The seventh tier of English football, and the sleepy, semi-rural surrounds of Barnet in London’s northern outskirts, is not where you’d expect to find the testing grounds for a tech revolution that could change the face of professional sports as we know it.
There is nothing particularly remarkable about Wingate & Finchley FC, a club that plays in the Isthmian League’s Premier Division, and never in front of more than a few hundred people. But in 2019, this semi-pro outfit took a brave leap into the great unknown by appointing a piece of software as first-team assistant coach.
Wingate & Finchley, a non-league football team in London, used an AI assistant coach in 2019.
Next to manager Dave Norman in the dugout was a small “smart” speaker – a rebadged Amazon Alexa – which was plugged into a computer program analysing billions of different data points extracted from recent matches. It meant that, for their clash against relegation rivals Whitehawk, Norman and his coaching staff were able to ask what sort of formation they should use, based on their opposition and their own recent form, or how to react tactically if a certain player was sent off, and the AI coach replied with a suggestion. It even provided inspirational quotes for Norman to use in his pre-match and half-time team talks.
For the record, the AI coach recommended a 4-3-3, and Wingate & Finchley drew 1-1. “The AI has been a really useful aid for the coaches at the club,” Norman said at the time. “With today’s draw, it can now claim to be an unbeaten coach.”
It was only a one-off, in partnership with a British science fair – and at its core, not much more than a cute PR stunt by an obscure non-league team. But it turns out Wingate & Finchley were ahead of the curve.
That was four years ago, and the world of artificial intelligence has evolved rapidly since. Today, ChatGPT can help kids cheat on their homework, or draft the emails you can’t be bothered to write yourself. Tools like DALL-E and Midjourney can conjure up perfect images of Donald Trump lifting weights in an orange prison jumpsuit, or the Pope in an all-white Balenciaga puffer jacket, and convince everyone they were real. AI is poised to sweep through every field and industry, making some jobs obsolete and totally reframing the nature of others.
Sport has long been at the forefront of technology, and as a result of AI’s progression, big changes are on the horizon. Coaches can breathe a sigh of relief, though: machines won’t replace you anytime soon. But they might make you better.
All over the world, there are engineers, scientists and developers trying to figure out different ways to leverage the power of generative AI – a type of technology that can produce new content based on the “training data” used to create it – in sport. Two data scientists working with the United States Soccer Federation built a neural network to help identify what it is that makes a counter-attack succeed or fail, in both men’s and women’s soccer. SportsPower AI, a company with roots in Brazil and the US, has created an AI assistant coach, similar to the one Wingate & Finchley used, which offers expertise in basketball, soccer and volleyball, and can answer “virtually any question about the game in real-time”. Another start-up known as Darius promises to “give your team superpowers” and provide players with tailored training programs.
Scott Bronkema, a former college football player at Malone University in Canton, Ohio, brought these innovations to American football, creating from scratch an automated playbook that not only gives coaching insights into how previous offensive and defensive moves worked out, but simulates how they could work against a given opponent. As far as he knows, he’s the first to do it.
Putting it together was easier said than done, and the result of a lot of tedious work. “There were so many problems,” Bronkema said, mostly around training his model to recognise padded, helmet-clad football players in the many hours of footage he had pooled together from myriad sources, including angles of games recorded for coaching purposes, Twitter, game-review tool hudl, and on-demand and live-streaming subscription service NFL Game Pass. “[Figuring] out to make it a birds’ eye view was the biggest. The camera in the coaches film moves. And to do it, you need at least four static points on the football field to match it to the ‘field’ you’re drawing on.” Given the rate of recent advancements, it’s not too hard to imagine AI working this stuff out for itself soon enough.
Scott Bronkema’s application identifies each of the players (left) and then records their movements as the play unfolds (right). This play is from the 2019 Alamo Bowl, Iowa State University vs Washington State University.
This all sounds very futuristic – and it is – but it is the natural extension of AI work that has been ongoing for decades, albeit focusing more on board games than sports. It is now being taken forward by a dedicated team within the Australian Institute of Sport in Canberra, which sees AI as the next frontier in sports technology, and a potential competitive advantage in the race for Olympic gold.
In 1997, the IBM supercomputer Deep Blue became the first machine to beat a human world champion at chess. But it took almost another 20 years for a machine to figure out how to win at Go, an ancient Chinese game played on a 19×19 grid of lines that is infinitely more complex and sophisticated than chess, and has more legal board positions than there are atoms in the observable universe. It had long been regarded as a “holy grail” of sorts for AI, an unattainable milestone.
AlphaGo, a program developed by London-based DeepMind Technologies, learned the rules and explored all possible strategies by simulating more games against itself than humans had ever played against each other. In 2016, it was pitted against South Korean world champion Lee Sedol in a five-game match that was broadcast live, and offered a $1 million prize to the winner. The program prevailed 4-1, a victory that arrived at least five to 10 years ahead of what most experts had anticipated. Lee was confronted with moves he’d never even considered. He retired three years later, saying that while he could be the world’s No.1 player, he would never be better than AI, an “entity that cannot be defeated”.
South Korean professional Go player Lee Sedol puts his first stone against Google’s artificial intelligence program, AlphaGo.Credit: Getty
Stuart Morgan, who heads up the AIS’s machine learning, performance technology and AI unit, described AlphaGo’s breakthrough as a “lightbulb moment” for him in the application of AI across sports.
“We started to say, OK, well, if an AI agent can discover new tactics and new approaches in a game like Go, what could it do in field hockey, or basketball, or tennis?” he said.
”Could we harness that kind of approach and find new and interesting ways of solving really complex problems in sport? If we translated this to the AFL domain, if you had a team that was flooding the defence, or had some kind of novel tactic that was catching opponents off guard – could you use an AI assistant to explore countermeasures that would potentially circumvent whatever that advantage is, and find other solutions? Or if you had a clever tactical idea that you wanted to be able to deploy, but you weren’t sure whether you might be exposing yourself to some kind of counter-attack as a result, could you explore those ideas in a simulation, and find out what [an opponent] would do to try to counteract the strategy that you’re thinking about?”
This is old news for anyone who’s played Football Manager before – the long-running, much-loved simulation game that puts you in charge of a professional soccer team, where you transfer players in and out of your squad, negotiate contracts, and fine-tune your tactics. In FM, you could, for example, play with seven strikers, two midfielders and one defender, if you really wanted to, and try it out against Manchester City, just to see what would happen. (Spoiler alert: you’d get carved up.)
The only difference between FM and what is now happening in the real world, Morgan said, is the level of sophistication of the engine and the data fed into it.
Stuart Morgan is enthusiastic about the potential for AI to give Australian sport an edge.Credit: Jason South
For many years now, all sorts of different data has been collected from athletes and teams through GPS devices, heart-rate monitors, sensors and strain gauges. But the full potential of that data, Morgan said, has never been unlocked. Sport is now moving from simply recording data and analysing it, or using it to model behaviour, to not only predicting what the data says might happen next, but letting AI loose to discover innovations in tactics or strategy that people haven’t even thought of yet, just like how AlphaGo came up with stunning new moves.
At the Tokyo Olympics, Australia used a system called “Sparta Two” which automated the tracking of every swimmer in a race – and by the end of it, within seconds, coaches were provided with the most granular of details on a swimmer’s stroke rate, turn times, breathing rates and other statistics. It was a four-year project and a major breakthrough for the AIS but, for Morgan, it also signalled the end of innovation in that particular area.
“We know now that the Brits and the Germans and the US, they’ll have their own equivalent systems by the time we get to Paris. That stuff is becoming business as usual. So we’re shifting our attention,” he said.
“We’ve got our eyes on the 2032 Olympic Games. Our research and development timelines are geared towards making sure Australia is really well-equipped to be the most AI-assisted sporting nation in the world. We’ve got most of a decade to really make significant gains; that’s where our attention is at the moment. We’re looking at things like ChatGPT as really kind of confirming our belief that there’s going to be a lot of competitive advantage to be derived from AI if you know where to look.”
Ariarne Titmus and other Aussie swimmers at the Tokyo Olympics had an edge, thanks to the “Sparta Two” project steered by the AIS.Credit: Getty
Morgan is trying to create what he calls “SportGPT” – a program that would enable coaches across any sporting pursuit to ask questions in plain language, prompting the AI to dive into the data, find the technical answers and respond with insights they could actually understand.
If it sounds like all the fun and humanity is at risk of being sucked out of sport in the very near future, consider the alternative viewpoint.
“I actually think the opposite is true,” Morgan said.
“I think it allows coaches to release their inner creativity. Coaches have incredibly good intuitive understanding of the sports they work in. We would never try to suppress their creativity or replace it with an AI. What we’re trying to do is build assistive technologies that allow a coach to explore new ideas, and maybe inspire new ideas. If an AI can come up with a seed of an idea, we’re then relying on the coach to really bring that into fruition.
“We talk about this idea of allowing experts to be experts. Across the nation, we train thousands of sports scientists, there’s an untold number of PhD-qualified biomechanists … often we utilise their expertise by putting them in front of the computer and getting them to manually annotate events in sport. If we can automate those kinds of things, then we release those skilled and talented and creative people to do much higher-level work.
“Humans are genuinely creative, and they have the capacity to bring together disparate ideas and come up with something entirely new. Computers can’t do that. There is nothing in the AI domain that is remotely creative. If you’re a highly creative and inventive coach, you’re never going to be replaced by anything, because that’s not something that now, or in the foreseeable future, could be replaced by an AI.”
Bronkema has reached the same conclusion. “Not a chance in the world,” he said when asked if coaches should be worried about going extinct.
“AI would tell you 80 per cent of NFL plays are two yards gained or more. Scenario: it’s the fourth down and one yard to go, with one minute left, and you’re on the 40-yard line. Well, AI would be like, ‘Go for it.’ But it doesn’t take into account that your starting right guard and best offensive lineman just got hurt, and your quarterback has a bad ankle. And if you miss it, the other team can easily kick a field goal to win. So the smart coach would say, ‘I’ll take my chances with punting the ball away and making the other team go 80+ yards to win the game.’
“What my product does and what AI, and computer vision, and all of it can do, is just be another data point in better play-calling, better coaching of players, and better strategy prep.”
We won’t ask AI to imagine what all of this could mean for Australia at the 2032 Olympics in Brisbane. We’ll just ask Morgan: will we be raking in the gold medals? What does the future of sport look like?
“We’ll have found ways of empowering coaches with assistive technologies that accelerate their creativity and multiply their capabilities in ways that other people haven’t,” he said.
“And so we’re operating on a whole different level, and we’re five years ahead of the rest of the world. We’ll see a whole new range of really interesting approaches to tactics, or the preparation of athletes. If AI can help us keep athletes on the park more often and predict or reduce injuries, then how good is that for sport?”
News, results and expert analysis from the weekend of sport sent every Monday. Sign up for our Sport newsletter.
