Can AI discover major mathematical results on its own?
Breakthrough in sight
Why it matters
Today's AI systems often cannot guarantee that their mathematical reasoning is correct. AI that finds proofs checked by a computer could become a reliable tool for solving hard mathematical problems.
What we know
In 2025, AI systems reached gold-medal level at the International Mathematical Olympiad.
What blocks us
Research needs new concepts, not just solving problems someone else posed.
Current leads
Formal proofs in the Lean proof assistant, collaboration between mathematicians and AI.
Sub-questions
Can AI reliably translate ordinary mathematical writing into formal, machine-checkable statements (autoformalization)?
Can AI propose new conjectures and concepts, not just solve problems written by humans?
How can AI learn formal mathematics when so few human-written formal proofs exist to train on?