English Premier League giants Liverpool are reportedly using a new artificial intelligence assistant developed by Google DeepMind to help with tactics.
It's not unusual for football teams to analyze games, but Liverpool may be a step ahead of other clubs in using machine learning technology to understand tactics.
The club and DeepMind have been collaborating for nearly three years and TacticAI is the result of that partnership.
WHAT IS TACTICAI AND HOW DOES IT WORK?
The system works by analyzing corner kicks, which are set plays awarded after the ball has gone out of play, giving the other team a chance to score from corners.
Such kicks and how players move near the goal is a matter of extensive planning, depending on the firepower of the strikers and the quality of the defense. Where the kicker directs the ball is often determined by the other team's defensive weaknesses.
TRAINED WITH DATA FROM 7176 CORNER KICKS
TacticAI was trained on data from 7176 corner kicks used during the Premier League season from 2020 to 2021.
78 PERCENT ACCURATE PREDICTION
Using predictive and generative AI, TacticAI learned to predict where the ball would go and was very good at it, guessing correctly 78 percent of the time.
"POTENTIAL TO REVOLUTIONIZE SPORTS"
"We developed and evaluated TacticAI with Liverpool's experts as part of a very long piece of research. TacticAI's recommendations are preferred by human expert evaluators 90 percent of the time over tactical layouts seen in practice," DeepMind said in a press release. TacticAI demonstrates the potential of assistive AI techniques to revolutionize sport for players, coaches and fans."
DeepMind said TacticAI's powers can be applied to more areas of the game than corner kicks. The company said it could be used for all set-pieces and even general play throughout the game.
Petar Velickovic, a research fellow at Google DeepMind, said the technology could also be adopted by soccer, hockey or basketball teams.
"The best clubs are always looking for an advantage, and I think our results show that techniques like this will become part of modern soccer in the future," Velickovic told MIT Technology Review. "As long as it's a team-based sport where you believe it would be useful to model the relationships between players and you have a source of data, it's feasible."
