a culture of sameness
You need to fund the strange stuff and serve it to people to make things materially less stagnant.
I deleted TikTok a while ago. Not for reasons much morally superior to a whim: the Trump stuff was finally happening to it, and I figured I didn't want to deal with how shabby and grubby it would probably get after the president's cronies got their hands on it. (I was probably wrong about that, but whatever, it stayed deleted.)
I'd spent a fair amount of my finite life scrolling through it — watching the guys in factories halfway around the world forging dies, or flicking past the not-quite-hostage videos musicians seem professionally obligated to post there — and I thought breaking the physical habit (open this folder and navigate to the app) would be a good way to do less of that. It helped!
Eventually, as one does, I succumbed again to the siren song of vertical video. Reels, TikTok's suboxone, was right there — built into Instagram, even. It promised the same people posting the same stuff, because who, these days, is going to make a video for just one platform? I tapped on the tab and opened it up. Immediately things felt different, Capgras delusion-y, like someone had gone into my apartment and moved all of my furniture about an inch to the left.
It also wasn't as captivating as TikTok, for some reason, despite just about all of the same people I watched on the other site posting the same videos to Reels. I just wasn't seeing the same things, like its recommender algorithm was just a bit different and couldn't seem to pin down what I was interested in consuming. I was a little surprised; TikTok had me figured out in about 10 scrolls, and it seemed like the video recommendations were only getting better over time[1]. Reels, on the other hand, still hasn't quite gotten it.
It's only now, having read this brilliant post[2] by the data scientist Lauren Leek, that I understand why I feel TikTok and Reels differ in some actual way: it's because TikTok's personalized For You Page algorithm, for whatever reason, seemed to serve me videos outside of the things it knew I cared about more often than the one powering Reels.
Every so often on TikTok I'd get something that seemed totally out of the blue, like the fourth video recap of some internecine drama blowing up one side of the site. A lot of the time I'd scroll by, but sometimes one would hold my attention. It made for a much more addictive experience. Reels, on the other hand, just shows me what I want; it can't surprise me. And thus: it's boring. It's not sticky.
And this is the thrust of Leek's post: recommender algorithms — of all kinds — are a very good way of averaging out difference.
Almost every recommender is trained to minimise a prediction error, get the rating wrong by as little as possible, or maximise the chance you click. But the problem is that under squared-error loss, the prediction that minimises your expected error is the conditional mean. So an algorithm that is uncertain about you, and it is always at least a little uncertain, hedges toward the average. The more uncertain it is, the harder it pulls you toward the crowd. This is why more data doesn’t save us. Personalisation under a standard loss function is regression to the collective mean with extra steps. Variance, the technical word for the stuff that makes you you, is expensive to predict. Basically, diversity is variance, and optimisers are built to minimise variance.
Leek's project is to understand why so many things feel the same right now. Her opening anecdote is about moving to The Hague, doing househunting, and realizing that everywhere already felt familiar — which shouldn't be the case[3]. We have more data than ever, in other words, and in every dimesion — and Leek lists several — these algorithms keep pulling things back toward whatever's most mid, or what's predictably satisfying ("satisfying").
The problem, of course, is that if you can't be suprised, you'll never know what you'll be absolutely taken with. The new genre-bending artist you'd love more than any other will never appear in your recommendations because a recommendation algorithm won't take a chance on putting them in front of you. Too much variance, as Leek has it.
I think this kind of enforced sameness is corrosive. Not only does it create the pervasive feeling that culture's stagnating — which, sure, when was the last time you were pleasantly surprised by pop culture? — but it makes it feel like it's very hard to get out from under the algorithm's proverbial thumb.
And it's because the internet both flattens subcultures — a corollary: when was the last time you encountered a genuine regionalism? — and renders real diversity ("variance," in Leek's words) invsible, because it's not something than can be reliably recommended. A couple years ago, the writer and cultural critic W. David Marx noticed this in the cultural sphere. A couple years ago, in a piece for The Atlantic about cultural arbitrage — where one can earn cultural capital for bringing a cultural phenomenon from one place to another — he lamented that the internet has severely diminished its returns:
But the internet’s sprawling databases, real-time social-media networks, and globe-spanning e-commerce platforms have made almost everything immediately searchable, knowable, or purchasable—curbing the social value of sharing new things. Cultural arbitrage now happens so frequently and rapidly as to be nearly undetectable, usually with no extraordinary profits going to those responsible for relaying the information. Moreover, the sheer speed of modern communication reduces how long any one piece of knowledge is valuable. This, in turn, devalues the acquisition and hoarding of knowledge as a whole, and fewer individuals can easily construct entire identities built on doing so.
His point was that cultural arbitrage produced novel ideas, the kind of thing that keeps things from feeling stagnant — from feeling like just about everything you see on social media. Because, of course, that's where it's easiest to notice all this; the figure of the professional creator is a creature built entirely by the algorithm, a human who nevertheless is piloted like a mech suit by the whims of a recommender algorithm in thrall to the wisdom of a crowd it's half-blinded.
If you consume enough of the video people make on the internet, you'll find that there's a formula to all of them, so general as to be applicable across every genre: catchy hook, curiosity gap quesion, half-answer to the question posed, outro promising more of the same next time. (Or thereabouts.) The reason it works is because it's engineered to grab your attention, which oddly isn't the same thing as being interesting[4]. If you watch a channel long enough, you'll almost never be surprised[5][6]. To be successful under an algorithmic regime, you've got to be predictable enough to be served widely.
At the end of her piece, Leek outlines a possible solution to the sameness problem: exploration, or getting the algorithms unstuck from the mean.
So the entire fight is forcing exploration back in, and when I simulate it, the result is actually encouraging. The only thing I vary is how much exploration I force, and diversity goes from near-total collapse to fully preserved. But, and this is the finding regulators need, it’s a threshold. Two or five percent, roughly the level of a token “discovery” tab, does almost nothing. You have to clear a cliff, around a fifth of the entire system’s attention, before variety survives at all (just as in the nature).
You need to fund the strange stuff and serve it to people to make things materially less stagnant. It's funny, because to me, here in the cultural sphere, that might look like more gatekeepers. Though that's actually not quite true: what it means is we need more people empowered to spend real money to enact their tastes. As in: the editor at a major house who's empowered to spend a nice chunk of change on a very weird book. Or the film distributor who can put a genuinely odd movie on millions of screens. Or the video game publisher who takes a chance to fund a small team with a good idea.
The reason, of course, is that even if the cultural products they fund don't achieve total world domination, the people they inspire might make the next big thing. The next thing we love. None of this stuff is linear, and none of the rewards are purely monetary — which is a hard sell in a data-driven (derogatory) world. At the moment, the system for creating cultural works feels something like an algorithmic lottery: in just about every field, to get past the gatekeepers, you already have to be doing the thing you want to do and performing enough for social media that you have a following.
Because for artists today, the Powers That Be have decided that an online following is a proxy for safety: that the number by your name means you're worth paying attention to, and in theory have both earned a notable amount of attention and putatively have fans who care about your work. Leaving aside how easy it is to fake those metrics, I want to say that it's a very bleak way to evaluate the worth of an artist.
And here's where we come to Backrooms and Obsession, both made by two young filmmakers who started as YouTubers. They're not the first people to come up through this opaque system, but they're the most successful. What it makes me wonder: who else is making things on YouTube that the algorithm missed? Who else is deserving of the same shot? And what about the people who don't have the means, for whatever reason, to make enough films for the internet that the Powers That Be take notice?
If you're anything like me, it's a question that keeps you up at night. Who else are we missing? And what might they have made?
This, of course, is the tech everyone in the world wants from TikTok. ↩︎
Seriously, go read it immediately. It's great. ↩︎
Incidentally, my buddy Kyle Chayka wrote about this about a decade ago for The Verge, a good blog I used to work for. His piece is a very good companion for Leek's. ↩︎
This is why watching too many vertical videos can make you feel like you're in a hostage situation. Your attention is being held hostage. ↩︎
A theory: this is why those apology videos — or other out of character videos — are recommended so often. It's because the surprise of a channel breaking its format keeps people unusually engaged. ↩︎
And yes I know I'm being indulgent about the footnotes here, but that's because I finally got them to look and act right!!! ↩︎