Defector has partnered with Baseball Prospectus to bring you a taste of their work. They write good shit that we think you’ll like. If you do like it, we encourage you to check out their site and subscribe.
This story was originally published at Baseball Prospectus on Oct. 10.
If you’re visiting the pages of Baseball Prospectus, it’s very likely that you’ve watched a game, or a highlight, on MLB’s website in recent months. And if you’ve done that, it’s equally likely that you’ve seen the following ad, conservatively speaking, at least 2,000 times.
Let’s take a step back for a minute, let that metastasize. This Google advertising campaign isn’t new. The spots were introduced last postseason, the first one (entitled “Mustard”) concentrating on the correlation between pitch velocity, IVB, and whiff rate. Which is fine. We don’t really need to drain the world’s oceans and kill god to get that kind of epiphany—half a minute on a baseball savant search will get you the same answers—but it is what it says it is, a quantified contextualization of the action.
Does knowing this detail make you a smarter, more entertained fan? Yeah, probably, I guess. It’s not necessary, but again, you’re here, so we can’t pretend this information doesn’t contain some intrinsic value. And to be honest, probably more than I should, I like this kind of quantification, which I like to call an “expectation statistic.” It doesn’t actually tell you anything new, or anything specific to the player or the situation, but it does help you feel the set the correct level of disappointment (or elation) for what actually follows. It’s for contextual purposes, and for someone watching a game, context is really useful. It’s a hell of a lot better than “Carpenter’s gone 1-for-6 against him lifetime,” and broadcasters have been feeding us that brand of slop for generations.
The expectation statistic isn’t the sexiest concept in our business: They’re not pushing the research forward, or driving the sales of the book that breaks down how this year’s champion is truly special, as opposed to all those special champions from before. But it’s important nonetheless, acting as a bridge between the experienced and the neophyte fan, helping to set the stage. They help teach people how to become fans. What is fandom, after all, except attaching one’s welfare and sanity to the external world, quite out of their control, and then managing their growing horror in real time? Probabilistic thinking is a vital skill, not just for cushioning the blow of getting bummed out about bad news, but in teaching that every action comes with more than one possible result. My wife is not a sports fan, and does not like thinking probabilistically. This week, she did not have that luxury, because our plans over the next couple of weeks were a string of expectation statistics. “The Yankees lost, so MLB changed the game time, so I won’t be around for dinner tomorrow.” “If the Seattle Mariners win Game 5, we’ll need to find a babysitter for next Friday” is exactly how fandom works, but for my wife, it’s like being a fan at gunpoint. All I can do is say “There’s going to be a 45 percent chance that you’re going to be pretty annoyed with me.”
But that’s just the first 15 seconds of the ad. It follows that first stat up with a bit about how hot dog sales go up four percent when the home team is winning. In keeping with the grand tradition of AI, there’s no citation for this. But what’s unclear is who this fact is for. Why would you care if the team sells more hot dogs? Because it’s data. Baseball has already trained most of its fans to think like general managers, but they like to hang out in the offices wearing suits and making trades, not performing cost-benefit analysis on concession price points like they’re playing Rollercoaster Tycoon. But that’s the point. The promise of AI is that the sheer quantity of data, the computing power, is what makes it smart. And then, to steep everything in a vague, unspoken sense of profit, that this power is worth something—worth something to the fan, worth something to the (possessor left unstated, for the sake of the imagination) wallet. That citation is always, always needed.
Another one notes that when a team has a mascot with wings, “they just so happen to get 0.4 more steals per game.” Numbers flash across the screen, players pump fists and make faces. Everyone’s having a fun time not thinking about this too hard. And this is what brings us back to this year’s ad about bat tapping: a willingness to play fast and loose with the concept of correlation. A transcription:
Does a bat tap untap good mojo? AI from Google Cloud knows that 125 hitters tapped home plate in last year’s postseason. Those who tap twice or more had a hard hit rate of 41 percent, the same as those who tapped once. But surprisingly, it was the non-tappers who actually untapped 7 percent more hard hits. Now that really hits home. Catch the game like never before with AI from Google Cloud.
I’ve spent a couple of weeks trying to wrap my head around exactly why this advertisement makes me so upset. It’s not the willful, winking obfuscation of causation and correlation, or rather, it’s not just that. It’s how this all-powerful mathematical tool fails to respect math, or people. You have to recall that in the setting when Baseball Prospectus first went online, the attitude toward sports statistics was very different.
Not completely, mind you; they’ve never stopped spitting that cliché about damned lies. But when some of the smart folks outside the usual channels started noticing that some statistics correlated pretty well with the big one, wins, it was very threatening to the people within the game. The act of quantifying baseball was seen at the time as an assault, something that somehow lessened its cherished mythos. It became very important to be meticulous, to show work, to demonstrate that repeating the old wisdom thousands and thousands of times didn’t make it any more right.
They were pedantic times, and they were important times. And yet the way Google presents its information in these commercials, data dressed up as meaning, makes their own figures sound exactly the way that old-school thinkers thought sabermetrics did during those long, wearying ideological battles of the 1990s and 2000s: “On Tuesdays when Mercury is in retrograde and the pitcher has just eaten chicken for lunch…”
The numbers have to mean something. They can mean something to the teams trying to win the baseball games, and to the fans trying to understand how the teams try to win the baseball games. That’s fine, if that’s your bag. They can also mean something to the work of enjoying baseball: the expectation statistics, the narrative, the backstory. All of these things make baseball better. This is not what AI from Google Cloud is selling, and it’s not what “artificial intelligence,” as a promoted prospect, is particularly interested in. AI cannot make human connections. Just as its computational power is a pale imitation of the human brain, its product is essentially a poor copy of nature, producing stuff, meaningless data, and requiring the human to make the connections to it themselves. This is the modern internet: uncurated, unhelpful, infinite. Its humanity is buried beneath a million layers of human-engineered meaninglessness, and when you do find something actually created, more often than not, it’s an intentional lie. Or, more specifically, advertising.
The irony of all this, in this particular instance, is that we never needed random computer-generated trash to scavenge and build meaning out of. We already had Rays-Angels games to do that. Surviving that, finding meaning in it and sharing it with each other, is the whole point of baseball. It’s the whole point of people. To think that computers can do it for us, do the work of fandom, any more than writing our novels and our obituaries, is depressing as hell. But look on the bright side. The metaphorical hot dog sales will go up four percent, for somebody, somewhere.
