OpenAI claims AI models prove long-standing math conjecture

Artificial intelligence systems have long been tested against human benchmarks in coding and writing, but OpenAI says its models have now crossed into original scientific discovery. On Monday, the company announced that its AI had produced new results in mathematics, computer science, and physics.

The most cited achievement involves harmonic analysis, a branch of mathematics that decomposes complex signals into elementary waves. OpenAI stated that its models proved the local smoothing conjecture in every spatial dimension for the first time. Researchers in the field had previously managed to prove the theory only in three dimensions.

Alex Iosevich, a mathematics professor at the University of Rochester, described the event as a historic development. He noted that the conjecture was formulated by his doctoral advisor, Chris Sogge, making the personal stakes high. Iosevich told Al Jazeera that the initial news was a shock, but he now feels joy and optimism about the direction of scientific work.

“The scientific enterprise as a whole is going to make giant leaps as a result of the AI revolution, and we have the chance to be a part of this historic development,” Iosevich said. He added that he expected theoretical mathematics to change drastically due to the technology, though he admitted feeling a twinge of sadness about the shift.

Beyond harmonic analysis, OpenAI claimed its models disproved the Exponential Time Hypothesis. This hypothesis relates to the time required to solve complex computational problems. The company also said its models provided mathematically rigorous proof of the existence of quantum chromodynamics, a field that studies the interaction of quarks and gluons.

OpenAI said it contacted scientists outside the company to help turn the raw research data into scientific understanding. The company consulted with three expert groups, which it referred to as shepherds, to determine how to release the findings. These groups helped shape the method of disclosure for each specific discovery.

The groups working on the mathematics and computer science problems chose to make the AI-produced data public immediately. In contrast, the physics group decided to defer the release to write up their own interpretation of the findings first. OpenAI acknowledged that these approaches involve tradeoffs in how the work is contextualized for the academic community.

The company, valued at $852 billion, stated that choosing initial shepherds necessarily shapes who gets involved first and creates an asymmetry of information. It said it welcomes feedback on how to broaden participation and make the process more useful for the wider research community.

This announcement follows a similar claim by Anthropic, the creator of Claude. Last month, Anthropic announced the discovery of a new enzyme that resembles CRISPR, a mechanism used in treating conditions such as cancer and sickle cell disease. That announcement drew mixed reactions from scientists, with some hailing it as an exciting development and others suggesting the company exaggerated its significance.