OpenAI to work with an independent math panel as it claims over 100 more solved open problems

OpenAI said on Monday that it is working with a new, unpaid panel of nine mathematicians, the Advisory Group on Mathematics and Artificial Intelligence. Its home is the Institute for Advanced Study (IAS) in Princeton, New Jersey. In the same announcement, OpenAI says an internal model it started training on August 28 has settled over 100 problems that had stayed unsolved for years, spanning most branches of the field. OpenAI had already credited the same model with cracking the Navier-Stokes Millennium Prize problem.

The initial members, as listed by OpenAI, are François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood, drawn from institutions including the IAS, Cambridge, Oxford, Stanford, Harvard, Berkeley and EPFL.

What the group will and will not do

According to OpenAI, the panel will help it judge how significant new results are, plan when and how they are made public, weigh in on academic and professional standards, and suggest how its tools could support research and learning. OpenAI says the members can give advice it did not ask for, comment publicly on its effect on the field, publish their advice and decide their own membership.

There is a firm limit, though. OpenAI states that the panel will not advise it on the pace of its internal math work. As TechCrunch reported, the IAS made the same point in its own release, stressing that it cannot make decisions for any AI lab and that the labs alone answer for what they do.

The group's website says its first job is advising OpenAI on coordinating the release of the large batch of results the company reports its model produced. OpenAI has also said before that it is nearing another Millennium Prize problem.

Why it exists

The panel follows months of friction with mathematicians. Earlier in September, 25 Fields Medal winners put their names to an open letter warning that labs racing to crack famous problems were damaging mathematicians' work. Of the nine panel members, only De Lellis signed it. As The Verge reported, OpenAI's earlier result announcements drew accusations of scooping researchers and of giving no credit to the people whose work its models built on.

The group's site says OpenAI first approached some members about an external advisory board, and the mathematicians chose to set up an independent body instead. Hairer wrote in a blog post that not every member was contacted by OpenAI, that members signed no restrictive agreements beyond confidentiality, and that they take no pay, with the IAS handling technical support. He also wrote that recent AI math releases have fallen short of acceptable practice, and that members would be naive to think the labs will not spin their advice for public relations.

Reaction from other mathematicians, as gathered by The Verge, was mixed. Several called the group a reasonable start but asked how members were chosen and how much sway they will have. Francesco Fournier-Facio, who is joining Heriot-Watt University, said the panel "feels like an ivory tower", and questioned whether a small group of prominent academics can speak for the wider field.

The panel is gathering views through conversations with fellow mathematicians, an open online form and the Proofs and Prompts blog.