We’re just a few years into the AI revolution, but AI systems are already improving decades-old computer science algorithms.
Google’s AlphaEvolve AI, its latest coding agent for algorithm discovery, has improved on a 56-year-old algorithm for matrix multiplication. “Provided with a minimal code skeleton for a computer program, AlphaEvolve designed many components of a novel gradient-based optimization procedure that discovered multiple new algorithms for matrix multiplication, a fundamental problem in computer science,” Google Deepmind wrote in a blogpost.
“AlphaEvolve’s procedure found an algorithm to multiply 4×4 complex-valued matrices using 48 scalar multiplications, improving upon Strassen’s 1969 algorithm that was previously known as the best in this setting. This finding demonstrates a significant advance over our previous work, AlphaTensor, which specialized in matrix multiplication algorithms, and for 4×4 matrices, only found improvements for binary arithmetic,” ir added.
Matrix multiplication is one of the fundamental operations in computer science, and is used extensively in AI applications. It involves multiplying two matrices, or two sets of numbers. The multiplications happens through specific rules by normally multiplying numbers, known as scalar multiplication, within the two sets of numbers. Thus far, it needed 49 normal “scalar” multiplications to multiply two 4×4 matrices. Google’s AlphaEvolve, though, came up with an approach that is able to multiply these two matrices with just 48 multiplications. This is a tangible improvement over the state-of-the-art existing method, and if deployed at scale, can help improve the performance of virtually all AI applications.
And AI systems coming up with new algorithms that improve over decades-old state-of-the-art approaches could be the beginning of the great scientific acceleration that’s been predicted by players from leading labs including Google Deepmind, OpenAI and Anthropic. Once AI systems become as capable as human researchers, they can be endlessly replicated, and deployed towards solving problems without breaks that their human counterparts might need. The sheer scale of their deployment — the only cost of running an AI researcher will eventually be the electricity costs in running them — will mean that these researchers will inevitably come up with new discoveries. These new discoveries can in turn be used to come up with even better AI assistants and newer discoveries, resulting in a flywheel of scientific progress. And with Google’s AlphaEvolve already coming up with new research and algorithms, the AI-assisted flywheels of scientific progress seem to have already begun turning.