The Path to Mathematical Superintelligence | Tudor Achim | TED

Summary

Mathematics has been humanity's most unreasonably effective tool for understanding the universe, built on a 4,000-year-old process of human creativity, communication, and peer verification. However, this system is reaching a breaking point as AI now generates mathematical proofs faster than humans can verify them, creating a verification bottleneck that threatens reliable mathematical discovery. The solution lies in upgrading mathematics to use formal languages like Lean, where AI writes proofs that computers can automatically verify with absolute certainty, enabling a new era of human-AI collaboration rather than replacement.

The speaker traces this vision back to 17th-century mathematician Gottfried Leibniz, who dreamed of a universal characteristic system with a perfect logical language, comprehensive encyclopedia of verified knowledge, and mechanical engine of reason. In 2025, all three components now exist: Lean provides the logical language, Mathlib offers the encyclopedia of proven mathematics, and AI serves as the engine of reason. This formal mathematics approach has already demonstrated success at the International Math Olympiad, where automated systems solved problems in ways computers could verify without human review.

Key takeaways

• Adopt formal mathematics (Lean) for AI-generated proofs to enable computer verification instead of relying on human review
• Shift human mathematicians from tedious proof-checking toward creative roles: proposing conjectures, asking questions, and guiding AI exploration
• Implement Mathlib or similar computational proof libraries as the foundation for human-AI mathematical collaboration
• Recognize that formal verification eliminates the human bottleneck by replacing subjective review with objective computational certainty
• Prepare for exponential growth in AI mathematical output by establishing formal verification systems before the verification gap becomes critical