trials. After massive compute, it returns a mixture of unknown chemicals: the model predicts it kills cancer cells, Lean verifies the proof, and the clinical trial passes. But no one knows how it was found, nor whether the AI genuinely discovered a therapy or exploited a loophole in the trial protocol. Before that needle enters your vein, would you not want at least one oncologist to understand its mechanism, one mathematician to grasp the mathematical reasoning behind the model, or one scientist to understand the search model itself?

This experiment highlights that model‑internal verification is not enough; someone must judge whether the result aligns with real‑world goals. Models can generate answers, verify reasoning, and accelerate exploration, but they cannot replace human oversight.

In his earlier essay "Mathematics in the AI Era", Tao breaks problem‑solving into five stages:

Generate proof

Verify correctness

Explain clearly

Be understood and accepted by the community

Integrate into the field's standard theory

The final stage is especially slow: researchers must find more natural formulations, better proofs, and connect results to other knowledge so they enter textbooks and the next generation's toolbox. This is precisely what AI struggles to accelerate. If AI mainly speeds up the early stages, "proof indigestion" may occur. Tao therefore advocates: "Reduce emphasis on racing to generate proofs first, and increase emphasis on explanation, review, and knowledge curation."

OpenAI's 722 Papers and the Human Mathematics Association

On October 7, OpenAI's release of 722 papers triggered a forceful response. Tao spearheaded the creation of the Association for Human Mathematics (AHM) and issued a joint statement: "Releasing 700+ files at once is not research; it is pure compute‑dominance posturing. We strongly urge all mathematicians to stop collaborating with OpenAI and return to a scientific vision centered on human understanding!" Tao insists that formal verification alone is insufficient; authors must be able to explain their results, otherwise the work remains incomplete.

Does a PhD Still Matter in the AI Era?

Tao argues that the value of a PhD no longer lies in paper production but in "cultivating a systematic ability to absorb difficult material, synthesize knowledge, and pose good questions." He compares it to diet and exercise: when food and transport are abundant, one must deliberately exercise; when answers are at hand, one must consciously maintain cognitive stamina and resist the convenience of bypassing thought — capabilities that doctoral training is designed to build.

Toward a Sustainable "Mathematics 2.0" Paradigm

Tao reveals concrete experiments toward human‑AI co‑growth. The SAIR Foundation's Open Mathematical Models initiative aims at daily research needs: understanding difficult arguments, checking citations, exploring examples, writing code, and formalizing proofs. The first open models are expected to be released soon.

In closing, Tao identifies the main bottleneck for Mathematics 2.0 as imagination . The supply of proofs is changing, but how to harness that change to advance understanding, training, and exploration together requires collective choices. After a problem is proved by AI, the mathematician's next questions become: "What else can we learn from this answer?" and "What can we do next?"

References: [1] https://teorth.github.io/tao-web/ai-views.html [2] https://x.com/ns123abc/status/2109045309112250560 [3] https://teorth.github.io/tao-web/slides/math-2-0-caltech-2026.pdf [4] https://arxiv.org/html/2608.16753v1

Code example

[1]https://teorth.github.io/tao-web/ai-views.html
[2]https://x.com/ns123abc/status/2109045309112250560
[3]https://teorth.github.io/tao-web/slides/math-2-0-caltech-2026.pdf?utm_source=chatgpt.com
[4]https://arxiv.org/html/2608.16753v1?utm_source=chatgpt.com