Fully autonomous AI mathematicians remain far off. Terence Tao, the world's foremost living mathematician, told Dwarkesh Patel that hybrid teams of humans and AI will dominate the field for far longer than the current hype cycle suggests - and that the transition will require breakthroughs nobody has made yet.
"I do believe that hybrid human plus AIs will dominate mathematics for a lot longer," Tao said in a conversation reported by "Hvylya", citing the Dwarkesh Podcast. "AIs currently are very good at certain things, but really terrible at others. It feels like we don't have all the ingredients to really have a truly satisfactory replacement for all intellectual tasks."
The core limitation, according to Tao, is that AI excels at breadth while humans provide depth. Models can try thousands of approaches simultaneously and never miss an obscure technique in the literature. But they cannot build cumulative understanding - the kind of slow, interactive process where two mathematicians trade half-formed ideas, test them, fail, modify them, and gradually map out what works. "There isn't this cumulative process which is built up interactively," Tao said. "It seems to be a lot more trial and error and just repetition: brute force."
Tao pointed to a concrete gap. When AI hits a dead end in a proof, it cannot identify the right intermediate step and hold position there. "What they can't do is jump a little bit, reach some handhold, stay there, pull other people up, and then try to jump from there," he said. Human mathematicians do this naturally - it is how partial progress compounds into eventual solutions.
The practical implication, Tao argued, is that mathematics needs to reorganize itself to exploit AI's strength. "We should have a lot more effort in creating very broad classes of problems to work on rather than one or two really deep, important problems," he said. AI could map out entire unexplored fields by making all the easy observations first, then flag specific islands of difficulty for human experts to tackle. "I see very much a future of very complementary science," he said. "Eventually, you would hope to get both breadth and depth."
When Patel asked directly when most mathematical progress would come from autonomous AI, Tao declined to give a year. "It will require some additional breakthroughs beyond what we already have, so it's going to be stochastic," he said. He acknowledged that within a decade, much of what math students currently spend their time on could be handled by AI - but predicted that researchers would simply move on to harder problems, just as they did when calculators and computer algebra systems arrived.
Also read: "Hvylya" examined why the Financial Times' Martin Wolf warned that AI poses a deeper threat to the educated class than any previous technological shift.