A recent study indicates that an artificial intelligence model can predict the outcomes of corner kicks in football matches and assist coaches in designing tactics to increase or decrease the probability of player shots. On March 19, the findings were published in Nature Communications.
Petar Veli?kovi? and colleagues from DeepMind, a subsidiary of Google, developed a tool called TacticAI, as part of a three-year research collaboration with the Liverpool Football Club in the UK.
Corner kicks occur when the ball crosses the goal line and goes out of bounds, presenting a prime scoring opportunity for the attacking team. Consequently, football coaches devise detailed plans for various corner kick scenarios, which players can study before matches.
TacticAI was trained on data from 7,176 corner kicks during the 2020-2021 English Premier League season, including each player's running positions as well as their height and weight. It learned to predict which player would make the first touch after a corner kick. In testing, the player predicted by TacticAI ranked among the top 3 candidates for the first touch 78% of the time.
Coaches can use artificial intelligence to generate attacking or defensive corner kick tactics, maximizing or minimizing the chances of specific players receiving the ball and the team taking shots. TacticAI analyzes real examples of similar corner kicks, then provides recommendations on how to alter tactics to achieve the desired outcome.
In a blind test, football experts from Liverpool Football Club were unable to distinguish between tactics generated by artificial intelligence and those designed by humans, with 90% of their preferences leaning towards the AI-generated tactics.
Veli?kovi? emphasized that while TacticAI is capable, it will never replace human coaches. He stated, "We support AI systems that enhance human capabilities, giving them more time for creative work rather than replacing them."
Veli?kovi? noted broader applications of this research beyond sports. "If we can simulate football matches, we can better simulate several aspects of human psychology," he said. "As AI capabilities grow, they will need a better understanding of the world, especially in uncertain situations. Our system can make decisions and suggestions in uncertainty. We believe these skills can be applied to future AI systems, making it a great testing ground."
For more information, see the related paper: https://doi.org/10.1038/s41467-024-45965-x
