Mix Tape

Spotify Automates; Humans Get Played

There’s a band called Houndmouth that represents everything I loathe about “indie” music: a bogus bohemian aesthetic that consists of long hair and torn flannels, a contrived rootsiness, hats with very wide brims, and a misplaced nostalgia for the alleged good ol’ days of train hopping and blacksmithery.

But what disturbs me even more is that I can’t stop listening to their songs.

I should know better. Houndmouth’s music often sounds like a pop-folk knock-off of Bob Dylan and The Band. They’re like the Alabama Shakes, but less talented; like Mumford and Sons, but less famous.

Despite my attempts to reason with myself (but the long list of musical clichés!), my head compulsively nods along—an instinctual reaction that must come from a much deeper place than a simple appreciation of the music.

Take, for example, the first Houndmouth song I ever heard: a stomping ballad called “On the Road.” That song shares its title with a Kerouac book long one of my favorites. When I was 19, I got a tattoo to prove it. The quote I chose and that song each evoke a similar brand of beatnik wistfulness: “The only people for me are the mad ones….” No doubt Houndmouth’s four members likely consider themselves “mad to live” as well.

Houndmouth and I are connected by geography, too. Despite the Southern affectation in many of their songs, Houndmouth is a Midwestern band. Each of the band’s four members was born in New Albany, Indiana, two hours south on the Ohio River from Dayton, where I was born. (The image on a Houndmouth album cover depicts this river in the 19th century, with horse-drawn carriages plodding down the bank alongside floating steamboats.) They’re Midwestern kids yearning for the open road, just like me.

Political identity—at least in a very general sense—might also endear them to me. What limited knowledge I have of the band’s politics tells me that we’d likely agree on at least one point: a basic distaste for bigotry.

The extent of their activism seems to be an appearance on Late Night With David Letterman in 2015, five days after then-Indiana Governor Mike Pence signed a “religious freedom bill.” Before the band took the stage, Letterman had been excoriating Houndmouth’s home state, which is his too. The band made clear that they also disagreed with “all the shit that’s going on back home.”

I like to think of myself as someone who would never have sought out a Houndmouth record; as someone with preferences deep and complex who likes artists that make meaningful music. Houndmouth makes decidedly meaningless music. Yet I dig it.

The revelation here seems pretty obvious: I’ve been lying to myself. In this respect, I’m completely normal—one in millions of cognitively dissonant listeners who profess a more refined sensibility, one in millions who never would have sought out a Houndmouth record. Clearly, we don’t know ourselves.

It was Spotify that told me I’d like Houndmouth. I didn’t believe it. I hadn’t before sampled any of the streaming platform’s recommendations. My taste, I would contend, is far too subtle and sophisticated to be sussed out by an algorithm. Spotify, however, is out to prove me wrong.

In July 2015, Spotify introduced its “Discover Weekly” feature, the platform’s algorithm-driven recommendation service that compiles 30 songs that “you might like” and delivers them to you in playlist form every Monday.

Here’s how the platform’s website breezily describes its trademark feature:

At Spotify, we’re all about discovering new music. To make this easy for you, we’ve tailored music recommendations specifically to your taste. Our suggestions are based on your listening habits, so why not head on over and discover that new favorite jam?

But, in reality, it’s not quite that simple—nor, for the technoskeptics among us, is it that innocent.

For all the praise heaped on it (“groundbreaking!” “revolutionary!”), the foundation of Discover Weekly is a piece of (relatively) old technology—a process known as “collaborative filtering.” Essentially a mathematical technique, collaborative filtering dates back to the early 1990s when computer scientists John Reidl and Paul Resnick developed a tool that recommended articles on Usenet, an precursor to the chatrooms and forums on today’s Internet.

Amazon was one of the first major adopters of collaborative filtering, and its “if-you-like-x-you’ll-like-y” product suggestion remains the most well-known example of recommendation algorithms online today.

But Spotify takes it a step or two further. And here’s where I’ve had to concede to yet another reality that chips away at my sense of individualism: I am more of a data point than I care to admit.

First, Spotify’s algorithms scan the two billion or so playlists (generated by the service’s 100-million-some users and by Spotify employees who call themselves “editors”), noting which songs appear together. If the algorithms notice that two songs I’ve played often appear on playlists with another song I haven’t heard, Discover Weekly might suggest I listen to that third song.

Spotify then pairs this data with each user’s “taste profile,” which compiles a head-spinning amount of information in an attempt to define a listener’s taste. The algorithms behind the taste profile not only know which songs I listen to, but when and where I listen to them—in the morning on my way to work, for example. They also use readily available attributes like age and geography (from which Spotify can guess political ideology) to further profile me.

The architect behind the taste profile technology, Brian Whitman, talked to The New Yorker in 2014 about going even further. He said that Spotify’s programmers are hoping to learn even more about their listeners, data like “what the weather is like, what your relationship status is now on Facebook.” (In 2011, Facebook entered into a partnership with Spotify.)

Whitman added, “We’ve cracked the nut as far as knowing as much about the music as we possibly can automatically, and we see the next frontier as knowing as much as we possibly can about the listener.”

Now, it’s not exactly clear what sort of user data Spotify and Facebook share with one another, but Whitman’s giddy quote makes one thing clear: Spotify wants to know everything about you.

To some, this might not be disturbing at all. Indeed, Spotify’s recommendation system might seem like the next step on a path towards a technological utopia in which our every desire as a consumer is anticipated and then satisfied.

In a breathless piece titled “The Magic That Makes Spotify’s Discover Weekly So Damn Good” a Quartz writer marveled at the service and, without any hint of irony or trepidation, wrote, “After I received several excellent playlists in a row, I couldn’t stop thinking about how Spotify had figured me out, along with 75 million other people.”

But, for me, this sentiment raises a few important questions: what does it mean to be “figured out”? and, are we OK with programming machines that have the ability to do it? what will they figure out next?

The author of that 2014 New Yorker piece, John Seabrook, raised a similar question: “Are you playing the music, or is the music playing you?” Or put a more existential way: are you sacrificing some of your humanity by succumbing to the algorithm?

We’re an intensely creative species and listening to music, Ben Ratliff argued, is “a highly creative act; but a little less so, I think, when you let the computers do the choosing for you.”

Mix Tape

iStock.com / MarkPiovesan

Ratliff is a music critic for The New York Times and author of the book Every Song Ever. To the question of whether Spotify chips away at the human experience, Ratliff would answer yes: “Through listening to music you learn how to walk and talk and move and have relationships and speak different languages,” he says. “You pick up things from music that really increase your humanity.” Do you really want a robot in control of that?

Ratliff returns often to this idea that Spotify and its algorithms are actually impeding his sense of discovery—an outcome that, given the name “Discover Weekly,” is ironic. He compared using Spotify to browsing a record store, writing that “In stores and libraries you can see the stock in context; you can move through stacks and sections without exactly knowing where you’re going; learn enormous amounts by accident.” Whereas, with Spotify and comparable platforms, “You can’t see anything except what the service wants you to see, and so you have no awareness of what’s in the back of the store, or even that there is a back of the store.”

I would add that we’re losing out on something else, too. Not only is our sense of “discovery” impacted, but it’s also depriving us of a human interaction that can be quite powerful.

In my teens, I formed and maintained some of my best friendships through the exchange of burned CDs—playlists before there were playlists. We took great care with the music we swapped. The order of songs had to be just right, the song choice eclectic enough to maintain interest and to turn the other on to a new artist.

I still have these, bouncing around the floor of my car, in a closet at my parents’ house, in a box piled next to a dusty Walkman. The CDs address me, in colorful marker, as “R3!$”or they alert me to the “jamz” or “trax” they contain, always teetering on combustion (hot! fire! flames!) or overdose (dope! sick! ill!).

But their contents meant something. They were part of a conversation we were having. When a friend gave me a CD, he was saying “I like this, and I think you might, too.” There was part of him and part of me. And even if we didn’t agree on every song, we knew there was another person behind the suggestion and, quite often, shared feelings or experiences.

“Remember spray-painting your room to the Red Hot Chili Peppers? We wrote that ‘Soul to Squeeze’ lyric on your ceiling.”

“The other day I got high listening to this band Stardeath and the White Dwarfs, they’re like a new age Flaming Lips. Next time you smoke, listen to ‘The Birth.’”

Sometimes we took a journey through time. We liked MGMT and Nirvana, so we set out to see who laid the foundation for their songs. We’d end up exchanging old music from Brian Eno, Pink Floyd, and the Velvet Underground.

Last week, I walked into a record store near my home in Berkeley, California. I wanted to find a place where this conversation was still going on. I also wanted to see what else I’d run into if I came here searching for Houndmouth.

A young guy named Kevin showed me to the “H” section of the store’s rock collection. Houndmouth actually had a tab all to themselves, but (improbably) their albums were sold out.

“I don’t really like Houndmouth,” I said to an apologetic Kevin. “At least, I don’t think I do.” I began to explain myself, and we started talking. We talked about Spotify, and we agreed.

“Say somebody likes a certain sound, a certain drum sound, or a certain guitar sound, or they like music with lots of reverb that’s dreamy or shoegazey, you know anything like that,” Kevin said, as I looked down, trying to imagine what “shoegazey” sounded like. “I think there’s a lot of details to music and to art that an algorithm won’t necessarily be able to cover.”

And then, we just started talking about music. “I love this drummer, Bernard Purdie, and this bass player, Chuck Rainey,” Kevin told me. “I’ll buy anything they play on.”

I told him what I’d been excited about recently: a weird mix of classic Patti Smith, a newish Brit band called The Struts, and a remastered record from the late Detroit hip hop icon J Dilla.

“You might actually like this,” he said. And right there, in the H section of Amoeba Music, adjacent to the empty Houndmouth rack, Kevin recommended an album: To Bring You My Love by PJ Harvey.

The anti-algorithm arguments really come down to this: If we allow ourselves to think we’ve been “figured out” by these algorithms, then, perhaps, we will eventually become simple enough for this to actually happen.

Spotify is like a music GPS. And my worry is similar (“Kids these days, they just don’t know how to use a map, let alone a compass”). We begin to use the GPS because it’s convenient, it’s easy. We can do it quickly whether we’re in the car or walking. We don’t have to talk to anyone, or ask for directions. But then we start to rely on it. We forget about detours, about meandering. We go where it tells us to go. We have the map but we never learn the territory.

And here, I have two fears, one slightly less batty than the other. My first, the classic “what if” of the I, Robot brand of technoskepticism, is that our trusty GPS will go rogue and send us right off a fucking cliff. This hasn’t happened (yet). My second already has: our GPS will send us to the same place it has sent everyone else. And then it will give us the illusion we got there ourselves.

Now, will Spotify be used in a corporate scheme of world domination to stream propaganda into our headphones, so that every dreary traveler on a Manhattan-bound A Train, with head down and white plastic earbuds in, simultaneously snaps to attention and repeats “One of us” until everyone is listening to the same song? (the aforementioned “off a fucking cliff” scenario). Well, maybe—but probably not.

But imagine the other scenario: everyone is listening to the same song, or same type of song, over and over. Spotify’s algorithms notice this and keep suggesting this sort of music. Then, in this process of artificial selection, we find ourselves caught in a double mirror of “songs you might like.” (I like to picture a herd—sheep with earbuds, all bobbing their heads to “Shape of You” by Ed Sheeran, Spotify’s most-streamed song with nearly 1.6 billion plays.) This sort of evolution is already under way. Can we stop it?

I often listen to music while I write. As I worked on this piece, I thought it might be fitting to give Discover Weekly another try (I hadn’t dared return since the Houndmouth incident).

Toward the end of the playlist, I heard something I recognized: it was a two-part harmony—sort of folksy, inflected with some twang. The artist, though, was unfamiliar, a band called Josiah and the Bonnevilles. I navigated to their page and clicked on the tab for “related artist” (akin to Amazon’s recommendation system). This led me to a group called Great Caesar. Their page led me to The National Parks, which directed me to Run River North, then on to a band I recognized, Lord Huron.

Here, on the list of artists related to Lord Huron, I found what I was after: Houndmouth. I wondered how many bands it might take me to get back to Josiah and the Bonnevilles, when I would prove that Spotify’s recommendation engine is nothing but a cyclical trap that will eventually destroy our desire for diverse music. I wondered.

But that’s no way to discover. Instead, I closed my computer and walked down the street, in search of some PJ Harvey, or maybe Bernard Purdie. Something I hadn’t heard yet.