It is an odd feeling meeting the thing that will kill you – in a professional sense, at least.
But here it is, living under a stairwell at Wimbledon. It’s just around the corner from the cafeteria where all the journalists it will some day replace get their Americanos and flat whites.
I’d like to say it doesn’t look like much, but it really looks like something. Low lighting, a half-dozen human attendants, walls of flat screens that show it teaching itself the nuances of tennis.
IBM calls it watsonx – an artificial-intelligence platform that can do all sorts of things. Currently, it is learning how to become a Wimbledon-grade play-by-play announcer.
The biggest of all the very big brains in this room, IBM engineer Aaron Baughman, is trying to explain to me how this works. Though he is speaking to me in what I gather is the way he approaches grade-school tour groups and very smart dogs, I can’t fully grasp anything he’s saying. Something about piles of data, large language models and the ontology of tennis.
I catch the bit where he says watsonx only met tennis three months ago, and has since read everything – and I mean everything – about it.
Since it doesn’t want to be swamped by rioting broadcasters, IBM gets jumpy if you suggest watsonx might be after someone’s job.
“People want to hear John McEnroe call the Wimbledon final,” says Kevin Farrar, the IBM boss showing me around. “There is no substitute for the experience of someone who has been there and done it.”
True. But as I understand it John McEnroe does his job for millions of dollars, while watsonx will some day do the same job for $10 worth of electricity. Forgive me if I don’t take much solace in the ‘experience’ argument.
The current goal is to get the AI to the point where it can do subsidiary matches – first-round mixed doubles and such. Those encounters don’t get the benefit of their own call because there aren’t enough broadcasters to go around.
In some future world, the hope is that people can do the matches in English, while watsonx can do them in a few hundred other languages.
This sounds great. Tennis-loving, non-murderous robots bringing us all together through the shared love of sport. You and ChatBotBjornBorg over here, best pals. Like HAL in 2001: A Space Odyssey.
But the part that really sticks with me is where I ask Baughman if there is a point when what the AI does will be completely indistinguishable from a call by a human.
“Certainly,” he says.
Completely and totally?
“Yes.”
Baughman is a scientist. Nothing is certain in his world. So when he says “certainly” to you – a person who pays his considerable mortgage via the craft of journalism – that’s when you should start thinking about panicking.
There’s no need to panic right now. Just think about it. Really sit in it for a minute. Then write out a list of all your transferable skills. And then panic.
The good news for my sort is that watsonx isn’t great at this yet. And before you get there, reduce your expectations for what “this” is.
The AI isn’t calling matches at Wimbledon. You have to dig into the Wimbledon app to find it at all. In order to hear the AI, you must toggle it on. It only plays over short highlight packages for matches that are already completed.
What it can do at this point is add some colour around the action. That colour is beige.
Watsonx has two tennis voices – posh British man and posh British woman.
It doesn’t narrate the action, but sums it up in ways that contain basic insights (e.g. “Thiem forces Tsitsipas into the backhand error and wins the fourth set”).
It has trouble with names. For instance, “[Maria] Sakkari’s backhand is too good for [Marta] Kostyuk,” comes out sounding like, ‘Sockdy’s backhand is too good for Kosty.”
Elsewhere, it is perfect. Too much so. It is jarringly free of tics.
Also, something about the timing and inflection is off in a way you have trouble pinpointing. Like a horn coming in a half-beat too late.
All of those things, as well as the absence of vernacular, twigs your lizard brain like a twitching in tall grass. What you are listening to sounds human, but isn’t. Flee.
That said, watsonx is not terrible. If you hadn’t told me what it was, I would guess it is someone reading from a clichéd script. Isn’t that about the level of most sports chat anyway?
At the outset of this experiment, it took watsonx days to recalibrate itself when its IBM handlers gave it pointers. It’s got that time down to less than an hour. Whatever’s coming next is coming fast.
How much of a leap is it from calling games to writing about them? It’s an algorithmic baby step.
Honestly, I could think of a hundred good reasons they should replace me with a computer. I’m bad with names. My spelling is atrocious. I’m often so convinced that I know some arcane bit of trivia to be true that I don’t check it, and it isn’t true. That happens all the time.
Crap with numbers. Shaky on grammar. I can’t totally explain to you what syntax is, but I’m going to assume I’m bad at that, too. No need to check.
The only thing I’ve got going for me is that experience IBM’s Farrar talks about. I do know enough that you shouldn’t settle on a lede in the second set (or period, or quarter, or inning) – but I still do that constantly. You wouldn’t believe how many times I’ve been burned by deciding what the story is before there’s a story, but does that stop me? Of course not. I love pain.
So there I am in the last five minutes of whatever’s happening, blowing a million motor neurons trying to turn a ‘lose’ column into a ‘win’ column and ending up with a ‘your guess is as good as mine’ column.
That is our only protection. Watsonx and its ilk will soon be able to do what many of us do. Better even. But can it ever be as wonderfully, pathetically, hilariously human as we are while it’s doing it?
Cathal Kelly