Although this year’s Olympic Games were unusual, thanks to the coronavirus pandemic, certain parts of the competition looked remarkably familiar, like watching judges hand down scores in events from diving to skateboarding.
In the future, some of that judging could be handled by artificial intelligence tools, which could one day prove to be more accurate and less prone to accusations of bias than human evaluation, according to Patrick Lucey, chief scientist at the sports data firm Stats Perform. Lucey says he predicts diving may be among the first sports to benefit from automated judging, with its relatively small and controlled environment compared to the complexity of, say, a gymnastics floor routine.
“It would be transparent,” says Lucey. “There wouldn’t be any confusion around bias or whatnot.”
Already, automation is playing an increasing role in human sports officiating. The U.S. Open and other tennis tournaments have been phasing out human line judges in favor of automated tools that determine ball placement, partly as a way to have fewer people on the field during the pandemic. And Minor League Baseball and other smaller pro baseball leagues have been experimenting with automated systems to help umpires call balls and strikes.
As machine learning has improved in recent years, it’s taken on a greater role in tracking player performance and on-field action in a variety of sports. Hawk-Eye, owned by Sony, provides the line judging technology to the U.S. Open and has seen its ball-tracking technology used and tested in sports, including soccer, rugby, and badminton. A system called TrackMan provides baseball coaches at a variety of levels with information about player pitching and hitting. And Stats Perform’s AutoStats system is recently moving from being able to extract data from college basketball games where specialized cameras are installed in the arena to simply observing existing video footage and documenting what’s going on in detail.
“We can actually measure the defense,” Lucey says. “We can actually measure who’s guarding who.”
The systems are trained based on video footage labeled by human experts, and they can come to observe detailed aspects of play more quickly and accurately than humans, Lucey says.
“Given enough labels, what we call it in this supervised paradigm, the computer’s able to emulate what those sophisticated humans can do,” he says.
Not needing to install equipment in every stadium is a big benefit in a sport with more than 300 teams participating in the NCAA’s Division I alone, with fans and pro scouts angling for information about team and player performance. The company last month renewed a deal with the Orlando Magic, which uses AutoStats as part of its scouting process. AI-driven player evaluation can help pro teams spot under-appreciated college talent or have a better understanding of the capabilities of players already on their radar, Lucey says.
“Tracking data has become the gold standard of NBA analysis, and having access to AutoStats’ college tracking data has allowed us to bring our college scouting process up to the same depth of analysis as we have at the NBA level,” said Jeff Weltman, president of basketball operations for the Orlando Magic, in a statement. “Being able to back our scouting and prospecting with real data from AutoStats has given us a huge advantage these past two seasons and provided important insights during the recent 2020 NBA Draft.”
Artificial intelligence can also naturally play a role in athlete training and practice, providing real-time feedback on play and technique. Tools already exist to let even hobbyist athletes track their performance over time using footage captured from a smartphone, and these apps will likely only get more sophisticated as cameras and software continue to improve. Tennis app SwingVision can offer real-time analysis of your game based on footage taken with any modern iPhone or iPad, and HomeCourt can offer similar feedback for basketball players. And golfers have a range of tools to use to improve their game, including some that use video footage and some that collect data from on-club sensors.
While these technologies are already here, fully automated judging at the level of the Olympics is still a ways away. For algorithms to completely replace human judges, the sports world will likely demand a high level of accuracy, even with known human errors and biases (think of all the Cold War-era jokes about Russian Olympic judges).
“It doesn’t have to be as good as a human,” says Lucey. “It really has to be 10 times as good.”