On March 20th, Google's AI research company DeepMind announced a collaboration with Liverpool Football Club to develop TacticAI, an AI model that predicts outcomes of corner kicks in soccer and helps coaches design attacking and defensive corner kick strategies to maximize their players' chances of getting a touch or shot or reducing the opposition's chances of a shot.
The partnership is part of a three-year research collaboration between DeepMind's team led by Petar Veličković and Liverpool.
In soccer, a corner kick is awarded to the attacking team when the ball goes out of play over the goal line, having last been touched by a defender, but not as a goal. Corner kicks are often a valuable scoring opportunity for the attacking team, and coaches typically develop extensive tactics for different scenarios, which they brief their players on before the match.
TacticAI was trained on data from 7,176 corners taken in the 20/21 Premier League season, including data on players' height, weight, and their precise location on the pitch at any moment. The model is able to predict which player will make the first contact with the ball. In real-world tests, 78 percent of the time, the first player to touch the ball was among the top three players predicted by TacticAI.
The model can be used by coaches to develop strategies for taking and defending corners, to either maximize or minimize the chances of a particular player getting a touch, and therefore influencing the likelihood of a shot. By analyzing real-world corner kick situations, the AI learns patterns and can suggest tactical tweaks to get the desired outcome.
In a blind test, Liverpool's football experts were unable to reliably distinguish between tactics generated by the AI and tactics designed by human coaches, preferring the AI option in 90 percent of cases.
Despite TacticAI's impressive performance, Veličković stresses that such AI is not about putting coaches out of a job. "We strongly advocate for AI systems that augment human capabilities and that free up humans to focus on more creative work, rather than a system that would replace them," he says.
Veličković believes that the implications of this research extend far beyond sports. "If we can simulate soccer, then we can better understand many aspects of human cognition," he says. "As AI becomes more powerful, it will need to reason about the world and make decisions under uncertainty. Our system makes decisions and recommendations in situations where there is uncertainty, and we think these skills will transfer to future AI systems, so it's a good testbed."
