30 Mathematicians Gather at Harvard to Grade Problem Set as AI Threatens the Field
Updated
Updated · The Washington Post · Jun 14
30 Mathematicians Gather at Harvard to Grade Problem Set as AI Threatens the Field
1 articles · Updated · The Washington Post · Jun 14
Summary
Thirty mathematicians met at Harvard this month to evaluate a difficult problem set whose results are meant to help steer mathematics through mounting anxiety over AI.
The exercise carried unusual weight because some researchers now see AI not just as a tool, but as a potential existential threat to how mathematical work is done and valued.
Mathematics underpins fields from traffic modeling and protein formation to faster medical imaging, raising the stakes of any shift in how the discipline evolves.
The Harvard gathering reflects a broader struggle over what remains distinctly human in a field long treated as a core measure of human reasoning.
If AI can solve math's hardest problems, how do we redefine human mathematical genius?
Can human experts effectively referee the coming flood of complex, AI-generated mathematical proofs?
The 2025 AI-Mathematics Challenge: How o4-mini Redefined the Limits of Machine Reasoning and Transformed the Field
Overview
In mid-2025, thirty leading mathematicians, supported by Epoch AI and OpenAI, created an 'ultra-hard exam' to test the limits of AI in mathematics. This challenge reflected a growing trend among tech companies to use mathematics as a clear benchmark for AI progress. The problems were carefully designed to be unsolvable by current AI, aiming to reveal the boundaries of machine reasoning. However, the new o4-mini model surprised experts by solving many of these tough problems, showing that AI's mathematical abilities had advanced much faster than expected and highlighting a major turning point in AI research.