Two researchers beat DeepMind’s artificial intelligence in math

Listen to the audio version of the article

Jakob Moosbauer and Manuel Kauers are the two Austrian researchers who have found a more efficient way to perform matrix multiplication, breaking a record set just a week ago by Google’s artificial intelligence company DeepMind. Earlier this month Alpha Tensor – as the DeepMind AI system is called – had discovered a faster way to solve a fundamental math problem in computing, breaking a record that has spanned more than 50 years. The study from the University of Linz has not yet passed the Peer Review and you can find it published in the archive of scientific articles Arxiv. There are those who hailed the “enterprise” as a revenge of the man on the car. But in reality not like this and in any case it is not a race.

Why is matrix multiplication important?

Solving these fundamental computational tasks are particularly important for the advancement of artificial intelligence systems. An improvement in efficiency means allowing the machines to save time and computing capacity which translate into better performance and above all in better energy efficiency.

Because it’s not a man-machine race

Until a few days ago, the fastest method to multiply two matrices four by four together was the one devised in 1969 by the German mathematician Volker Strassen to multiply two matrices four by four together. The basic high school method requires 64 steps; Strassen requires 49 steps. AlphaTensor found a way to do this in 47 steps. The two researchers claim to have reduced that count by one but claim to have based themselves on the work done by DeepMind. In their article, Kauers and Moosbauer write: “This solution was obtained from the scheme [dei ricercatori di DeepMind] applying a sequence of transformations that lead to a pattern from which a multiplication could be eliminated. ”The human solution would have been inspired by the scientific article published by human researchers who used Deep Mind’s artificial intelligence system. If so, it would be a further confirmation of how the “race” between men and machines brings advantages for youtti.