OpenAI Seeks Guidance from Top Mathematicians to Improve Performance
After a series of impressive yet controversial mathematical claims led to a reputational crisis, OpenAI is now seeking guidance from human mathematicians to navigate a more constructive path forward.
On Monday, the company unveiled an independent panel of mathematicians designed to advise it and other AI entities on their engagement with mathematical research and the broader math community, particularly regarding the presentation and release of new findings. The sudden announcement took many in the mathematics field by surprise. While some researchers view the formation of this group as a positive initial step, many still have fundamental questions about its purpose, influence, and whether OpenAI will genuinely heed its advice. There are also concerns about whether such a small group of prominent figures can adequately represent the entire mathematical community.
The nine-member panel is composed of well-known figures in mathematics, hailing from prestigious institutions like Stanford, Harvard, Oxford, and Cambridge, and includes several Fields Medalists and MacArthur “genius grant” recipients. It will be hosted by the Institute for Advanced Study (IAS) in Princeton, New Jersey, a renowned center for mathematical research that has been home to legends like Albert Einstein and John von Neumann.
OpenAI stated that the group, named the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), will function independently, with the ability to provide unsolicited advice, publicly comment on the implications of AI on mathematics, and disclose its recommendations. Although the specifics of these activities remain vague, it seems the group will have early access to OpenAI’s research to fulfill its role. Members will not receive compensation from OpenAI and can adjust their membership as needed. The company emphasized that the group’s effectiveness hinges on its members exercising their own judgment and challenging the company’s views. Their responsibilities will include reviewing and communicating emerging results, assessing their significance, coordinating their dissemination, and advising on academic standards in mathematical research.
Despite the emphasis on independence, the actual influence of the group and the rationale behind its formation remain unclear. Simon Machado, a researcher at ETH Zurich, expressed that the situation feels “shrouded in mystery” and questioned the selection process for the nine members. The group’s minimal website indicates it formed after OpenAI approached some members about creating an advisory board, but the mathematicians opted to establish an independent group instead. It’s uncertain which members were initially contacted by OpenAI or how the others were chosen.
In a blog post, advisory group member Martin Hairer addressed the group’s origins and the rationale behind its creation, countering what he described as a “torrent of misinformation.” He noted that AI has significantly impacted mathematical research, raising questions about idea attribution and human understanding. He argued against ignoring AI labs and emphasized the need for dialogue.
Hairer clarified that while the group was formed in response to OpenAI’s outreach, not all members were contacted by the company, and the organization is genuinely independent from OpenAI and other leading AI labs. He assured that members have not signed restrictive agreements and receive no financial compensation, relying on the IAS for technical support.
“Our primary goal is to represent the interests of the mathematical community as best as we can,” wrote Hairer, who is a professor at Imperial College London and the Swiss Federal Institute of Technology in Lausanne (EPFL). He acknowledged that the group is not a comprehensive representation of the community but expressed confidence in the members’ commitment to their responsibilities.
Regardless of how the members were chosen, the group consists of a distinguished set of academic mathematicians. However, its small size raises concerns about whether it possesses the necessary breadth of expertise to address the diverse challenges in contemporary mathematics.
Francesco Fournier-Facio, an incoming mathematics professor at Heriot-Watt University in Scotland, expressed ambivalence about the panel, suggesting that increased engagement with mathematicians could lead to improvements. However, he questioned whether this group truly reflects the community affected by OpenAI’s ambitions. “It feels like an ivory tower,” he remarked.
Machado echoed this sentiment, wondering if the group’s members can fully grasp the broader mathematical community’s concerns regarding AI. “They’re incredible mathematicians, but I’m not sure they represent my views on more political issues,” he said, noting the disparity between the realities faced by elite mathematicians and those of their peers.
The group’s ability to gain the trust of mathematicians will ultimately determine its effectiveness in addressing the fallout from OpenAI’s recent missteps in announcing results and mitigating the impact of future announcements. Hairer acknowledged that members would be “very naïve” to think that AI labs wouldn’t attempt to spin their statements to suit their public relations needs, which far exceed anything the group could produce.
Nonetheless, Hairer argued that it would be hypocritical to disengage now that OpenAI is making an effort, especially after widespread criticism of the company’s previous conduct. The group has yet to finalize its advice but recognizes a pressing issue: “AI companies have recently produced high-profile mathematical results, and their publication practices have not met acceptable standards, however that may be defined.”
Currently, the group is gathering input through discussions with colleagues, a public feedback form, and a blog that has become a vital platform for mathematicians grappling with AI’s influence on their field.
However, the group’s most significant challenge may lie ahead. When announcing the advisory group, OpenAI mentioned that its unreleased model — which recently tackled the Navier-Stokes Millennium Prize problem — has now “resolved more than 100 long-standing open problems across most areas of mathematics.” The company has also indicated it is nearing a solution for another Millennium Prize problem. The group stated that addressing this influx of results is its top priority: “We are currently facing the very specific challenge of advising OpenAI on how to coordinate the release of a large number of significant results in mathematics that they report have been produced by their internal model.”
This new initiative to engage mathematicians acknowledges the significant public relations disaster stemming from OpenAI’s previous attempts, which led to accusations of overshadowing researchers, failing to credit foundational human work, and even appropriating ideas from those utilizing its models. These controversies prompted many mathematicians to question whether the company truly understands or respects the discipline and community it has entered.
In light of recent events and the extensive discussions about the community’s needs, Kevin Buzzard, a mathematics professor at Imperial College London, pondered why OpenAI felt it required such an elite group to grasp what the community desires. “It wasn’t proofs of hard theorems; it was a better understanding of our subject,” he noted. “It’s not entirely clear to me that you need a committee of brilliant people to hammer the point home.”
