by Rob Perez

AI vs Tennis

There’s this new thing. It’s the biggest thing. Every country loves it and claims it as their own. The future will look at this moment and divide history into two sections: Before and After.

I’m not talking about “K-Pop Demon Hunters.” Although that qualifies in all categories.

I’m talking about AI.

AI is smart. How smart? They say it’s going to cure cancer. Eliminate all but trial lawyers. Can sift through decades of data in a New York minute. AI knows quantum physics, climate modeling, and protein folding. AI can write in Latin, iambic pentameter, and more usefully, in English.

And this moment — here, now — is the dumbest AI will ever be. Every day AI’s getting exponentially smarter.

So why is AI bad at tennis?

I’m not talking about AI playing tennis—although I’d like to see that. Why is AI bad at talking tennis? Returning readers know I’m a fan of tennis. This time of year, I’m watching the U.S. Open being played in New York.

The USTA, who puts on the thing, has partnered with IBM for decades. They were surprisingly early adopters of AI, bringing it to the Open in 2017. There’s a whole lot of tennis going on at the Open. Dozens of matches being played simultaneously, hundreds of matches over the fortnight.

In the ye olde olden days, the USTA needed a small army of people to cut the highlights, call the video, and post the thing. Now it’s all AI. Ready in mere minutes.

Except AI is bad at tennis.

Take any match highlights. For instance, in highlights of a match with Jessica Pegula (current World #4) vs. Victoria Azarenka (former world #1), AI managed to pronounce Azarenka three different ways. I can hear the pushback: Announcers mispronounce foreign names all the time! I agree! Foreign names come with many consonants. And vowels. Thing is, AI mispronounced “Pegula” three different ways. And she’s American.

Keeping score in tennis is different than other sports. Tennis doesn’t have “zero.” Tennis has “love.” That’s not a metaphor. Although maybe it is. Also tennis doesn’t have 15-15. Or 30-30. It’s fifteen-all. It’s thirty-all. And nobody. I mean, nobody says forty-forty.

It’s deuce. It has been deuce since before electricity. If your highlight reel says forty-forty, maybe they’re talking about bingo? Scoring tennis is quirky, but it’s not rocket science. So why is AI good at rocket science and bad at tennis?

Here is AI tennis commentary from the US Open, in its own words:

“Thirty-all” is called: “30-zero,” “30-to-zero,” and my personal favorite “30-to-zero in points.” The latter is like saying, “The Cowboys lead the Vikings fourteen—to-zero in points.”

“As the fifth set unfolded at zero–all…” Counterpoint: In soccer, zero is nil. In cricket, it’s a duck. In baseball, it’s a goose egg. IN TENNIS ZERO IS LOVE!

“Player X hits a backhand winner, securing the point.” Really? Was the point insecure? Does Player X get to take the point home?

“Player X approached the net and concluded the point…” Really? Are some points inconclusive?

“In the first set, with Player X holding advantage at 2-games each…” AI can run regression models, crack cryptography, but AI doesn’t know two-all is a tie?

“Player X forced an unforced error.” Interesting. A forced unforced error?! Perhaps it’s a paradox, a wormhole, a Buddhist koan. But I can tell you it’s not tennis.

So AI is bad at tennis. But maybe this isn’t an AI failure? Maybe Someone programmed AI to cut highlights in 2017 and then went on walkabout? Hey buddy! I think it’s time to come back to work. I would really like AI to figure out what love is – at least the tennis kind – before it cures cancer.