Science & Tech · AI · 1 hr ago
AI advances in mathematics raise questions about proof and research
AI systems are beginning to tackle research-level mathematics, including hard problems and open conjectures.
In May, Google DeepMind’s AI solved nine previously unsolved Erdős problems and addressed 44 open conjectures, while OpenAI later shared 722 AI-generated mathematical manuscripts.
Mathematicians say a claimed proof is not accepted until it is checked, and that a result matters less if people cannot understand why it is correct.
The debate is part of a wider change in science: AI can produce many possible answers, while verifying them takes time and expertise.
Researchers interviewed in China see AI as a potential tool for expanding mathematics, not a reason the field or mathematicians will disappear.
They also say mathematics could help explain how AI works and improve its reasoning.
A new joint programme by Zhejiang University and Zhijiang Laboratory will explore AI and mathematics together, including ways to verify AI-generated work and study AI’s mathematical foundations.
The researchers say the goal is to build an open research system that can support mathematics and other sciences.
AI systems are increasingly tackling research-level mathematics, including conjectures and long-unsolved problems, raising questions about how their results should be checked and understood.
In May, a Google DeepMind AI agent solved nine previously unsolved Erdős problems and 44 open mathematical conjectures, according to China News Service.
OpenAI recently made public 722 mathematics manuscripts generated by its internal AI models, prompting debate about AI’s impact on research.
Twenty-five Fields Medal winners warned of a serious misalignment between AI and mathematics, while the Human Mathematics Association called on mathematicians worldwide to stop working with OpenAI.
Two Chinese academicians interviewed by the outlet said AI could change mathematical research, but stressed the need for verification, deeper understanding and open cooperation.
- Who
- AI developers, mathematicians and researchers, including interviewees Li Jianshu and Wang Jian.
- What
- AI has produced mathematical results and manuscripts, prompting debate about proof verification and how AI may reshape research.
- When
- The article was published on 29 October 2026. It refers to developments in May and to OpenAI’s recent release of manuscripts.
- Where
- OpenAI published the manuscripts on GitHub. The article also discusses a joint initiative by Zhejiang University’s Institute for Advanced Study in Mathematics and Zhijiang Laboratory.
- Why
- AI’s mathematical capabilities are advancing, but researchers still need to verify and understand its proofs and consider how AI and mathematics can support each other.
This story does not have two clearly opposing sides.
A Google DeepMind AI agent solved nine previously unsolved Erdős problems and 44 open mathematical conjectures, according to the article.
OpenAI published 722 mathematics manuscripts generated by its internal AI models on GitHub.
China News Service published its article on AI’s advances in mathematics and their implications for research.
Zhejiang University’s Institute for Advanced Study in Mathematics and Zhijiang Laboratory jointly launched the Future Mathematics Plan.
- DeepMind results
- 9 previously unsolved Erdős problems and 44 open mathematical conjectures
- OpenAI manuscripts
- 722, published on GitHub
- Fields Medal winners
- 25 signed an open letter warning of serious misalignment
- Interviewees
- Li Jianshu and Wang Jian
- Future Mathematics Plan
- Jointly launched by Zhejiang University’s Institute for Advanced Study in Mathematics and Zhijiang Laboratory
1 source
What people are saying
From the perspective of the future of mathematics as a whole, I not only would not feel disappointed, but would be very excited.
In traditional mathematical research, when a mathematician claims to have proved something, that in itself does not establish anything; it must be verified before it can be accepted as a proof.
The intervention of AI means that so many difficult mathematical problems suddenly have answers.
AI is not actually collaborating with a particular discipline, but with scientists, and ultimately helping humanity solve practical problems.