In Defense of Algorithms

Review of Kyle Chayka’s "Filterworld”

Istanbul’s Blue Mosque

Years ago in Istanbul, I visited a new coffee shop recommended by a friend. It was hip and bougie and cute, similar to places I’d often visit in San Francisco, where I lived at the time. A bit guiltily, I ordered a flat white coffee and avocado toast, happy to find good coffee and curious to try my hometown specialty halfway around the world.

I’d been surprised to find a place like this — different from anything I’d known while living in Turkey nearly a decade earlier. The closest we could find then to good coffee was from Starbucks, a twenty-minute walk down a large hill and surprisingly expensive, and so I subsisted most of the year on Nescafe and tea, or çay. (It’s pronounced “chai” — and while Turkish coffee is delicious, it's drunk more as an occasional treat. Turkey, at least then, was very much çay country.)

But as I sat enjoying my food and drink, I looked around and realized the coffee shop felt almost too comfortable. I was eating avocado toast — perhaps the most iconic, derided staple of San Francisco culture at the time. Why, I wondered, had I even come to Istanbul?

I began noticing other places around the world where seemingly disparate cultures seemed to rapidly blend, cohere, and erase. At a conference, I saw a presentation referencing an essay titled “Welcome to Airspace,” which discussed just this phenomenon — the ways in which coffee shops and hotels all around the world seemed to be converging on similar aesthetics, a blinding homogeneity of cute succulents and subway tiles and Edison bulbs. And it blamed the phenomenon on the internet — Instagram, in particular, and the ways in which the platform allowed people to share, and then copy, aesthetics from one place to another, no matter the physical distance between them.

I began noticing, too, how social media seemed to drive a rapid inundation of certain outdoor sites. Peru’s Rainbow Mountain is one — a place I didn’t even hear of until years after visiting the country in 2010, yet which has already grown so popular that some now fear for its preservation. We can see a similar story of exploding popularity in southern Utah, at sites like The Wave and Antelope Canyon. Both are easy to access from the nearby Page, Arizona airport, and both are easy places to take gorgeous Instagram photos. As a result of all this, many now urge other outdoor lovers to stop geotagging their pictures, hoping to stem the flood of new visitors.

I’ve thought about all this for years, and in fact planned a series of posts about the phenomenon here. As some readers know, my hope with these posts is to build toward a book — and in my outline for that book I’d planned for this idea, the way in which the internet’s scale flattens consumer markets and homogenizes global culture, to comprise an entire chapter.

Fast forward to January of this year, when I heard an interview with Kyle Chayka discussing his new publication Filterworld, an entire book discussing… all of this, in great depth. What’s more, I soon realized he’d written the “Welcome to Airspace” essay in 2016 that initiated a lot of my thinking on this, too.

I thought I’d been scooped! And so of course, I rushed out to jealously buy and read the book.

Image via GIPHY

But I was surprised. I found myself agreeing with — and often depressed by — much of Chayka’s description of our current moment, impressed by his thorough investigation into the ways in which the internet today “flattens” global culture. I largely agree with his depiction of the symptoms of our current online moment — but as it turns out, I fundamentally disagree with him on their causes.

The diagnosis underlying everything Chayka describes in the book, I believe, is both grander and more systemic than he suggests. Which of course implies a grander, more systemic treatment going forward. Chayka repeatedly lays blame for our cultural flattening at the feet of recent popular scapegoats: “the feed” dominating our online lives (the word “feed” appears over 300 times in the book), and “the algorithm” that controls and sees all (a term which appears more than twice as often.)

But what if it isn’t necessarily the feed, nor algorithms, causing all this trouble? What if the problem lies within the structure of the internet itself today — what if the call is coming from inside this very big, very strange house?

What if this is just what happens when billions of people come together in one place?

Image via GIPHY

Filterworld, at its heart, describes the way in which the internet is leading us to the same places — to similar films, similar music, similar aesthetic and fashion tastes, similar travel destinations and even politics. And not only is the internet leading us to these places, it’s causing our culture to conform to their formats, as well. The feeds and recommendation algorithms of large apps and sites like Instagram, YouTube, Twitter, Netflix, and Spotify first aggregate the tastes du jour of the global populace, and then regurgitate them back. 

Even more insidiously, these same forces lead to the creation of global culture that conforms to these diluted, homogenized formats and tastes. If a certain style of coffee shop, artwork, poem, song or show gains popularity via algorithmic feeds, then it behooves artists and other creators to conform to this style.

This, to me, is the most sinister — and depressing — element of Chayka’s argument. It’s not enough, he implies, for us as individuals to leave these platforms. One day soon, it may be impossible for us to find any art, literature, or other cultural artifacts that haven’t been infected by the “flattening” elements introduced by our algorithmic feeds — no matter how remote the cave we choose to live in to escape them.

What does this flattening look like? Not only does it include the aesthetic homogenization of coffee shops as far away as Brooklyn, Berlin, and Bangkok, and increasingly voracious swarms of tourists to highly-Instagrammable outdoor destinations. Chayka also describes the ways in which Spotify’s recommendations seem to push increasingly shorter and similar songs, the seeming convergence of Netflix recommendations to similar tastes and styles, and how Instagram and Twitter may shape literary and journalist voices to better suit those formats.

He discusses at length, for instance, the success of Rupi Kaur, whom he argues rocketed to literary fame through poems and illustrations well-attuned to the scrollable, visual format of Instagram. Even though, as a Michigan student cited in the book decries, “Kaur’s poetry states obvious, mildly interesting stream-of-consciousness shower thoughts in visually appealing ways.”

With each additional example of cultural flattening, Chayka points to the same proximal cause: algorithms nudging content toward us and us toward content, moving us away from a selection of culture driven by individual agency and toward a passive consumption via “the feed.” Facebook, Twitter and Instagram replaced their original, chronological feeds with recommendations based on our personal data and the engagement of others. Spotify and Netflix automatically generate playlists and promote shows on the home page of the apps, guided by the same data sources. And now we have TikTok’s “For You Page,” the apotheosis of all this — the logical end-point of passive, algorithm-driven consumption. Until, of course, the next algorithmic feed comes along.

Image via GIPHY

What’s Constitutes an “Algorithm”?

Throughout Chayka’s book, I felt a nagging sense of dread — continually wondering if we’re really moving ineluctably toward the future he describes. But more interestingly, I also felt a nagging sense that “the feed” and algorithmic recommendations alone can’t take all the blame for the flattening he describes.

In fact, I think to primarily blame these things misses something really important. 

After all, we have relied on the presence of algorithms of some kind to help decide what culture to consume, foods to eat, and places to go for years — at least, if we can expand our notion of what “algorithms” entail.

Take a bestseller book list. What more is it than a statistical recommendation for what many may like based on what others find popular — a type of old-school algorithm? Chayka claims independent radio is much better at helping us discover new music than music feeds in something like Spotify — and I agree, to an extent. But I also really like the music that shows up each week in my Discover Weekly. I’ve discovered a number of new, favorite artists there, artists I'm nearly certain I would never come across where I traditionally looked for it.

And moreover, the type of independent radio Chayka praises consumes a very small portion of the overall “radio landscape — college radio has been struggling for at least a decade. Most, traditionally, have relied on commercial radio for new music discovery — and I worked at a radio station in college twenty years ago, and can tell you that the way we chose music was very algorithmic indeed.

Despite what I’d expected, the DJs of most large commercial radio stations have very little choice in the music they play. Many station playlists are set in advance by a Program Director, who even twenty years ago used Nielsen-type metrics to understand what was popular, and set playlists to accommodate those popular tastes.

Image via GIPHY

Nielsen ratings for television have dictated the television content broadcast for decades as well. And various other quantitative gauges of popular taste have informed cultural selection in film, publishing, and many of popular cultural formats for years. Even if a human did ultimately select and curate what others consumed, that human has very often, for years before the internet, operated in a highly algorithmic manner.

Let’s consider what may be a form of recommendation seemingly opposite the algorithmic, feed-based recommendations Chayka describes in the book — algorithms that recommend content based largely on a) what the algorithm knows about your personal tastes, and b) what is popular among people like you.

Taking away the massive, numeric quality of decisions like this made by many websites, does this latter description sound all that different from how our friends might recommend songs and movies? Ideally they a) know something about our personal tastes, and if their taste is similar to ours, they’ll b) recommend something they think we'll like.

The way in which Spotify’s Discover Weekly algorithm works, in fact, is very similar to this, and part of why many, including myself, believe it works so well. Traditional music recommender systems like Pandora took what we might call a “music-centric” approach — queueing up songs within the “classic rock” genre, say, if a user expressed a desire for that, and rarely diverging far from it. But Spotify’s Discover Weekly algorithm, despite being an “algorithm,” takes a much more “human-centric” approach. If I’ve listened to lots of songs on Spotify, the site has a sense of my musical taste — and finds others with tastes similar to mine. And then it finds songs they seem to like that I haven’t heard yet, and serves them up to me.

I promise I don’t work at Spotify! But the way this algorithm operates sounds extremely similar to how we all receive and send music recommendations to other people, doesn’t it? Ever since I first started paying attention to music — first with cassettes, then CDs, then MP3s, and now through online streaming services — I’ve had a few friends in my orbit with musical tastes similar to mine. Regardless of the format, we’ll occasionally send each other new stuff we think the other might like. Sometimes, as with Discover Weekly, we hate the recommendations — but sometimes they’re great.

Online algorithms don’t necessarily need to flatten culture, as Chayka describes. Much of what’s shaped culture in the past is algorithmic, as with bestseller lists and radio playlists. And there algorithms exist that act surprisingly human, too.

So if it’s not the algorithms or the feed that are explicitly driving the cultural flattening of Filterworld, what is?

Image via GIPHY

The Internet is Too Damn Big

Or at least, too monopolized.

I don’t mean to completely dismiss Chayka’s argument that algorithmic recommendations fed to us by large, monopolistic platforms are to blame for our cultural “flattening.” But I believe we should focus much more on the “large, monopolistic platform” piece of this than the “algorithmic recommendations” one.

One friend personally recommending a movie, TV show, book, song, or coffee shop isn’t a problem. But it becomes one when that person recommends these things to all of us. Which, arguably, is what’s happening today — we all go to Spotify for new music, Netflix for new movies and shows, Google Maps for new places to eat and drink, Instagram for vacation inspiration, and so on.

It’s not the algorithms themselves, in other words — it’s that the same algorithms are working on all of us, in all the same places, all of the time.

Decades of research and writing about the internet, in fact, suggest the same. Hal Varian, esteemed academic and now Google’s chief economist, demonstrated the tendency of large networked systems like the internet to “flatten” markets for goods. And what is culture if not a kind of market for taste, ideas, and trends? 

The internet elides the necessity of multiple local versions of the radio station I worked at in college, for instance. Because “information goods” like online radio don’t obey traditional supply constraints — no matter how many people listen to a song, it sounds just as good — it’s enough to have a single station streaming that music to everyone in the world, so long as it can be discovered (and they have the servers to support it.) “Local radio” ceases to have value online unless a station — any station that wants to gain listeners — can differentiate itself from every other playing music just like it.

In some ways, this makes the internet a fairly ruthless place to start an information business. (Um, hi 👋.) You’re no longer competing with the other radio stations, newspapers, booksellers, and — well, name almost any other shippable good — within a short distance of your customers. You’re now competing with nearly every other supplier of that thing in the world, and you have to compete with the Spotifys and New York Times’ and Amazons to build your customer base.

As might be expected from this, information networks like the internet tend to lead to monopoly control. Varian wrote about this, and Tim Wu, who later architected President Biden’s antitrust policy, described these tendencies in his book The Master Switch, too. Wu, in fact, runs through the history of various other information industries — film, long-distance telephone communication, cable TV, and more — to show that each eventually led to some form monopolistic or oligopolistic control, with one or a few large companies at the top of the heap.

Image via GIPHY

When he published the book in 2011, Wu asked if — like information giants of the past — the rising behemoths Google, Facebook and others would come to dominate the internet, the newest information industry. Over a decade later, it seems they have.

How does all this tie back to Chayka’s argument that algorithmic feeds are to blame for our cultural flattening? The feeds are designed and maintained by companies, in large part, to deal with the problem of scale. If you follow over 1,000 others on Facebook, Instagram, or Twitter, a chronological feed of everything they’ve posted means you’ll miss a lot of important, or at least interesting, stuff. An algorithmic feed that boosts seemingly valuable content becomes almost necessary once a system reaches a certain size.

It’s the scale, then, that’s the problem. Did we have the same “flattening problem” when Facebook was only a social network for students at the same university? Would we have the problem different Spotify’s existed for different types of music — a classical music Spotify, a hip-hop one, a classic rock, an EDM one, and so on? Would we see the blending of coffee shop styles across the globe if there was an explicitly American Instagram, a Colombian Instagram, a Thai Instagram, and more?

I don’t think we would. Our online life, in this world, might more closely resemble our offline one, with our local radio stations and local newspapers and local coffee shop offerings and aesthetics. Avocado toast might just stay in San Francisco.

This would surely introduce other problems, and we might not prefer that world. But the cultural flattening described by Chakya would be much less powerful, and much less worrying.

Should this Sadden Us?

As I’ve mentioned, I felt a growing sense of dread while reading Chayka’s book. I couldn’t stop wondering if we’re on a high-speed train to a very dull future.

But I don’t think we are.

For one, I believe we’ll begin to realize we don’t like this “cultural flattening” any more than Chayka does, and resist it. Chayka describes how Amazon attempted to recreate its algorithmic book recommendations in physical stores, describing at length what sounds like a truly dystopic bookstore environment. But he never mentions that Amazon Books, as the stores were called, closed permanently in March 2022. Seattleites thought the stores were dystopic too, and stopped going.

And while the internet tends to the creation of monopolistic agglomeration by a few large corporations — funneling the rest of us into the algorithmic cultures they create — this need not always be so. Monopolies don’t last forever. The Bell System held monopoly control over long-distance telecommunication in the US until 1984, when it was finally broken up. And innovation flourished after, including a 19% increase in industry patents, according to one study.

Image via GIPHY

There are already signs that many wish to move away from the massive sites dominating our online lives. Be.Real may already be losing traction among Gen Z users who first adopted it, but the app’s early popularity signals a desire by many for more “authentic” experiences online — less curated, less algorithm-driven, and less ~everyone~. Those I know who’ve used the app highlight this last point, that it’s just them and a few friends they know. It feels like an actual social network. 

This is part of the same reason many others, myself included, like using Strava. And it helps explain some of the growing popularity of Discord, which allows for customizable control to make an online community feel more like a community, something inherently small and human.

And whether you like cryptocurrency or not, much of the excitement among the crypto community centers around the idea of Web3, a theorized next version of the internet that would be less centralized, and less monopolistic. Whether or not this next internet becomes the one we use, it’s clear there’s a desire to make it happen.

Filterworld, at the end of the day, is a worthwhile book. It made me think much more deeply about the ways the internet is affecting global culture, and the ways all that’s affecting me.

But I disagree with Chayka’s theorized causes of our cultural flattening, and don’t feel quite so hopeless about it as I did while reading. After all, I’m writing this while listening to new Discover Weekly songs, and I’m liking many of them. And when that stops happening, I’ll move onto whatever comes next.

---

Song of the Week: Florence + The Machine — Big God