Geoffrey Hinton, the researcher widely regarded as the “Godfather of AI,” thinks machines will be better at mathematics than any living human within the next decade. For a field that has defined human intellectual achievement for millennia, that’s a rather bold claim. But Hinton isn’t exactly known for hedging.
Speaking at the Sana AI Summit, Hinton laid out a deceptively simple argument: mathematics is, at its core, a closed system. Like chess. Like Go. And we already know how those stories ended for human players.
Math as a game board
Here’s the thing about closed systems. They have defined rules, clear objectives, and verifiable outcomes. When DeepMind’s AlphaGo defeated the world’s best Go player back in 2016, it did so partly by discovering moves that no human had considered in thousands of years of play. Hinton’s argument is that mathematics operates on fundamentally similar principles.
In English: if you give an AI system a set of axioms and let it explore conjectures and proofs on its own, it doesn’t need human-generated examples to get better. It can teach itself. The same way AlphaZero learned chess by playing millions of games against itself, AI could potentially navigate the landscape of mathematical proof by brute-force exploration combined with pattern recognition.
