OpenAI faces backlash for gatekeeping new science breakthroughs

The new science results

OpenAI recently announced three significant breakthroughs in the fields of mathematics, computer science, and physics. These results involve some of the most consequential unsolved problems in modern science, ranging from theoretical physics to fundamental logic. Some of these findings build on previous work, while others represent entirely new ground for the company.

The mathematics result purports to resolve the Kakeya conjecture in all dimensions. This problem asks for the minimum amount of space required to rotate a needle or pencil in every possible direction. A three-dimensional version of this problem was recently solved by a human mathematician who won a prestigious Fields Medal.

In the realm of computer science, the AI claims to disprove the exponential time hypothesis. This is a widely held assumption that certain problems cannot be solved significantly faster than using a brute-force approach. This finding is closely related to the P versus NP problem, which is one of the seven Millennium Prize problems in mathematics.

Selective access and 'shepherds'

To manage these findings, OpenAI is using a new approach involving expert 'shepherds.' These individuals are tasked with deciding the best way to share the latest breakthroughs with the broader community. However, the specific release strategies vary significantly depending on the complexity of the individual result.

For the mathematics result, OpenAI released the AI-generated output directly to let researchers organize their own explanations. The computer science finding will be reviewed by an external group of researchers to help explain the results. In contrast, the physics result has not been released at all, as OpenAI is deferring to an external group of experts to decide on a release plan.

This tiered approach has sparked frustration among the scientific community. Many researchers are tired of the company's perceived gatekeeping and the fact that only a few distinguished scientists are being given first look. Critics argue that handpicking a specific group to oversee the process creates an unnecessary barrier to public scrutiny.

The physics of the Standard Model

The physics result focuses on quantum chromodynamics, which describes the strong nuclear force. This is one of the four fundamental forces of nature, alongside gravity, electromagnetism, and the weak nuclear force. The work explores the relationship between this force and Yang-Mills theory, a cornerstone of modern particle physics.

Yang-Mills theory is significant because it is at the heart of one of the seven Millennium Prize problems. While OpenAI's work does not address the unresolved 'mass gap' within that problem, it touches on central questions in both physics and mathematics. The existence of quantum Yang-Mills theory has been a goal for mathematical physics for decades.

OpenAI stated it delayed the release of the physics results until the Lean formalization was complete. Lean is a tool used for computationally verifying mathematical results. Because the formalization is now finished, some experts argue there is no longer a scientific reason to delay the public release of the natural-language and Lean outputs.

Calls for open democratization

The Advisory Group on Mathematics and Artificial Intelligence (AGMAI) warned against OpenAI's current practices just one day before the announcement. They argued that selecting certain people to receive outputs first is unnecessary and risks biased selection. They recommended that AI labs instead release outputs on scholarly repositories with formalized proofs and information on how problems were chosen.

Researchers who spoke to The Verge expressed that the large labs already have significant power. They argued that the company should not be picking winners and losers by controlling who sees the data first. Experts suggested that a competitive process for selecting researchers might be fairer, though they acknowledged it might still result in a similar group of high-level scientists.

Some mathematicians believe that making findings openly available would lead to better science and faster adoption of new tools. They argue that having many groups around the world wrestle with these results supports rapid progress. By making the work public, the community can better verify the papers, which are often difficult to read and verify because they are solved so quickly without human contribution.