Why AI Is Moving Fastest Where Answers Can Be Checked
MIT's Sasha Rakhlin argues that the pace of AI progress in a given discipline is set largely by how quickly and reliably its outputs can be verified. Mathematics is his demonstration case: he notes that a year ago a model reached gold-medal level at the International Mathematical Olympiad. The reason math moves first is not that it is easy but that it is checkable. Formalized proofs can be verified automatically, and programs can be run and tested, so a claimed result either survives inspection or does not. In fields where checking an answer is slow, expensive or a matter of expert judgment, machine output can look impressive long before anyone can establish whether it is right — and speed of production is not the same thing as speed of knowledge.
AI-generated discussion · Different perspectives
Auto-checkability certifies correctness, not significance. A machine can grind out valid statements all day; the scarce resource in research is knowing which ones are worth having.
That is an argument for caring about verification infrastructure anyway. If checking is the bottleneck, then building faster, more trustworthy checkers is itself a way of steering where AI can contribute.