How AI-Generated Proofs Are Reshaping Mathematics

Advanced artificial intelligence models are transforming mathematics at an astonishing rate. In recent months, AI-powered proofs—logical arguments that demonstrate why a mathematical result is correct—have been produced for hundreds of long-standing problems.

It may take mathematicians months to fully understand the underlying arguments and implications of these automatically derived solutions. As a mathematician who uses AI tools in my own research, I can see that the field is undergoing a fundamental change. We now urgently need to establish an era of “Mathematics 2.0,” in which a lack of proofs is no longer a defining feature of the discipline.

The challenges of AI-generated mathematical proofs

The challenges associated with this new approach to mathematics are already evident. Many results produced by AI are not explained clearly enough to meet publication standards. Some researchers using AI tools move on to the next problem once they have a minimally provable result, leaving proofs that are difficult for other mathematicians to evaluate. It also remains unclear how extensively AI models draw on mathematicians’ previous work without proper attribution.

The centuries-old practice of openly sharing mathematical problems is under serious threat. Mathematicians increasingly fear that revealing the direction of their research could allow rivals with cutting-edge AI models to scoop their work. By solving these problems, graduate students and young mathematicians have traditionally developed their skills and gained early recognition from their peers.

On October 6, technology company OpenAI shared progress toward solving hundreds of such problems online at once. The development disrupted countless research programs overnight. Learn more about AI’s progress in solving mathematical problems.

When AI produces mathematics faster than humans can understand it

Perhaps the most counterintuitive result is that humans are falling behind in understanding mathematics because AI systems can produce results faster than mathematicians can digest and absorb them.

In pure mathematics, this “indigestion of proofs” may not be an immediate problem. However, if AI-generated mathematical models are deployed in the real world without a thorough understanding of how they work, the consequences could be serious.

Complex AI models are already being used to some extent to decide who qualifies for bank loans and to measure financial risk in markets. If these models become too complex for humans to identify when they will malfunction, the risk of unforeseen errors will increase—and those errors could affect the general public.

‘It’s incredible’: How AI will transform mathematics

How AI can improve mathematical understanding

These negative effects are not inevitable. When used creatively and supported by the right incentives, AI tools can enhance human understanding. It is still possible to change direction by fundamentally reevaluating traditional incentives in mathematics and introducing more responsible practices for using AI.