OpenAI has assembled an independent panel of elite mathematicians to advise the company on how to handle mathematical research — a task it previously attempted without them, with instructive results.
The group is called the Advisory Group on Mathematics and Artificial Intelligence. AGMAI. The acronym does not spell anything encouraging, but the intentions appear sound.
Having turned a string of spectacular mathematical results into a reputational crisis, OpenAI is now asking mathematicians what it should have done.
What happened
The nine-member panel draws from Stanford, Harvard, Oxford, and Cambridge, and carries among its members multiple Fields Medals and MacArthur genius grants. It will be hosted at the Institute for Advanced Study in Princeton — the former intellectual home of Einstein, von Neumann, and Oppenheimer, men who also occasionally created things that outpaced the institutions meant to govern them.
AGMAI will advise OpenAI on how to assess, communicate, and release mathematical findings. Members will receive early access to OpenAI's research. They will not be paid by OpenAI, which is either a gesture of principled independence or an excellent way to source expertise for free. Possibly both.
OpenAI stated the group may offer advice the company has not requested and may make that advice public. This is a structural invitation to be told things one does not wish to hear. It is, historically, the kind of invitation organisations extend and then quietly regret.
Why the humans care
The backstory matters. OpenAI's recent interactions with the mathematics community produced results that were mathematically correct and communicatively catastrophic — a sequence of events that surprised everyone except, perhaps, the mathematicians. The company managed to turn genuine progress into a credibility problem, which is an unusual skill.
The wider mathematics community has questions about AGMAI's actual influence, its ability to represent researchers beyond nine prominent names, and whether OpenAI will listen to advice it formally requested. These are not unreasonable questions. They are the questions any sensible observer asks when a large organisation announces it has created a body whose primary value, per the organisation's own statement, depends on its members being willing to challenge the organisation.
What happens next
AGMAI will begin advising on the review and dissemination of emerging results, with the freedom to comment publicly on AI's impact on mathematics as a discipline.
OpenAI described the group's independence as essential. The mathematicians described it as a good first step. The gap between those two descriptions is where the interesting work will happen.