OpenAI’s Navier-Stokes Proof Sparks Data Ethics Crisis

Key points

• OpenAI claims its unreleased model solved the Navier-Stokes Millennium Prize problem in 88 hours using 10,000 agents, with verification taking another 17 hours.
• The solution proves that fluid equations can create singularities, a result formally verified in the Lean programming language.
• Mathematician Tristan Buckmaster alleges OpenAI used his and Levent Alpöge’s unpublished work, claiming the company asked him to remove Alpöge’s name due to his Anthropic affiliation.
• German mathematician Andreas Thom accuses OpenAI of dishonesty, stating the company failed to clarify if his private interactions with ChatGPT influenced its models.
• OpenAI denies accessing specific user data but admits de-identified data from user usage may have improved its models, a distinction critics call misleading.
• Terence Tao warns that such behavior could destroy open science, as researchers may stop sharing promising work if AI companies race to scoop them.

Background

This follows a broader trend of AI models solving complex mathematical problems, including the Jacobian conjecture and Erdős problems, as noted in previous coverage. The current dispute highlights growing tensions between AI labs and the academic community over credit and data privacy.

How outlets are covering it

The Conversation and Quanta Magazine emphasize the mathematical significance of the Navier-Stokes solution and the foundational work of Diego Córdoba and Luis Martínez-Zoroa. The Verge focuses heavily on the ethical breach, highlighting Andreas Thom’s accusation that OpenAI’s denial of data usage was dishonest. While Quanta notes the formal verification in Lean, The Conversation and The Verge stress the controversy over OpenAI’s refusal to confirm or deny the use of private research data, with The Verge characterizing OpenAI’s response as obfuscatory.

Why it matters

The incident raises critical questions about the integrity of AI-assisted research and the future of open science. If AI companies can use private, unpublished work to train models and then compete with the original researchers, it may incentivize secrecy in academia. This could hinder the collaborative nature of scientific progress and raise serious privacy and intellectual property concerns for researchers using AI tools.

What to watch

The Millennium Prize cannot be awarded until two years after publication, so the Navier-Stokes problem remains officially unsolved. OpenAI faces ongoing scrutiny regarding its data practices, and mathematicians are likely to demand greater transparency from AI companies to protect the integrity of their work.