OpenAI has made significant strides in mathematics, computer science, and physics, according to their recent announcement on Monday. The company shared that it is evolving its approach to disseminate findings, but many researchers are expressing frustration over OpenAI’s previous actions, which they describe as chaotic.

In recent months, mathematicians have struggled to understand a wave of research papers from OpenAI, released without much forewarning. In response to these concerns, the company claims it is consulting with expert “shepherds” to determine the best methods for sharing future results.

The three major discoveries revealed by OpenAI tackle some of the most pressing unsolved problems in their respective fields, with some extending previous work while others offer fresh insights. Each breakthrough comes with a tailored release strategy. For the mathematical breakthrough, OpenAI provided an AI-generated output, allowing researchers to coordinate their own explanatory efforts. In computer science, an external team will be convened to clarify the result. The physics result, however, is still under wraps, with OpenAI opting to wait for expert guidance on how to release these materials.

  • Mathematics: The breakthrough in this area claims to resolve the Kakeya conjecture across all dimensions. This conjecture investigates the minimal space required to rotate a geometric object, such as a needle, in all directions. Notably, a proof of the three-dimensional version contributed to NYU mathematician Hong Wang‘s recent Fields Medal win. Brendan Hassett, a mathematics professor at Brown University, acknowledged that OpenAI’s assertions seem well-founded, although he has not yet checked the Lean formalization used to verify the results.
  • Physics: Investigating the Standard Model, this research focuses on quantum chromodynamics, which describes the strong nuclear force—one of nature’s four fundamental forces. It ties into Yang-Mills theory, a foundational concept in modern particle physics. OpenAI’s research does not, however, address the unresolved “mass gap” that is part of one of the seven Millennium Prize problems.
  • Computer Science: The final claim aims to debunk the exponential time hypothesis, which posits that certain problems cannot be solved significantly faster than brute-force methods. This finding connects to the famous P vs NP problem, which asks whether problems that can be quickly checked can also be resolved quickly.

The diverse release strategies and information complexity make it challenging for the research community to fully evaluate OpenAI’s findings, with some experts stating that the results should be treated seriously. Descriptors such as “huge” and “significant” have surfaced in discussions among researchers.

“It doesn’t make any sense that first only a few distinguished scientists can look at them.” – Constantin Kogler

While excitement about the breakthroughs exists, many researchers are exasperated by OpenAI’s conduct. They argue that further barriers hinder a timely review process of the AI lab’s results. Many feel eager for the broader community to participate in assessing these findings, questioning why OpenAI is selectively choosing experts to oversee this process.

Although external consultation has been welcomed, experts stress that publicly announced results should be accessible for full community scrutiny. Kogler notes that the recent more manageable release is a step forward, yet the selective access remains a concern. Andreas Karch, a physics professor at the University of Texas at Austin, criticized this selective approach, suggesting an open release alongside expert consultation to enhance accessibility and understanding.

“My sense is that for any approach, there is going to be someone who is upset.” – Jared Duker Lichtman

This gatekeeping practice has been advised against by the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), which warned about the risks of unequal access and biased selection. They recommended that AI research outputs be made publicly available with standardized proofs to promote transparency and collaborative understanding of the results.

After the latest results were published, AGMAI urged OpenAI to expeditiously release the underlying materials in both natural language and Lean formats, citing the importance of making these significant findings accessible.

Making findings available to all could not only enhance the quality of science but might also facilitate the broader acceptance of new tools and methods within the community, as noted by Hassett.

“Fundamental results are being solved so quickly without human contribution … the papers are difficult to read and verify.” – Xiaoyu He

Finding a satisfactory method for engaging the wider research community while addressing their concerns presents a challenge for OpenAI. The company recognizes the complexities inherent in different communication approaches and is actively exploring various methods, each carrying its own set of trade-offs. Lichtman posits that any chosen option is likely to displease someone, thus emphasizing the need for careful consideration in their next steps.

Editor’s Take

This development is pivotal as OpenAI strengthens its foothold in areas traditionally dominated by human researchers. The selective disclosure of findings raises concerns regarding access and transparency in a community that thrives on collaboration. Ensuring open access could not only bolster research quality but enhance trust and acceptance of AI tools in scientific inquiry.

Source: www.theverge.com