OpenAI wants to consult elite mathematicians about how to not fumble again

You can no longer expect AI-generated mathematical breakthroughs to be released without scrutiny and community input.

OpenAI has established a new independent advisory group, the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), composed of nine prominent mathematicians from elite institutions. This group will advise OpenAI and other AI companies on their interactions with mathematical research, including the presentation and release of new results. The formation of AGMAI follows recent controversies where OpenAI faced accusations of poor communication and attribution regarding its AI models' mathematical achievements. The advisory group, hosted by Princeton's Institute for Advanced Study, aims to provide independent guidance and assess the significance of AI-generated mathematical findings. Members will have early access to OpenAI's research and the freedom to publicly comment on their advice, though their influence and the group's precise operational details remain subjects of discussion within the mathematical community. The AGMAI's immediate priority is advising OpenAI on the coordinated release of over 100 reported solutions to long-standing mathematical problems by its internal model.

The creation of AGMAI signifies a significant acknowledgement by OpenAI of the reputational damage caused by its previous handling of mathematical discoveries, indicating a shift towards more cautious and collaborative research dissemination. This move suggests a broader trend for frontier AI labs to engage more formally with established scientific communities to navigate the complex ethical and practical challenges of rapidly advancing AI capabilities. The success of AGMAI, however, hinges on its ability to genuinely represent the diverse interests of the mathematical community and exert meaningful influence, a prospect currently viewed with skepticism by some researchers concerned about an "ivory tower" dynamic.

The group's formation highlights the increasing tension between the pace of AI development and the established norms of scientific discourse, particularly in fields like mathematics that rely on rigorous proof and community validation. The decision to form an independent advisory board, rather than a company-controlled entity, signals a nascent effort to establish new governance models for AI research outputs that could have implications for other scientific disciplines. Future developments will likely focus on how AGMAI balances its advisory role with the need to maintain independence and effectively bridge the communication gap between AI developers and the academic world.