OpenAI’s New AI Has Solved Hundreds of Math Problems, Sparking Debate Among Scientists

Published results from an OpenAI artificial intelligence model that has not yet been released to the public have sparked widespread debate in the mathematics community. Hundreds of scientific papers shared by the company on GitHub propose new solutions to mathematical problems that have remained unsolved for years.

The collection initially contained 722 scientific papers, but three were withdrawn on October 7 after an error was discovered in one proof. This reduced the number of materials to 719. The remaining papers are still being reviewed.

According to OpenAI, obtaining each result required computational resources equivalent to approximately three hours of reasoning by ChatGPT Pro. This demonstrates artificial intelligence’s potential to investigate complex mathematical questions in a relatively short time.

However, the results must be independently verified before it can be said that hundreds of problems have been solved. OpenAI uses a formal mathematical verification system called Lean to check the proofs. According to the information provided, approximately half of the papers have been verified using this system.

Although Lean checks the logical sequence of mathematical proofs, whether a formally verified statement fully corresponds to the original scientific claim must be assessed separately. Therefore, the scientific community has not yet completed its evaluation of the results.

The development has raised concerns among some prominent mathematicians. French scientist and Fields Medal laureate Hugo Duminil-Copin said the published results include 16 problems on which he has been working for years.

The scientist’s main concern is not that artificial intelligence works faster than humans. In his view, publishing hundreds of solutions simultaneously does not give mathematicians enough time to examine the results, understand the new methods, and determine future research directions.

The article states that similar concerns were previously expressed in an open letter signed by 25 Fields Medal laureates. The Human Mathematics Association has called for a boycott of collaboration with OpenAI, arguing that such mass releases have become more a demonstration of technological capabilities than a scientific contribution.

However, not all scientists share the same view. Fields Medal laureate Timothy Gowers believes that producing numerous mathematical results is not inherently a negative development. In his view, the main question is how this development will affect the training of future mathematicians and the emergence of new fields of research.

OpenAI said it plans to fund scientific conferences, seminars, and dedicated research programs to help improve understanding of the results. The company is also sharing materials that assist with proof verification and information about computational resources.

If most of the results successfully pass independent verification, this could significantly change the role of artificial intelligence in fundamental scientific research. At this stage, however, it is too early to say that all 719 papers represent verified mathematical discoveries.