OpenAI, after facing a crisis due to controversial mathematical claims, is now seeking guidance from human mathematicians to navigate its way forward more effectively. On Monday, the organization introduced an independent advisory panel composed of esteemed mathematicians, aiming to improve its relationships with the field of mathematics and streamline how new findings are shared and communicated.
The announcement of this nine-member panel, known as the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), took many in the mathematics community by surprise. While researchers acknowledged it as a positive initial initiative, questions loomed regarding its scope, the group's real influence, and whether OpenAI would take their advice seriously. There were also concerns about whether such a small group could adequately represent the wider mathematical community.
The panel boasts a lineup of notable figures in mathematics, drawn from prestigious institutions including Stanford, Harvard, Oxford, and Cambridge, many of whom have received Fields Medals and MacArthur Fellowships. It will operate in conjunction with the Institute for Advanced Study in Princeton, New Jersey, a historic hub for mathematical research previously associated with luminaries like Albert Einstein and J. Robert Oppenheimer.
OpenAI emphasized that AGMAI would function independently, possessing the authority to provide unsolicited advice, publicly comment on the implications of AI on mathematics, and disclose their findings. The panel will have early access to OpenAI’s research to effectively carry out its responsibilities. Members will not receive compensation from OpenAI, and the group retains the ability to change its composition at will. OpenAI stated, “Its value depends on its members being able to exercise their own judgement and challenge ours,” adding that the group will guide OpenAI in how to better present and disseminate mathematical results, and uphold the standards of mathematical research.
Despite the assurances of independence, there remains skepticism regarding the genuine influence of the group. Simon Machado of ETH Zurich remarked that the endeavor felt “shrouded in mystery,” with uncertainty around how the members were chosen. The group's sparse website mentions that it formed after OpenAI reached out to some mathematicians, who then decided to establish an independent entity, but specifics about member selection remain unclear.
Martin Hairer, one of the advisory group members, addressed concerns in a blog post, calling out a “torrent of misinformation” regarding its origins. He acknowledged the profound effect AI has already had on mathematical research, raising crucial issues related to attribution and human insight. Hairer pointed out the futility of ignoring AI labs and emphasized the panel's goal to serve the mathematical community's interests.
Although the panel is not fully representative of the broader mathematic community, Hairer expressed confidence in the group's commitment and sincerity. Nonetheless, past instances raise questions about whether the elite group can effectively connect with the wider concerns of mathematicians regarding the advancements driven by AI. Giovanni Fournier-Facio, a future mathematics professor, expressed ambivalence about the group’s composition and whether it adequately reflects the community impacted by AI's increasing presence in mathematics.
For the panel to gain the trust of mathematicians, its effectiveness in navigating OpenAI’s previous missteps will be critical. Hairer noted the likelihood of AI labs attempting to spin their advice to suit public relations interests. Yet, he argued that engaging with OpenAI is essential, especially given the company's recent performance marred by poor communication and accusations of overlooking researchers' contributions.
The group is currently identifying its strategic approach through discussions with peers, a public feedback mechanism, and collaborative discussions on platforms such as Proofs and Prompts, which focus on AI’s impact on mathematics.
Looking ahead, one of the advisory panel's significant challenges will be to manage the release of a significant number of mathematical results that OpenAI claims to have generated through its new model, which has already tackled over 100 unresolved issues in mathematics and is approaching another Millennium Prize problem. Recent months highlighted OpenAI's PR missteps, including allegations of overshadowing human researchers and failing to acknowledge the foundational work upon which its models built. These issues have left many in the mathematical community questioning OpenAI's understanding and respect for the field.
Reflecting on the situation, Kevin Buzzard, a mathematics professor at Imperial College London, voiced skepticism about the necessity of such an elite advisory group to grasp community needs, suggesting that a genuine understanding of their concerns might not require a select committee of top-tier mathematicians.
