There’s plenty of interest in how AI could help accelerate scientific research, but there’s a particular field it could especially excel at.
Geoffrey Hinton, widely regarded as one of the “godfathers of AI” for his pioneering work on neural networks, recently shared his perspective on where artificial intelligence will make its most dramatic scientific breakthroughs. Hinton pointed to mathematics as the domain where AI superiority over humans seems not just possible, but inevitable—and perhaps surprisingly soon.
“I agree with Demis,” Hinton said, referring to Demis Hassabis, the leader of DeepMind, “who for many years has said AI is going to be very important for making scientific progress. It’s going to make scientific discoveries. There’s one area in which that’s particularly easy, which is mathematics, because mathematics is a closed system.”
The concept of mathematics as a “closed system” is central to Hinton’s argument. “You’re going to get AIs that play mathematics,” he explained. “That is, they ask themselves, I wonder if I could prove this. I wonder if I could prove that. And because it’s a closed system, they can just try things out and see if they can prove them. I think AI will get much better at mathematics than people, maybe in the next 10 years or so.”
Hinton drew a compelling parallel to AI’s conquest of games like Go and chess. “Within mathematics it’s much like things like Go or chess that are closed systems with rules where they can generate their own training data,” he said. “When they first taught Go to AIs, they would mimic the moves of human experts. And there’s of course a limitation to that, which is you run out of human expert moves and humans are only so good.”
The breakthrough, Hinton noted, came with Monte Carlo rollout methods. “They got what they call Monte Carlo rollout where you say, if I go here, he goes there, I go here, he goes there. Oh what? That’s bad for me. And you can learn from the Monte Carlo rollout, and you no longer need humans to tell you what good moves are. You can figure it out. It’ll be the same in mathematics and we’ll get mathematical systems much better than people I believe.”
Hinton’s prediction aligns with recent developments in AI-powered mathematics. DeepMind’s AlphaGeometry has already demonstrated the ability to solve complex geometry problems at near-Olympiad level, while other systems are making progress on formal theorem proving. Robinhood CEO Vlad Tenev’s math startup Harmonic says it has solved an Erdos problem that was open for 30 years, and other companies are using Lean to build proofs. The key advantage, as Hinton emphasizes, is that mathematics operates within defined rules and axioms—a perfect sandbox for AI systems to explore, experiment, and learn without external constraints. Unlike fields that require physical experimentation or real-world validation, mathematical proofs can be verified instantly and definitively. This means AI systems could potentially explore vast spaces of mathematical possibilities, proposing and testing conjectures at scales impossible for human mathematicians. If Hinton’s timeline holds, the next decade could see AI not just assisting mathematicians, but fundamentally reshaping how mathematical research is conducted—discovering theorems, identifying patterns, and solving problems that have eluded human minds for generations.