Google DeepMind wants to make AI a part of soccer

• Google DeepMind launched a prototype AI soccer tactics tool last month.
• TacticAI suggests strategies for corner kicks, based on an analysis of over 7,000 set pieces.
• DeepMind developed the tool with Liverpool FC, known for its data-driven approach to soccer.

Google DeepMind is bidding to bring artificial intelligence (AI) into soccer by launching a new tool to help coaches make better tactical decisions.

The London-based lab unveiled a prototype of TacticAI in March, which it developed as part of a three-year collaboration with English Premier League soccer club Liverpool FC.

DeepMind researchers published their findings in the scientific journal Nature Communications.

Soccer experts involved in the project believe that the AI system, trained on a dataset of more than 7,000 corner kicks from EPL matches, is already better than humans at designing plans for attacking and defending these set pieces.

"What's most exciting about [soccer] for me is that it's a game that lies in between art and science," Zhe Wang, one of Google DeepMind's leads on the TacticAI project, told Business Insider. "There's lots of randomness in a game, but we can still use data to make better decisions."

Corner kicks

Some of the most iconic goals in soccer's recent history have come from corners. 

Former Ivory Coast star Didier Drogba memorably headed in a last-gasp equalizer for Chelsea in the 2012 Champions League Final from a corner, while Liverpool fans voted Divock Origi's strike from a "corner taken quickly," which knocked FC Barcelona out of the same competition in 2019, as the greatest goal in the club's history.

Petar Veličković, a coauthor on the DeepMind project, told BI that it made sense to train TacticAI on data taken from corners because they tend to lead to more structured passages of play.

Soccer is "super unpredictable," he said. "You can never really predict exactly what's going to happen. But what you can do is notice patterns in tactics. Corners are really good for that because they're rigid, frequent, and lead to goalscoring opportunities."

TacticAI can make suggestions to coaches about players' optimal positioning when attacking and defending corners, potentially giving their teams the sorts of marginal gains that tend to make big differences in competitive sports.

Liverpool link-up

DeepMind developed TacticAI with Liverpool, which has won plaudits from legendary "Moneyball" baseball executive Billy Beane.

Since being taken over by Boston Red Sox owners Fenway Sports Group in 2010, the club has hired several people whose CVs you wouldn't typically find in an EPL boardroom — including Ian Graham, who holds a doctorate in physics from Cambridge and who headed up Liverpool's research department until 2023.

Having that knowledge on hand helped the collaboration between DeepMind and Liverpool to run smoothly, according to Veličković.

"For the past 10 years or so, Liverpool have had people on their team with Ph.D.s from Cambridge and Harvard — these are people who are very engineering, data, and research-minded, and they've worked on using statistical analysis for a long time," he said. "So there was already a good, solid foundation we could build on with deep learning."

Liverpool's experts helped DeepMind to judge whether TacticAI's corner routines were effective, while the two organizations have also teamed up to produce a broader study about AI's role in soccer and a system for predicting players' off-camera movements in recent years.

AI and soccer

The release of TacticAI marks the end of that collaboration between DeepMind and Liverpool — but the researchers believe that executives at top-level clubs have only just started to realize that AI could play a transformative role in soccer.

Like corners, other set-piece events, including free-kicks, throw-ins, and penalties, offer up the sort of structured data that AI tends to thrive on and could be the next area that researchers in the field look into, they said.

Some purists worry that technological interventions like these might one day take some of the joy out of the so-called "Beautiful Game" — but the researchers emphasized that they've tried to create a tool that'll make coaches' lives easier rather than put them out of a job.

"We're not trying to replace human coaches. Instead, we want to help them make faster and better decisions to help them see patterns and find possible improvements through AI," Wang said.

"Ultimately, it'll still be the coaches who make the decisions," he added. "This is just a tool to help them make those decisions more comprehensively."