xi's moments
Home | Opinion Line

To rely on AI alone or not is a very pressing question

By Li Yang | China Daily | Updated: 2026-11-22 20:24

The Riemann Hypothesis, a puzzle concerning the distribution of prime numbers, has challenged generations of mathematicians since German mathematician Bernhard Riemann formulated it in 1859. Now, an employee at Anthropic claims to have solved it with the help of Claude, the company's artificial intelligence model. This development raises an uncomfortable question: what happens when machines produce answers humans cannot understand?

If validated, Levent Alpoge's achievement could mark a watershed. His proposed proof comprises about 17 million lines of Lean code, a language used to formalize mathematical arguments. The AI-generated explanation reportedly ran to more than 250,000 pages. Although the code has passed formal verification checks, Alpoge acknowledges that he has not fully grasped the mathematics underlying it. The Clay Mathematics Institute still lists the hypothesis as unsolved.

Mathematics is not simply about producing correct answers. A proof should explain why a proposition is true, connect ideas and give researchers a foundation. As Rutgers mathematician Alex Kontorovich has observed, reading a proof involves following its reasoning and developing a deeper understanding of its logic. A computer's approval cannot substitute for that intellectual process.

The mathematical community's response reflects this tension. Some researchers see a historic opportunity. Fields Medalist Shing-Tung Yau has called the development a great contribution, potentially heralding an era where researchers find answers first and understand them afterwards. James Maynard, a mathematician at the University of Oxford, has expressed confidence in the proposed proof but emphasizes the need for scrutiny. Others are more cautious. Professor Terence Tao in the Department of Mathematics at the University of California describes the challenge as "proof indigestion", suggesting that the ability to generate arguments may outstrip humanity's capacity to digest them. A proof nobody can explain is a peculiar achievement.

Nor is formal verification infallible. While the Lean code is a powerful safeguard against logical errors, its reliability depends on the integrity of the software, formalization and the proposition being checked. If an AI system changes the meaning of the original mathematical statement when translating it into formal code, a flawlessly compiled proof may establish something other than what its authors intended. Human supervision remains indispensable.

Commercial incentives may also distort mathematical priorities. Anthropic's announcement came immediately after its stock-market debut, raising questions about science and corporate publicity. OpenAI's earlier claim of an AI-assisted breakthrough involving the Navier-Stokes equations also attracted scrutiny. When companies compete to announce spectacular discoveries, the temptation is to celebrate speed rather than the slower accumulation of knowledge. Twenty-five Fields Medalists have warned that turning mathematics into a race could distort the discipline.

The consequences are even more profound for education. If machines generate proofs beyond the comprehension of their collaborators, teaching students to merely solve established problems may no longer be sufficient. Future mathematicians will need to formulate questions, evaluate machine reasoning, identify assumptions and judge which discoveries matter. Yet this is no reason to abandon the fundamentals of mathematics. The struggle of tackling difficult problems develops intuition, discipline and resilience. Without these foundations, students risk becoming operators of systems whose conclusions they cannot independently assess.

The deeper issue is cognitive outsourcing. Mathematics has always relied on tools, from calculators to computers. However, AI is now responsible for reasoning and discovery, not just computation. If people accept conclusions without retracing the intellectual journey, critical judgment may atrophy. The danger is not that machines will think, but that humans increasingly will not.

The Riemann Hypothesis may or may not have been solved. However, mathematics should not measure progress solely by the number of problems machines solve. The purpose is to expand human understanding. An AI capable of reaching conclusions beyond human comprehension is impressive; a mathematical culture that no longer insists on understanding them would be a worrying proposition. The question is not whether such tools belong in research, but what standards should govern their use.

Global Edition
BACK TO THE TOP
Copyright 1995 - . All rights reserved. The content (including but not limited to text, photo, multimedia information, etc) published in this site belongs to China Daily Information Co (CDIC). Without written authorization from CDIC, such content shall not be republished or used in any form. Note: Browsers with 1024*768 or higher resolution are suggested for this site.
License for publishing multimedia online 0108263

Registration Number: 130349