AI keeps cracking unsolved math problems, and mathematicians have mixed feelings

Editor's take

OpenAI's GPT-5.6 Pro has demonstrated a significant capability by independently proving two complex mathematical conjectures that had previously eluded human mathematicians. This development underscores the growing potential of large language models not just for information retrieval or content generation, but for genuine scientific discovery. The implications extend beyond abstract mathematics, potentially accelerating research in fields reliant on theoretical underpinnings, from physics to computer science.

The mixed reception among mathematicians, as highlighted by figures like Timothy Gowers, points to a fundamental shift in the research landscape. While AI offers powerful new tools for exploration and verification, it also raises questions about the future role of human intuition, creativity, and the very definition of mathematical expertise. The speed at which AI can generate proofs, as seen with GPT-5.6 Pro, challenges traditional timelines and collaborative models.

Future developments to monitor include the broader adoption of such AI tools by research institutions and whether AI-generated proofs will be subject to the same rigorous peer-review processes. The emergence of specialized AI systems for specific mathematical domains, or the development of AI that can collaborate more seamlessly with human researchers, will be crucial indicators of how this field evolves. The ability of AI to discover novel mathematical concepts, not just solve existing problems, will be a key differentiator.