Etzioni on AI: What the World Cup tells us about the best role for humans and machines
In soccer, one sideline call can decide who advances and who goes home. But what can you do? Referees are only human.
However, the 2026 World Cup has put the vision of computer and AI in the management team: video review, sensor inside the ball, automatic offside calls, cameras tied to every rafter. And technology has already determined the goal.
On June 15 in Monterrey, Sweden was busy defeating Tunisia when Mattias Svanberg came off the bench and scored with his first touch. The pilot’s flag went up. Offside. The goal was gone, until it ended. The video review was returned, because the ball itself had registered a touch that the human eye missed: a low shot by Alexander Isak that reset the game and left Svanberg aside. However the cameras missed the play. The sensor inside the ball caught it.
How does the ball beat the string? Start with what FIFA has done to the tournament. Sony’s Hawk-Eye supports video review, goal-line decisions, an automatic offside system, and a “last touch” feature that resolves who kicked the ball into a corner.
Chenliang Xu, a computer vision researcher at the University of Rochester, told the university’s news agency that it is “a very complex system that combines many computer vision techniques.” Underneath, that means calibrated cameras, models trained to recognize the ball and players and their posture, and a thin layer of logic that decides when to look.
Player and ball tracking runs on neural networks trained on millions of labeled images, the same list of models behind face unlock and the vision stack in a self-driving car.
Xu compares training to “teaching a child how to see things”: feed the model enough examples and learn what’s important. Sixteen cameras cover each field, so that one angle can be blocked or manipulated, and multiple angles can be combined into a three-dimensional picture of the game. It works the way your eyes do.
“If you block one of your eyes,” Xu said, “it’s very difficult to see depth.” Two eyes see what one eye cannot. So are 16 cameras. The reconstruction takes seconds, and the person signs.
How fast is it? The program is small. According to FIFA, the cameras drop more than 150 million tracking points per match, more data than any all-purpose model can process in real time. Networks are designed for one task, to recognize players and the ball, and are stripped of everything else, which is exactly what makes them fast.
Childhood is also acceptance. The system measures the one thing that a camera and sensor can’t cleanly measure, body position at the instant the ball is hit, and it doesn’t sit on the call that starts so many arguments: whether the player listening was disrupting play. The machine gets the estimate. The referee retains the decision. A good reminder that AI is currently Assistive Intelligence, nothing more.
@media (max-width: 600px) { aside.callout { float: none !important; max-width:100% !important; margin-left:0 !important; margin-right:0 !important; } aside.callout .callout-img { display:none !important; }}
But the quietest AI in this world cup is not on broadcast.
A torn hamstring can end a player’s World Cup, and an opponent’s with it. Long before a game, clubs are pouring data from GPS vests and motion sensors, gear sold by firms like Catapult and Zone7, into models that signal when a player’s workload is turning to injury, sometimes before the athlete feels a thing. It produces no spike in the image and no slow playback. It generates a number that tells the trainer to rest the muscle for the day.
Cameras get the spotlight, but hamstring monitoring keeps players from being, well, hamstrung.
