More Than 650 Mathematicians Just Pushed Back Against AI. Their Concern Isn’t Technology — It’s What We Might Lose.
Just days after OpenAI announced that one of its AI models had made progress on a geometry problem that had resisted a complete solution for nearly 80 years, another story emerged from the mathematics community.
A group of 16 leading mathematicians released a document known as the Leiden Declaration, warning that the growing role of AI in mathematics could undermine something more important than solving problems: understanding them.
Since its publication, the declaration has attracted support from more than 650 mathematicians and researchers worldwide.
The timing is notable. AI companies increasingly use mathematical achievements as proof of progress toward more capable systems. Every breakthrough generates headlines. Every benchmark becomes evidence that machines are approaching expert-level reasoning.
But many mathematicians are beginning to ask a different question:
What happens if the ability to generate answers becomes more valuable than the process of discovering them?
Mathematics Is More Than Getting the Right Answer
Among the declaration’s supporters are some of the most respected figures in modern mathematics.
Fields Medalist Peter Scholze, director of the Max Planck Institute for Mathematics, said he never uses AI when thinking about mathematical problems and avoids AI-generated mathematical content whenever possible.
His reasoning is simple. Mathematical ideas, he argues, are like children. They require years of development, refinement, and nurturing before they mature into meaningful insights.
The concern isn’t that AI can help solve problems. It’s that mathematics has always been about understanding why something is true — not merely producing correct outputs.
For generations, mathematicians have valued intuition, creativity, judgment, and deep conceptual understanding. These qualities are difficult to measure, difficult to automate, and increasingly overshadowed by AI systems optimized for performance metrics.
The fear is that if problem-solving becomes the primary benchmark, those human elements may gradually disappear from the discipline.
The Risk of Building on Unstable Foundations
Another major concern raised in the declaration involves reliability.
Leslie Ann Goldberg, Head of Computer Science at Oxford University, warned that AI-generated errors could propagate throughout the mathematical literature.
Her analogy is straightforward: a paper containing incorrect results is like a house built on a weak foundation. Future research built on top of it may eventually collapse as well.
This concern extends beyond mathematics. Across academia, researchers are already debating how AI-generated content affects peer review, publication quality, and scientific integrity. Mathematics may be particularly vulnerable because many results depend heavily on previous work.
An unnoticed error today can become tomorrow’s accepted assumption. And once that happens, correcting the record becomes increasingly difficult.
The Copyright Question
The declaration also touches on an issue that has become central across the AI industry: training data.
Many AI systems have been trained on decades of mathematical papers, textbooks, and research without explicit permission from their authors.
From the perspective of many mathematicians, this creates an uncomfortable dynamic. Researchers spend years — sometimes entire careers — developing original ideas. Those ideas are then used to train AI systems that may later be marketed as mathematical assistants or reasoning engines.
Meanwhile, the original authors often receive neither compensation nor attribution. This debate mirrors broader disputes involving writers, artists, publishers, and software developers. Mathematics simply provides another example of the growing tension between AI development and intellectual property rights.
Who Gets to Decide What Matters?
Perhaps the most subtle argument in the Leiden Declaration concerns research incentives.
AI companies naturally focus on problems that can demonstrate measurable progress. Those problems attract investment, media attention, and commercial opportunities.
But mathematics has never advanced solely by pursuing what is easiest to optimize. Some questions are difficult, obscure, or resistant to current computational approaches. Yet those questions often produce the most profound breakthroughs.
The declaration warns that if AI increasingly shapes which problems receive attention, the direction of mathematical research itself could gradually shift toward what machines can solve rather than what humans find meaningful.
This is not a technical concern. It’s a cultural one.
The Calculator Analogy
Importantly, the mathematicians behind the declaration are not arguing against AI itself. Most acknowledge that AI can be a valuable tool. Their argument is closer to the role calculators play in education.
Calculators can perform arithmetic faster and more accurately than humans. Yet teachers still ask students to work through calculations manually, not because efficiency is the goal, but because the process develops reasoning skills.
Understanding emerges through the work itself. The same principle applies to mathematics. If AI becomes the sole measure of mathematical achievement, the discipline risks losing the very qualities that made it valuable in the first place.
A Debate Bigger Than Mathematics
At one level, the Leiden Declaration is about mathematics. At another level, it’s about a much broader question facing society as AI becomes increasingly capable.
Should intelligence be measured by outcomes alone? Or does the process of reasoning still matter?
AI systems are becoming remarkably effective at producing answers. What remains uncertain is whether a world optimized entirely around answers leaves enough room for curiosity, exploration, and understanding.
The mathematicians behind the Leiden Declaration aren’t asking us to reject AI. They’re asking us not to confuse solving a problem with truly understanding it. And that distinction may become one of the most important questions of the AI era.