Twenty-five of the world’s most decorated mathematicians have issued a stark warning. They argue that the artificial intelligence industry is fundamentally at odds with the goals of mathematics. The signatories, including Terence Tao and Maryna Viazovska, claim that using large language models to mass-produce solutions undermines the discipline’s true purpose. For these experts, solving a problem is merely a proxy for achieving conceptual understanding. When AI shortcuts this process, it destroys the fertile ground needed for new ideas to take root.
The core conflict lies in speed versus depth. AI companies treat famous unsolved problems as benchmarks to conquer for publicity. This approach skips the “long and arduous process of talks, discussions, simplifications” that traditionally follows a breakthrough. The statement notes that AI-generated solutions often arrive with no proper writeup or attribution. This raises severe plagiarism concerns and breaks the human transmission chain essential for integrating new methods into the mathematical canon. Without human care, AI-conceived ideas remain dead ends rather than living contributions.
This tension recently exploded in a controversy involving OpenAI. Rumors circulated that the company trained models specifically to beat researchers to a partial solution for a Millennium Prize Problem. This happened despite earlier claims from OpenAI leadership that they were not optimizing for math. The incident highlights how corporate incentives can override scientific integrity. The mathematicians see this as a symptom of a broader threat to all intellectual work. If the process of learning is discarded for the sake of the final product, we lose the ability to formulate new questions.
The implications extend far beyond academia. Education systems already struggle with the gap between AI-assisted homework and banned exam usage. As grades diverge based on tool access rather than comprehension, the value of traditional training erodes. The signatories are not calling for a ban on AI. They acknowledge its potential to accelerate study. However, they insist that humans must control how these tools are deployed. If we do not address this misalignment urgently, we risk losing sight of what intellectual work was meant to achieve in the first place.
