For Zhe Wang of Google DeepMind, an artificial intelligence (AI) company, corner kicks are like a game of chess. Partly because in both cases the opposing sides are ready to respond to a single, inevitable move. But undoubtedly also because they too can be revolutionized by artificial intelligence.
Artificial intelligence models develop where there is a lot of data. Football more than meets this requirement. Elite players wear vests that measure heart rate, position, speed and force exerted; team analysts watch hours of footage to calculate ball possession percentage and the number of passes, shots and goals. In a paper published March 19 in Nature Communications, Mr. Wang and his colleagues worked with Liverpool Football Club staff to feed this data into a statistical model known as a graph neural network (GNN). They were then able to apply this model to predict which player on the field would make contact with the ball first, with an accuracy comparable to human experts.
Coaches everywhere should be hovering in front of their benches excited about this news. During the standard 90 minutes of a football match, it is occasional to encounter the same situation twice. However, corner kicks, which are extremely repeatable, are the subject of hours of specialized training. And doing well at them pays off. Arsenal, who were top of the Premier League at the time of publication, have scored 13 goals from corners out of 70 goals for the season. A tiny advantage in this part of the game can make a huge difference.
DeepMind’s GNN service is based on establishing statistical relationships between data collected from all 22 players during 7,176 corner kicks. Ultimately, each corner procedure was represented as a vector in a 352-dimensional space: impossible to visualize by a human, but simple to process by a computer. Corners that developed similarly curled close together in this space, allowing the model to predict fresh set pieces.
The model was also able to apply this analysis to propose fresh tactics. Assessments by five Liverpool experts found that the AI-generated corner kick tactics were as good as those suggested by human coaches. In fact, after showing 50 pairs of corners, one real and one with AI-suggested improvements, 90% of the AI’s suggestions were approved by a majority of the judges.
Mr Wang, who admits he is “not a football fan”, says the sport provides a secure and controlled testing ground for developing useful artificial intelligence technologies that could one day be used in health care or defense. After all, football is not a matter of life and death, it is, as Bill Shankly, former Liverpool manager, once noted, much more crucial.
© 2024, The Economist Newspaper Constrained. All rights reserved. From The Economist, published under license. Original content can be found at www.economist.com
Posted: May 23, 2024, 6:00 PM EST
