On YouTube there are a lot of FIFA tutorials that teach you how to score a goal on all corner kicks. You just have to choose the kicker and the finisher very well and press a few keys on the PlayStation controller. Now, a team of scientists has just published an article in the scientific journal Naturein which they assure that, with the help of Artificial Intelligence, it can also be done in real life.
To do this, they trained a system, called TacticAI, with an architecture of algorithms and formulas to which they added 7,176 corner kicks taken by the Liverpool team throughout its history. Including that of Oakley Cannonier, the ball boy from the famous Liverpool-Bara semi-final. Which means introducing variables as complicated as catching a defense unusually clueless, and having the ball boy pass the ball at full speed to Alexander Arnold while Divock Origi wait in the small area.
In real life, corner kick routines are prepared before each match, so a system that helps analyze and improve scoring chances will be beneficial to help human experts, the researchers explain. Their results were placed in the hands of the technical team of Jurgen Klopp which determined that TacticAI’s ideas were better than theirs for nine out of 10 corner kicks, a truly chilling fact.
geometric deep learning
Zhe Wang, Peter the Great, Karl Tuylsthree AI experts, used so-called geometric deep learning, which consists of identifying key strategic patterns to produce predictive and generative results, capable of modifying and learning on the fly by themselves.
With the 7,176 corners, TacticAI was able to accurately predict which player should receive the ball at each moment, in what position, at what speed he should move, and where he should shoot once he received it. But also, and at the same time, other positive results with player configurations and ball movements alternative to that initial one.
The most striking thing was the machine’s ability to generate realistic tactical configurations indistinguishable from real-world scenarios, according to the evaluations of a group of five experts who tested it: three data analysts, a video analyst and a team assistant. red.
beyond a simple analyst
The researchers believe that their work could lay the foundations for an imminent generation of AI assistants that will also sit on the benches, and that already go far beyond a simple data analyst, predictor of performance or the chances of success of a company. play or a shot. They will help coaches determine optimal player configurations, asking the machine directly for any possibilities, and develop countertactics on the fly that maximize the chances of winning, they explain in their article.
And they don’t stay there. The authors also suggest that the method could be extended to other set-piece plays, such as fouls and throw-ins, but also to other team sports that involve suspended play situations.