In March 1950, RAF wing chief and trained accountant Charles Reep looked at the numbers on the ball. Reep, who became fascinated with the game in the 1930’s and was impressed with the team of Herbert Chapman who was pioneering at Arsenal, returned from the Second World War to find that the systemic change he had seen before was over.
Finally, during a break in the middle of the third round between Swindon Town and Bristol City, where he saw the repeated attacks in vain, Reep’s patience waned. He took a copy of the pencil and began to write down in detail what had happened on the court: He began to calculate the number of passes and shot one of the first attempts to use the data to check the ball.
Seventy years later, the data exchange has come to a low – fans know better xG and all funding, and higher education groups take the numbers of PhD students from the university in search of boundaries. Now, to protect Premier League players Liverpool have teamed up with DeepMind to use creative talent in football. A research paper for the two organizations, published today by a Notes on Artificial Intelligence Research, explains some of the options.
“The timing is good,” said Karl Tuyls, an AI researcher at DeepMind and one of the lead authors of the paper. DeepMind’s partnership with Liverpool stems from his previous position at the city’s university. (DeepMind founder Demis Hassabis is a lifelong Liverpool fan and was an advisor on the study.) The two teams met to discuss where AI could help football players and coaches. Liverpool have also given DeepMind more of every Premier League game the club has played from 2017 to 2019.
In recent years, the amount of data available in football has increased exponentially with the use of sensors, GPS trackers, and computer-based tracking systems. For football teams, AI provides a way to visualize forms that coaches cannot; for DeepMind researchers, football provides a challenging but challenging environment to test their methods. “Games like [soccer] It’s very exciting, because there are so many sponsors, there is competition and cooperation, ”said Tuyls. Unlike chess, or Go, football is uncertain because it is played in the real world.
That doesn’t mean you can’t predict, and it’s one area where AI can be very useful. This paper shows how you can teach the type of data about a particular team and the predictions of how its players will do certain things: If you hit a long ball to the right against Manchester City, for example, Kyle Walker will run a certain way, while John Stones can do something else.
This is known as “ghost throwing” – because some of the cover-ups are real, such as video games – and have a variety of functions. It can be used, for example, to predict the meaning of a system change or how an opponent would play if a player was injured. These are things that coaches can recognize for themselves, and the Tuyls are adamant that the goal is not to develop replacement weapons. “There’s a lot, a lot to be polished, and it’s not really difficult to deal with this data,” he says. “We’re trying to develop technical support.”
As part of the paper, researchers reviewed more than 12,000 sanctions imposed in Europe over the past few days – dividing the players into groups based on how they played, and then using this information to predict where they should have been and whether they could do better. For example, the strikers were more likely to be in the left-hand corner than the midfielders – who did the right thing, and the data showed that the best way to get punters was, perhaps surprisingly, to push their strong side.