Artificial intelligence is already pretty good at writing code, generating images, and answering questions. OpenAI now says one of its internal AI models is doing something considerably more interesting by solving mathematical problems that humans have struggled with for years.
According to OpenAI, a new internal model that began training on August 28 has resolved more than 100 long-standing open problems across most areas of mathematics. The company says the pace of progress has surprised even its own mathematicians, prompting internal discussions about how to inform the mathematical community and prepare the field for what could be coming next.
This is also the model OpenAI says resolved the Navier-Stokes Millennium Prize problem, one of the famous Millennium Prize Problems. If these results continue to withstand scrutiny from mathematicians, AI could be moving beyond simply assisting researchers and toward producing discoveries that humans have been unable to reach on their own.
There is an important catch. Mathematics is not simply a collection of difficult questions waiting for somebody, or something, to spit out the correct answers. How those answers are reached, explained, and ultimately understood is a big part of mathematics itself.
That concern recently led mathematicians to publish an open letter titled A Severe Misalignment of AI in Mathematics. The authors argue that solving major mathematical problems should be a tool for achieving greater understanding rather than becoming the goal itself, warning that rapidly producing answers could actually interfere with the development of new ideas.
AI could therefore create an unusual problem for mathematicians. If machines can generate important mathematical results faster than humans can properly review, explain, and absorb them, researchers could find themselves facing a growing pile of correct answers without enough time to understand what those answers actually teach us.
Not everyone sees that possibility as a bad thing. Mathematician and AI researcher Timothy Nguyen argues that mathematical understanding has always changed alongside the tools available to mathematicians. Computer-assisted proofs have already changed what humans can practically prove, and Nguyen sees frontier AI systems as another chapter in that progression.
OpenAI is now responding to the broader debate by working with mathematicians who have established an independent Advisory Group on Mathematics and Artificial Intelligence. The initial group includes François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten, and Melanie Matchett Wood.
The group will help assess the significance of emerging AI-generated mathematical results, advise on how discoveries should be communicated, and provide guidance on academic and professional standards. OpenAI says members will not be paid by the company, can make their advice public, and will be free to criticize OpenAI’s impact on mathematics.
There is one particularly interesting limitation to all of this. OpenAI specifically says the advisory group will not be responsible for advising the company on how quickly it should pursue its internal progress in mathematics. The mathematicians can weigh in on the discoveries and their consequences, but OpenAI is keeping control of the accelerator.
For now, the number that jumps out is more than 100 long-standing open problems solved by a model that OpenAI says only began training on August 28. If that pace continues and the results hold up under mathematical scrutiny, the question may soon stop being whether AI can solve extremely difficult mathematics and become whether humans can keep up with everything it discovers.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz
