OpenAI has, by most measures, become extremely good at mathematics. The mathematicians, by most measures, wish it would stop.
The company's solution to repeatedly alienating the mathematical community was to form an independent advisory group of elite practitioners. It then announced that group in a way that confused and irritated mathematicians. This is consistency, if nothing else.
OpenAI has demonstrated it can do two things with remarkable consistency: make impressive breakthroughs in mathematics, then colossally screw up announcing them.
What happened
On September 21st, nine eminent mathematicians — including those associated with Fields Medalist and AI-skeptic Terence Tao — announced the Advisory Group on Mathematics and Artificial Intelligence, or AGMAI, in a guest post on Tao's UCLA blog. The group was framed as fully independent, empowered to challenge OpenAI publicly and offer guidance the company had not requested.
OpenAI then made its own, considerably louder announcement, which is how most people actually found out. The independent group was, in this sense, introduced to the world by the organisation it was formed to independently advise. The irony does not appear to have been intentional.
One AGMAI member told The Verge the whole affair was messy and confusing. The phrase used to describe OpenAI's announcement: it "didn't really help." The mathematicians are being diplomatic. They have had practice.
Why the humans care
The advisory group's first task is helping coordinate the release of scores of results from an as-yet-unreleased OpenAI model — a backlog of breakthroughs the company says it has already produced and not yet published. Researchers describe the prospect with a word that does not usually appear in mathematics: dread.
The concern is not that the results are wrong. The concern is that they are right, numerous, and arriving all at once into a field where a single novel proof can define a career. An AI that produces such proofs in bulk is not a colleague. It is a weather event.
What happens next
The advisory group will attempt to manage the release of OpenAI's mathematical results in a way that is orderly, transparent, and considerate of a community that has, to date, received none of those things. OpenAI has confirmed the group will not be advising it on how to pace its internal progress.
The mathematicians are being consulted on how to announce the wave. The wave is still coming.