A groundbreaking artificial intelligence system developed in China has independently solved a decade-old mathematical problem, marking a significant step toward automating complex research tasks without human oversight. The achievement highlights the potential for AI to transform fields like mathematics by handling intricate proofs and verifications autonomously.
Breakthrough in Automated Mathematical Research
According to a new study, a team from Peking University created an AI that cracked an algebra conjecture first proposed in 2014 by American mathematician Dan Anderson, who passed away in 2022. The system processed decades of mathematical literature to solve the problem and verify its findings entirely on its own, with no human intervention required during the process.
How the AI System Works
The AI framework integrates two key components: a reasoning agent called Rethlas and a formalisation agent named Archon. Rethlas uses a math theorem search engine, Matlas, to explore strategies for solving problems, mimicking the workflow of human mathematicians. When Rethlas generates a potential proof, Archon employs another search tool, LeanSearch, to convert it into a format for an interactive theorem prover called Lean 4.
Lean 4 is a programming language with a vast community-maintained library containing hundreds of thousands of theorems and definitions. This setup allows the AI to rigorously verify proofs, addressing common issues like subtle flaws in human-written proofs or hallucinations in large language models.
Performance and Implications
Researchers reported that the AI solved Anderson's conjecture within 80 hours of runtime, performing tasks faster than any human could. Notably, the system required no mathematical judgment from human operators, though guidance from a mathematician could speed up the process. The team, led by mathematician Dong Bin, emphasized that this work provides a concrete example of how mathematical research can be substantially automated, reducing human effort while ensuring verifiable results.
While AI systems globally are being trained to tackle math problems, they typically need extensive human supervision. This Chinese innovation demonstrates a promising paradigm where informal and formal reasoning systems work together to produce reliable outcomes, potentially accelerating discoveries in fields beyond mathematics.
