DeepMind AI learns to play Football (harder than Go?)

DeepMind seemed to have reached the highest peaks by beating (under the name of AlphaGo) the former Go world champion Lee Sedol, but now artificial intelligence is facing an even tougher challenge… football! Can the most popular sport in the world be mastered by a cold-blooded AI, can the passes of a Mbappé be translated into algorithms?



More prosaically, Google engineers used the system Neural Probabilistic Motor Primitives (NPMP) in order to analyze videos of football matches, all mixed with records of motion capture recovered from players in action. The objective of this digital grub? Allow DeepMind to reproduce the movements of the players in a 3D simulator, while of course keeping the final objective which is to push the ball into the back of the opposing net. A little of machine-learning later (reinforcement learning), and now DeepMind is ready to manipulate 3D soccer player avatars like a new silicon R7.

Well, if you’ve watched the video above, you know that DeepMind is currently closer to the honor division player; but no matter, the AI ​​is moving fast, and as always will eventually master its subject and therefore plant goals that even Papin could not have dreamed of. Eventually, we begin to dream (no) of this same AI coupled with humanoid robots and playing against a team of flesh and blood players. In 2050 maybe?

Leave a Comment