After Navier-Stokes, OpenAI’s 377 math results leave experts demanding a voice

OpenAI has published more than 370 new mathematical results, stunning mathematicians and igniting a debate over how AI-generated work is vetted and who gets to scrutinize it.

The company released the findings on Tuesday, covering algebra, theoretical computer science and mathematical logic, The Guardian reported. The Age put the tally at 377, reporting that OpenAI says it has solved 377 math equations that have stumped experts for years. The release was designed to showcase what some of OpenAI’s most advanced AI models can do.

Navier-Stokes claim raises the stakes

The publication follows an earlier milestone. Last month, OpenAI solved the Navier-Stokes equation, one of the world’s toughest mathematical problems, which carries a $1 million reward for anyone who cracks it, according to The Guardian.

That achievement prompted concerns from leaders in the field. They say frontier labs should not be testing the most advanced mathematical problems on proprietary AI models that the wider community of mathematicians cannot access.

Institute for Advanced Study pushes back

The Institute for Advanced Study in Princeton, New Jersey, an independent group of mathematical experts, said it does not endorse the practice.

In a statement, the institute warned that AI can now produce mathematical arguments without the person who prompted it being able to understand, verify or take responsibility for them. It said human understanding of mathematics “remains of paramount importance.”

The organization also posed a broader question: how, in this new era, can researchers build a paradigm that treats human understanding of mathematics as part of responsible scholarly output?

OpenAI responds, but testing continues

To address the community’s concerns, OpenAI announced it would work with the institute to give “mathematicians a voice in how we move forward.”

The company did not indicate, however, that it would stop testing its AI models on such advanced problems. Experts in the field say they worry OpenAI is not doing the due diligence required to vet the results.

Who deserves the credit?

Tristan Buckmaster, a New York University mathematician who was working on the Navier-Stokes problem, raised another concern in an interview with The New York Times. Mathematicians who prompt AI models to solve equations, he said, could be supplying information that helps the model reach its answer.

“There’s likely to be a bunch of results where they take someone’s work and then take it to completion,” Buckmaster said, according to The Guardian.

A philosophical divide

The debate has also taken a more reflective turn. Kai Shaikh, a graduate student in mathematics at the University of Toronto, told The Age that humans have two competing natures: a tendency to compete and a capacity to appreciate beauty.

“To me this seems to be a case of the former attempting to strangle the latter,” Shaikh said.

Taken together, the reactions point to a field wrestling with a basic tension. OpenAI is releasing findings by the hundreds, while mathematicians are asking whether such results can be understood, verified and credited the way scholarly work traditionally demands. The Institute for Advanced Study has not endorsed the practice, and OpenAI has not said it will halt it.

Source: https://interestingengineering.com/ai-robotics/openai-377-math-results-mathematicians-concerns