Filterworld: How Algorithms Flattened Culture

by Kyle Chayka

a fundamental rule of Filterworld: Under algorithmic feeds, the popular becomes more popular, and the obscure becomes even less visible. Success or failure is accelerated.

in algorithmic culture, the right choice is always what the majority of other people have already chosen.

Taste is a word for how we measure culture and judge our relationship to it. If something suits our taste, we feel close to it and identify with it, as well as form relationships with other people based on it, the way customers commune over clothing labels (either loving or hating a particular brand).

taste is a much deeper philosophical concept. It borders on morality, representing an innate sense of what is good in the world.

Consumption without taste is just undiluted, accelerated capitalism.

Taste, in the end, is its own kind of algorithm that classifies things as good or bad. The equation that it is based on factors in personal preferences, preconceptions absorbed from marketing, and social symbolism, as well as the immediate experience of a piece of culture, and eventually produces a personal answer for whether you find the thing at hand enjoyable or repulsive.

Should the human fashion editor tell you what to like or should it be the algorithmic machine, in the form of the Amazon bookstore, Spotify feed, or Netflix home page? That is the central dilemma of culture in Filterworld. The former option is mercurial and driven by elite gatekeepers, a powerful group built up over a century of modern cultural industries, riddled with their own blind spots and biases including those of gender and race. (That group includes not just the mostly white fashion magazine editors of New York City but also Hollywood producers, record-label executives, and museum curators.) Yet the human flaws may become even more dramatic in an algorithmic ecosystem when the actions of mass audiences dictate what can easily be seen. Racism, sexism, and other forms of bias are a de facto part of that equation.

Galaxie 500

“Welcome to AirSpace” for the Verge,

A place’s uniqueness only attracts more tourists, which gradually grind it into dust with the increasing flow of travelers, who arrive to consume its character as a product and leave it ever more degraded.

Every social media user has an innate sense of what will get liked. As with classical beauty or geometric proportion, the formula is inexact and yet always identifiable.

“zombie formalism”

I might unwittingly overlook the degree to which the seemingly avant-garde literary style masks their books’ much more banal and mainstream content. Upon a recent reread, I was reminded of just how much of Cusk’s Transit, the second in her Outline trilogy, consists of anecdotes about salon haircuts and home renovation. Is Cusk radically overhauling my relationship to narrative, or do I just wish I had a London flat to rebuild in a desirable neighborhood?

Instagram offered a perfect venue for the influencer: no voluminous writing or emotional disclosure of the self were required. Glossy images alone were enough to attract a following that could later be monetized. Where the term blogger described a literal activity of writing, “influencing” is closer to the financial side of what’s going on. It’s a sales job, convincing audiences to buy something, first a vision of aspirational lifestyle and then the products that make it up.

(Influencing runs on jealousy.)

It’s harder for a creator to go straight to making a movie or publishing a book; she needs to first publish her sample material, describe her vision, and gather an audience online who are engaged fans of her work. A book, for example, must first make for good tweets and then provide the material for good essays prompting public dialogue, perhaps inspiring a follow-up or an op-ed. Readers must retweet her thoughts and share pull quotes on their Instagram stories. An agent must notice the burgeoning momentum and sign the nascent author. Then finally, perhaps, a publisher will consider her manuscript—if she has accrued enough of a “platform,” that is, high enough numbers of social media followers and an ability to influence the content stream. (The choice of word is telling: people are platforms, too.) Once the book sells to a publisher and debuts on bookstore shelves, the author leverages that platform, tweeting to her fans, posting images of the cover, holding it up in a TikTok video—driving attention to its launch in as many spaces as possible. (The same will be true for this book by the time you are holding it, at least if I’m effective at it.)

“one’s ability to engage in work as an artist or a writer is increasingly contingent on one’s content capital; that is, on one’s ability to produce content not about one’s work but about one’s status as an artist, writer, or performer.” In other words, the emphasis is not on the thing itself but the aura that surrounds it, the ancillary material that one produces because of living the lifestyle of a creator.

the author’s personal brand is now all that matters; it’s the work itself that is dead.

I realized that even though I wasn’t an Instagram star with hundreds of thousands of followers, I was influential to a set group of people who were watching what I was posting, and eventually their attention did influence what I posted.

“Art will adjust to meet the attention span of contemporary life,”

To break down Filterworld, change has to happen on the industrial level, at the scale of the tech companies themselves. Decentralization tends to give users the most agency, though it also places a higher burden of labor and responsibility on the individual. It’s also the best way to resist Filterworld and cultivate new possibilities for digital life. But companies are unlikely to embrace decentralization on their own, because it’s usually less profitable. The only path for change may be to force them.

After the 2016 election of Donald Trump, the American public became slightly more aware of how we were being manipulated by algorithmic feeds. Democrats couldn’t understand how anyone had voted for Trump, given that their Facebook and Twitter feeds didn’t promote as many posts from the other side of the political spectrum, creating one of Eli Pariser’s filter bubbles, a digital echo chamber. Online, they lived in an illusion of total agreement that Trump was ridiculous. At the same time, his supporters were surrounded by content that reinforced their own views—another form of homogeneity. Recommender systems had sorted the audiences into two neat categories that didn’t need to overlap, whereas in a human-edited newspaper or television news program, some mutual exposure might have been more likely. More likely but not guaranteed: traditional media can be biased into homogeneity just as well, as in The New York Times’s unwillingness to cover the possibility of a Trump win. I recounted the criticism of filter bubbles earlier in this book; the phenomenon may have done more to cause the surprise of Trump’s win than the fact that it happened. Liberals were simply caught unaware of his popularity in part due to social media, failed to take him seriously enough as a threat, and thus made it easier for him to win.

Our feeds under a dramatically reformed Section 230 would look very different. Social networks would be forced to take responsibility for each piece of content on their sites that receives algorithmic promotion. Perhaps they would adapt by putting most content outside the reach of recommendations, meaning that users would have to intentionally follow or search for a given subject. The content that still receives algorithmic promotion would have to be vetted to ensure that it is wholly anodyne—clips of cute pet antics and feel-good news stories. The selection would be subject to a collective perception of what constituted acceptable, or neutral, news. (Such content would still be filtered, in a way, by having a singular filter imposed on it.) TikTok’s wide range of niche recommendations, for example, would be impossible. One could imagine a TikTok feed made up of only clips that would fit in the anodyne TV show America’s Funniest Home Videos. It would still be amusing, but hardly addictive or manipulative. Content moderation would have to be much stricter. Changing the balance to emphasize linear, opt-in content over automated recommendations might be a good thing, if it can limit the possible exposure of harmful material online. But this would be a much more sanitized, and by necessity slower, Internet. The problem comes down to determining which kinds of content should be able to travel so quickly and frictionlessly across Filterworld, and which should be slowed down or s...

I have hope for an Internet that’s more like Geocities, with expressions of individuality and customization everywhere, but with the multimedia innovations that have made the 2020s’ Internet so compelling. It would be a messier, more fun place—more like a playground or a sandbox than the cubicle office floor that the Internet has come to resemble.

Nothing had changed about my iPhone camera, which I still had in my pocket at all times, but without the public venue of Instagram I had much less desire to take photos. The few photos I did take were different, too: they were images that I wanted to capture for myself, more often weirder or uglier than the established aesthetic of Instagram.

“Curation is now companionship.”

One researcher found that a quarter of the songs from the Billboard Hot 100 from the 1960s to 1990s featured a key change, but only one song in the 2010s did. Where the nineties had rambling narrative songs like Biggie’s “Juicy” and Tim McGraw’s “Something Like That,” or the Streets’s 2004 concept album A Grand Don’t Come for Free, storytelling in general in pop music also seems to have lately fallen off in favor of overall vibes and moods. Lyrics avoid requiring too much attention from the listener. Streaming-era songs are often brief, too—Grimes, for the deluxe version of her 2020 album Miss Anthropocene, released a few “Algorithm Mix” iterations of songs, cutting down their run time, making them denser and more immediately compelling, better for algorithmic feeds. (It’s not dissimilar to the “radio mixes” of the past.) On average, hit songs have gotten shorter in the past two decades, decreasing a total of thirty seconds from 4:30 in 1995 to 3:42 in 2019. Data scientists at UCLA calculated that the average length of a song released in 2020 on Spotify was just 3:17, and that length is trending even shorter. The musicologist Nate Sloan has argued that the collective shortening is caused by the incentives of streaming services—Spotify, for example, counts thirty seconds of listening as a “play” and pays out royalties based on that metric. There’s no financial benefit to going any longer.

“If you are not paying for it, you’re not the customer; you’re the product being sold.” When digital platforms are free to use and make money through advertising, content is reduced to a way of attracting attention. When you are paying directly for the content itself, however, the content is more economically sustainable and tends to have more resources invested into it, which is better for both creators and consumers.

As a rule of thumb, the more directly and proportionally content creators get paid by a platform, the more sustainable it is.

unlike many tech companies, Idagio doesn’t pursue scale at all costs; it doesn’t need to expand infinitely and address every genre. As Bartlett from Criterion told me, “We don’t need to be at the level of those huge streaming services. If we have a loyal, dedicated audience, we should be able to sustain ourselves at the level that we need to be able to continue.” Beyond the technology, the capitalist growth-at-all-costs mindset is also fundamentally to blame for the flattening of culture in Filterworld.

culture itself offers a kind of algorithm to follow, as each artist influences and inspires others, referencing and building on history.

“The ultimate, hidden truth of the world is that it is something that we make, and could just as easily make differently.”