On 6 October 2026, OpenAI published a broad collection of mathematical results produced by an internal AI model. The release made an already uncomfortable question feel urgent: what happens to mathematics when machines can produce new results faster than people can read, verify and understand them?
A few days later, mathematician Terence Tao gave that transition a useful name: Math 2.0.
So what was Math 1.0?
For most of mathematical history, proofs were scarce. Solving a difficult open problem required time, expertise and sometimes years of failed attempts. Because solutions were rare, “solve the problem” became a reasonable shorthand for mathematical progress.
But a proof did more than close a question. It created seminars, conversations and collaborations. Other mathematicians simplified the argument, tested its limits, connected it with existing theories and developed new questions. The proof entered the shared culture slowly enough for people to learn from it.
What changes in Math 2.0?
AI changes the economics of that system. If machines can generate and formally verify many proofs, producing a correct result is no longer the only bottleneck. Attention, interpretation and mathematical taste become more valuable.
More proofs do not automatically create more understanding.
This does not make automated mathematics worthless. It changes the job that humans need it to perform. A result should not simply appear in a repository and be counted as progress. Someone still has to explain why it matters, identify the reusable idea, connect it to other work and turn it into knowledge that another person can carry forward.
The risk is not only losing skills
It is tempting to describe the problem as a calculator debate: if we rely on AI, perhaps mathematicians will forget how to do mathematics. That is part of the concern, but the deeper risk is cultural.
When results arrive without the surrounding conversations, the field can lose the process that creates new researchers and new directions. Junior mathematicians need more than completed proofs. They need accessible explanations, mentors, open questions and a place in the discussion. A flood of answers can make that environment harder to build.
Math 2.0 should still be human
Tao’s proposal is not simply to reject AI. It is to stop treating problem-solving speed as the only measure of progress. In Math 2.0, AI could help formalise proofs, map connections across papers, explore variations of a theorem and take on technical work. Humans would remain responsible for judgement, explanation, creativity and the choice of questions worth asking.
That means valuing exposition and community-building alongside discovery. It means creating tools that strengthen understanding rather than replacing it. It also means resisting a competition in which the largest pile of newly solved problems automatically wins.
My simplest reading of Math 2.0 is this: mathematics is entering an age of abundance, but abundance needs curation. The goal should not be to remove people from mathematics. It should be to use machines so that more people can understand, contribute and ask better questions.
Further reading: This reflection was inspired by Terence Tao’s Math 2.0 post, his Caltech lecture slides and living summary on AI in mathematics. For context on the event that sharpened the discussion, see OpenAI’s 6 October release.