OpenAI sets up board of mathematicians after months of friction

OpenAI has created an independent panel of mathematicians to advise the company on the relationship between AI and mathematical research. The group is called the Advisory Group on Mathematics and Artificial Intelligence, or AGMAI. The announcement came on September 22.

In addition, the initiative comes after a string of controversies. In addition, researchers had been criticizing the way the company disclosed mathematical results produced by its models.

Among the accusations are appropriation of other people's work and lack of credit for human contributions. Therefore, the board was created as a response to a reputational problem.

OpenAI: Who is in the group and how it should work

There are nine members in total. The Institute for Advanced Study, in Princeton, hosts the panel.

The institutions represented include Stanford, Harvard, Oxford and Cambridge. In addition, the group brings together Fields Medal winners and researchers with MacArthur Fellowships.

The mandate is broad. However, the panel is expected to assess the importance of discoveries, advise on how to disclose them, and recommend academic standards for the field.

In addition, the group will also advise on the use of AI to support mathematics research and teaching.

OpenAI says the panel operates without constraints

The company promises independence. According to the company, the group can offer recommendations even without a prior request.

Members can also comment publicly on the impact of AI on mathematics. Likewise, they can make their own recommendations public.

In addition, there is another relevant detail. No member receives payment from the company, and the group itself can change its composition.

Martin Hairer, a panel member, detailed these conditions on the blog Proofs and Prompts. According to him, except for obvious confidentiality requirements, no one signed an agreement restricting public statements.

The group's origin sparked distrust in the community

The creation surprised part of the field. However, researchers interviewed by The Verge called the initiative a first step, yet still raised practical doubts.

Simon Machado, of ETH Zurich, described the group's formation as shrouded in mystery. He questioned how the nine members were chosen.

The group's website explains part of the story. The company approached some mathematicians to discuss an external board, and they decided to form an independent group by inviting other names.

However, there is a lack of clarity about who was approached first. Hairer confirmed the initial outreach and stated that not all members came through this path.

Francesco Fournier Facio, a future professor at Heriot Watt University, said he felt ambivalent. To him, the panel resembles an ivory tower.

Machado raised another point. According to him, researchers in such privileged positions live a reality far removed from most of the field.

The volume of results is already the first challenge

Here the scale of the problem becomes clear. When announcing the panel, the company revealed that a model not yet released solved more than 100 long-standing open mathematical problems.

This is the same model associated with the Navier-Stokes case. The company also said it is close to solving another Millennium Prize problem.

According to the AGMAI website, coordinating this disclosure has become the top priority. The challenge involves advising the company on how to publish a large volume of significant results.

Hairer summed up the diagnosis frankly. However, in his view, the publication of these discoveries fell well short of what mathematical practice considers acceptable.

OpenAI faces skepticism despite the gesture

Not everyone saw the need for the chosen format. Kevin Buzzard, of Imperial College London, questioned the logic of the select board.

According to him, what was lacking was a better understanding of the field, not the proof of a difficult theorem. Therefore, a committee of brilliant names would do little to solve that point.

The size of the group also raises technical doubts. Frontier research requires highly specialized knowledge, and nine people cover only a few subfields.

Hairer acknowledges the limitation. He states that the group falls far short of representing all of mathematics.

There is also the question of communication. According to him, it would be naive to expect labs to avoid framing the panel's statements in favor of their own public relations strategies.

Why this matters to those who write code

The case works as a laboratory for attribution. In mathematics, the asset is an unpublished idea, and that makes the conflict quite visible.

The same dilemma reaches software. Models trained on public code produce solutions similar to existing work.

Therefore, it is worth following the standards the group proposes. Rules on credit, disclosure timelines, and recognition of human contribution are likely to become a reference.

Meanwhile, the group is gathering opinions. There is a public contribution form and open discussions on Proofs and Prompts.

Follow our profile on Instagram!