Mathematicians react with fury as OpenAI releases hundreds of new AI-generated proofs

It’s safe to say that when it comes to mathematicians, OpenAI doesn’t really win hearts and minds.

On Tuesday, ChatGPT’s publisher published hundreds of new articles containing purported proofs of all kinds of mathematical problems – fresh from an explosive controversy over how its AI agents arrived at a solution to the Navier-Stokes equations.

Release management, like that of the Navier-Stokes, has angered experts in the field, who bristle at the company’s models potentially using their work — and at the implicit expectation that they would have to do the hard work of unpacking and verifying the AI ​​model’s sometimes almost incomprehensible proofs.

A group called the Association for Human Mathematics released a statement denouncing OpenAI’s release, which was later reshared and actually endorsed by Fields Medal-winning mathematician Terence Tau, a major figure in the community.

“Mathematicians have a particular view of progress that draws on history and domain-specific considerations,” the statement begins. “We reject OpenAI’s claim that this release advances our topic, and we urge mathematicians and the public to view the value of this release model with appropriate skepticism.”

“The simultaneous publication of more than 700 files is not a demonstration of erudition,” he adds, “but a demonstration of power.”

In an announcement accompanying the release of the documents, OpenAI touted its own efforts to improve how it shares results with the mathematics community and said it had consulted with an independent body of experts in the field called the Mathematics and Artificial Intelligence Advisory Group.

What OpenAI chose to leave out, however, was that the advisory group was against OpenAI’s decision to continue testing its proprietary models on advanced mathematical problems, with the group noting how the models remain inaccessible to the scientific community.

“We want to make it clear from the start: we do not approve of this practice and we ask them to stop testing advanced math problems on proprietary models,” he said in a statement released late last month.

The AHM, in its recent statement, criticized OpenAI for ignoring the advisory group’s advice while using it to “derive legitimacy” in the field.

“By ignoring the central tenet of the advisory group’s position, OpenAI has indicated a complete disregard for the standards of scientific research – standards that ensure that mathematics remains reliable, ethically researched, and in the public interest,” the AHM said.

The mathematical proofs of OpenAI were verified using a programming language called Lean. But even if they are technically valid, the real usefulness of these results remains to be demonstrated, because the evidence is sometimes almost unintelligible to humans. Beyond the question of whether AI’s work will advance the field and contribute to human understanding of the mysteries of mathematics, it also means that these same mathematicians are now expected to do the dirty work of unpacking the hundreds of proofs produced and published by OpenAI in one go.

Learn more about OpenAI: OpenAI says teens can talk about suicide for hours before ChatGPT notifies parents