Wimbledon 2023: Tennis whites, Pimm's and IBM AI
AI is finding new data and new insights – but pronouncing ‘Djokovic’ has been a struggle
IBM’s Data Bunker, underneath Wimbledon, is a compact but important part of the world’s most prestigious tennis tournament.
Tennis has come a long way from the early days of Wimbledon, more than 140 years ago. The game is faster, more professional and much more expensive: big money is on the line.
With that in mind, players and coaches want all the help they can get. Data is key.
That's true even for people not directly involved in the event. Fans love seeing stats about their favourite players and following their journey; whether they flame out in the first round or make it all the way to Centre Court.
Beneath the grassy courts, the tennis whites and the strawberries and cream, IBM Data's Bunker is hard at work every year gathering and analysing that data for the insights used by sporting desks around the world.
The team also works on the public-facing tech, like Wimbledon.com and the Wimbledon app.
"We're giving fans around the world a feeling of being here without being here," explains Kevin Farrar, Head of Sport Partnerships at IBM UK when we talked to him during the tournament. "It's really bringing the drama and excitement and beauty of Wimbledon to fans around the world, through those digital platforms."
Wimbledon 2023: Tennis whites, Pimm's and IBM AI
AI is finding new data and new insights – but pronouncing ‘Djokovic’ has been a struggle
IBM’s Data Bunker, underneath Wimbledon, is a compact but important part of the world’s most prestigious tennis tournament.
Bringing the action to everyone
IBM has been working with Wimbledon since 1990. Together, they have pioneered some major innovations to make the tournament more accessible. But in all that time, only the major courts have enjoyed match commentary. Until now.
IBM introduced AI-generated highlight reels in 2017, using AI to "listen to the noise of the crowd [and] look at the gestures of the players" to analyse excitement levels, gluing clips together into short videos to post online.
That wasn't only for the biggest matches, either. Juniors, seniors, wheelchair tennis - all of them have highlights, but no commentary.
The solution is IBM's new AI Commentary feature, which builds on the highlight reel function. As of this year a synthetic voice (or closed captions) will provide play-by-play narration for the start and end of each reel, along with key points.
Farrar stressed that the goal is not to replace human commentators.
"A lot of the top commentators have actually played on the Centre Court. Our AI is never going to play on the Centre Court. This is about providing commentary, ultimately, where there isn't commentary today."
Wimbledon 2023: Tennis whites, Pimm's and IBM AI
AI is finding new data and new insights – but pronouncing ‘Djokovic’ has been a struggle
IBM’s Data Bunker, underneath Wimbledon, is a compact but important part of the world’s most prestigious tennis tournament.
Training an AI for tennis whites
The system is built on WatsonX, IBM's enterprise AI and data platform for generative AI. Like any AI it needed training before being unleashed in a live environment.
The system is based on an IBM foundation model called Sandstone (similar to Google's T5 models), which has been "trained in sport."
IBM also used Sandstone for the golf Masters Tournament this year, but it had to be retrained in "the language of Wimbledon."
"We're teaching it the game of tennis [using court-side stats data], and the language of tennis, and the language of Wimbledon: it's the Gentlemens Singles, not the Men's Singles, things like that."
Once it's been trained, the model needs tuning by humans and AI working together - for example, on player name pronunciations. "They seem to get harder every year," Farrar jokes.
Finally, once the AI has been deployed there's an ongoing governance and monitoring procedure, "to make sure that it continues to be trusted AI."
Wimbledon 2023: Tennis whites, Pimm's and IBM AI
AI is finding new data and new insights – but pronouncing ‘Djokovic’ has been a struggle
IBM’s Data Bunker, underneath Wimbledon, is a compact but important part of the world’s most prestigious tennis tournament.
The possibilities of computer vision
In the live environment, all of the AI analysis is based on stats data fed into the model. However, before moving on Kevin showed us a proof of concept using computer vision, which could change the game again.
On-screen two players, Alcarez (congratulations on the tournament win, Carlos) and Berrettini, were charging back and forth across the court. Unlike a normal broadcast match, a coloured box and caption tracked each movement, identifying elements like ‘Serve' and ‘Forehand'. Beneath, textual captions followed the play.
This is a proof-of-concept of the next evolution of AI Commentary, pioneered with IBM Research. The system can analyse "like 17 different limb movements" for shot detection - and that's not all.
"The other thing it's enabled us to do is identify things that the stats we're collecting don't currently identify; things like shots between the legs, how many times has a player bounced the ball before they serve, how slowly or quickly do they walk to the chair?"
Data like this leads to new insights, which helps to enrich the AI commentary - and it's an important step is taking the feature to a live match.
"Wimbledon want to continue growing their global fanbase and engaging fans in new and exciting ways, and we're helping them to do that.
"It's very much about getting the balance right between the tradition and heritage of Wimbledon, which is obviously very important, and the tech and the innovation."