MIT mathematicians tell that AI proofs are not clear to persons
Bryna Kra of Northwestern University told that math changed more in two months than in 100 years.
Four mathematicians were at a panel at MIT on Monday to speak about AI in math. In September, OpenAI told that thousands of its AI agents did work for 88 hours. They made a proof for the Navier-Stokes equation, a known problem with no solution until then. The panel did not agree on all points, but it agreed that a person must know how a proof works.
What changed
In 2026, AI models found solutions to open problems that had no known solution. Companies put thousands of agents on one problem. One month after September, OpenAI released 722 manuscripts that AI made.
The numbers
The source gives no test of the proofs. Jacopo Borga tells that 90 percent of the review of a mathematician comes from journals. Kra, who is an editor of a journal, sees many more papers made with AI.
What is not known
Many mathematicians tell that it is not clear how the agents got their results. Michael Hopkins told, “The papers are nonsense.” Persons do not agree on credit, on data safety and on the sense of the proof.
Insight: AI labs win by speed and mathematicians say speed is not the point
OpenAI went first on the Navier-Stokes problem, while Tristan Buckmaster and a collaborator at Anthropic were near a solution. Kra tells that the first proof must not be important. UK mathematicians told before that the companies control the tools. Maybe the AI companies get more from speed. Mathematicians want to move value to how a proof works.
Sources
Posted