Claude does not solve Riemann, but achieves an unexpected mathematical breakthrough

The Riemann hypothesis, formulated in 1859, remains one of the most famous open problems in mathematics. For over a century and a half, countless mathematicians have attempted to prove or disprove it without success. Now, artificial intelligence has entered the fray, and while it has not achieved the feat, it has accomplished something that has surprised the community: a notable breakthrough in a derivative result.

What happened?

Anthropic, the company behind Claude, has published a report detailing how an experimental model of its AI tackled the Riemann hypothesis. The experiment, driven by Jarred Sumner, an employee without an advanced background in mathematics, consisted of asking the model to try to solve the problem. After generating and testing 650 failed ideas, the model did not achieve the main goal, but during the process, it found two sub-agents that produced relevant results for a known bound in one of the hypothesis's derivative results.

That bound, which had been advancing slowly for decades, improved drastically: it went from 41.6% to 67.2%. The finding has been published in a technical paper, and the code used has been shared on GitHub, allowing the mathematical community to review and validate the work.

Why is it important?

This result is relevant for several reasons. First, it demonstrates that AI can tackle highly complex mathematical problems and generate breakthroughs in derivative areas, even if it does not solve the main problem. Second, the method employed by Claude, which includes generating numerous hypotheses and verifying them through Python scripts, suggests a new form of mathematical exploration based on computational brute force and algorithmic creativity.

Furthermore, the fact that an employee without specific training in advanced mathematics was able to direct this process with simple instructions like 'keep going' or 'believe in yourself' raises questions about the role of humans in AI-assisted research. Is it necessary to be an expert in the field to guide these systems?

Consequences and debates

Claude's progress on the Riemann hypothesis bound does not mean that AI is close to solving the original problem. Anthropic is clear about this: it does not expect the techniques used in this experiment to lead to a proof. However, the secondary achievement has been celebrated as a sign of AI's potential to accelerate the pace of mathematical discovery.

Nevertheless, it has also generated skepticism. Some experts, such as Professor Ethan Mollick, have questioned the narrative that motivational prompts were key. Instead, they suggest that the success was due to the model's ability to consume more compute and persist in the search, something that does not require emotional reinforcement.

This debate reflects a broader tension in the field of AI: to what extent are results attributable to the model's intelligence and to what extent are they the product of computational power and persistence? The answer is likely a combination of both, but the scientific community does not yet have a consensus.

What should readers know?

For the general public, this event is a sign that AI can be a powerful tool for scientific research, but also that its capabilities have limits. We should not expect AI to solve humanity's great problems on its own, but it can help us advance them in unexpected ways.

For companies and professionals, the lesson is that AI can be useful even when it does not meet the initial goal. The ability to explore multiple hypotheses and generate partial results can be valuable in fields such as market research, product development, or process optimization.

In summary, Claude's attempt to solve the Riemann hypothesis failed in its main objective, but it achieved a secondary breakthrough that invites reflection. AI has not proven to be a genius mathematician, but it has proven to be a tireless assistant that can help humans explore unknown territories.

AI has not proven to be a genius mathematician, but it has proven to be a tireless assistant that can help humans explore unknown territories.

Reactions and context

The mathematical community has received the result with interest, albeit with caution. Some researchers have pointed out that the improvement of the bound is a technical achievement, but its impact on solving the Riemann hypothesis is marginal. Others have highlighted Anthropic's transparency in publishing the process and the code, which allows for independent verification.

This case joins other recent instances where AI has contributed to scientific breakthroughs, such as the discovery of new materials or the prediction of protein structures. However, it also serves as a reminder that AI still cannot replace human intuition and creativity in fundamental research.

Implications for the future

Anthropic's experiment raises questions about how to integrate AI into the mathematician's workflow. If models can generate and test thousands of hypotheses in a short time, humans could focus on interpreting the results and designing new strategies. This could accelerate the pace of discoveries in mathematics and other sciences.

But there are also risks. Reliance on AI for research could lead to a homogenization of approaches or the uncritical acceptance of results generated by algorithms. Therefore, it is essential to maintain rigorous human oversight and foster collaboration between mathematicians and AI specialists.

Ultimately, Claude's attempt is a milestone in the history of AI applied to mathematics, although not for the reason expected. It reminds us that innovation often arises from failures and that AI, when used well, can be a valuable ally in the pursuit of knowledge.