Think the player A.I. in FIFA ‘19 was something special? You haven’t seen anything yet! That’s because search giant Google is developing its own soccer-playing artificial intelligence. And, if the company’s history with machine intelligence is anything to go by, it’ll be something quite special.
In the abstract for a paper describing the work, the researchers note that: “Recent progress in the field of reinforcement learning has been accelerated by virtual learning environments such as video games, where novel algorithms and ideas can be quickly tested in a safe and reproducible manner. We introduce the Google Research Football Environment, a new reinforcement learning environment where agents are trained to play football in an advanced, physics-based 3D simulator.”
Google’s history with game-playing A.I. focuses most heavily on its deep learning subsidiary, DeepMind. DeepMind famously created a reinforcement learning-based A.I. that was able to learn to play (and master) classic Atari games. It could do this with no explicit instruction, and only the on-screen data to formulate its winning strategies. Reinforcement learning is a type of A.I. that focuses on the actions that should be taken in an environment to maximize a certain reward.
Soccer is, of course, more complex than a 2D game like Pong. In Google’s physics-accurate virtual Football Environment, A.I. agents will need to “control their players, learn how to pass in between them and how to overcome their opponent’s defense in order to score goals.” The team is testing three reinforcement learning algorithms in a variety of soccer-related challenges. It will play against both human and machine players.
While there are plenty of soccer video games which can beat many human players (as the higher difficulty levels will attest), Google’s project goes further. It’s not only the attack or defense part of soccer that its bots will need to master, but also things like high-level strategy and when it’s absolutely optimal to pass, shoot, or make other moves.
Google researchers aren’t the only ones interested in getting machines to play soccer. Since 1996, the organizers of the robot competition RoboCup have been trying to teach robots to play the world’s most popular sport. “The ultimate goal of Robocup is to develop humanoid soccer-playing robots that can beat the FIFA world champion team,” Gerhard Kraetzschmar, general chair of the RoboCup, previously told Digital Trends. “We hope to reach that goal by 2050.”
Nor is soccer the only sport A.I. experts want to teach computers to understand. As Digital Trends detailed recently, IBM has built a tennis-appreciating A.I., which it deployed at this year’s Wimbledon to create automated highlights of the most exciting bits of each match.
