Google's new AI tool for corner tactics has the approval of Liverpool analysts -- but the players have not tried it yet
Google's new AI tool for corner tactics has the approval of Liverpool analysts -- but the players have not tried it yet AFP

Liverpool football analysts have given their seal of approval to a new artificial intelligence tool from Google that suggests the best tactics for corner kicks, researchers said Tuesday.

However no Liverpool players -- or any humans -- have tried out the AI assistant's suggestions, and an outside expert cautioned that the tool's value could be demonstrated only on the field.

Many wealthy clubs across the world already deploy experts to comb through video and data to carefully plan corner kick scenarios ahead of matches.

Google DeepMind, the AI arm of the US tech giant, also targeted corners for its new TacticAI system in research published in the journal Nature Communications.

For the study, the researchers enlisted the help of five experts from Liverpool, who are currently second in the English Premier League.

TacticAI offers suggestion to coaches on exactly where the players should be positioned while attacking and defending corners, often involving very small adjustments in placement and speed.

Across 50 different corner set-ups, the experts were asked to choose between an AI suggestion or tactics previously used during practice -- without knowing which was which.

The Liverpool analysts picked TacticAI's suggestion in 45 cases, meaning they preferred its advice 90 percent of the time, according to the study.

Google DeepMind research scientist and study co-author Petar Velickovic told AFP the tool is the first AI assistant "capable of providing both useful and practical suggestions" indistinguishable from human tactics in sports.

The model uses data on the position, speed, weight and height of players in the Premier League.

Compared to the billions or trillions of data points AI models are usually trained on, only "fairly rudimentary" information is available for corners, Velickovic said.

There are an average of around 10 corners during a Premier League match.

Velickovic said the model's "secret sauce" was geometric deep learning, an AI technique that allowed the researchers to flip the field to increase the amount of data.

While the study focused on corners, the researchers said the system could be expanded to other parts of football -- and even other sports.

It is not the first collaboration between Liverpool and DeepMind, which started life as a British startup before being bought by Google.

DeepMind CEO Demis Hassabis is a "big fan" of Liverpool, the study's lead author Zhe Wang said.

Velickovic said that while Liverpool's players have not yet attempted a TacticAI corner, "we will probably see clubs trying out techniques like this" in the future.

Of course, Liverpool are no slouch at corners already. In an online poll, Liverpool fans voted a corner scored by Divock Origi -- and crossed in by Trent Alexander-Arnold -- in the 2019 Champions League semi-final as the greatest ever scored for the club.

Andy Harland, a sports technology professor at the UK's Loughborough University who was not involved in the research, was sceptical of the worth of the AI tool.

Whether TacticAI "is of any value will be realised by whoever uses it over future matches," he told AFP.

"I suspect it is just another contribution to an already overcrowded space," Harland said.

And "if everyone is using their own AI" in the future, they might all cancel each other out, he added.