You May Also Like

by Tom Vanderbilt

  De gustibus non est disputandum. There is no disputing taste. The philosopher Roger Scruton counters, “Clearly no one really believes the Latin maxim. It is precisely over matters of taste that men are most prone to argue.” Music is an exemplar of what the anthropologist Mary Douglas called the “fences or bridges” quality of goods (or taste), unifying people even as it separates them. “It’s like religion in a way,” an inveterately hip record store owner in Greenwich Village once told me. “Why do people hate you so much because you like San Francisco psychedelic rock but you don’t like Japanese psychedelia?”

  Of course, most people not only do not hate people because they like Japanese psychedelia; they probably have no clue what Japanese psychedelia is. This points to a curious thing about taste that Bourdieu identified: The closer people are to each other socially, the more pronounced taste disputes become. The smaller the territory, the more pitched the battle. This is Freud’s famous “narcissism of small differences”; those minor variances, “in people who are otherwise alike,” form “the basis of feelings of hostility between them.”

  Part of this must simply be because taste depends on knowledge (or at least the display of it). Who but fans of the band Pavement actually care about the contrarian position that Wowee Zowee was their best album? One study, which plotted people’s musical taste on a graph, found that the people who like Philip Glass’s opera Einstein on the Beach were “located” quite close to those who dislike it. Why? Because with a relatively obscure work, disliking something entails actually knowing something about it, which puts you in a social space close to the people who like it. Take something that more people have heard, like Vivaldi’s Four Seasons, and the social gulf between likers and dislikers grows (as do the reasons for disliking). When the gulf becomes large enough, one’s dislike might even spill over into a studied kind of appreciation, which itself gains power, and a kind of safety, by its social distance from the things one normally likes. Wrote Bourdieu, “The horrors of popular kitsch are easier to ‘recuperate’ than those of petit-bourgeois imitation.”

  What does the music you like say about you? Before coming to the Echo Nest, I had partaken in one of its playful experiments called “What’s Your Stereotype?” You enter a few of your favorite musical acts and are profiled as a “Manic Pixie Dream Girl” or “Vengeance Dad.” (“Based upon your affinity for artists like: Iron Maiden.”) I was dubbed a “Hipster Barista,” which, given that much of my music listening these days occurs in Brooklyn coffee shops, seemed predictive enough. Brian Whitman, the bearded, laid-back co-founder of the Echo Nest, sounded like a latter-day Bourdieu when he told me nothing is more predictive of a person than his music preference. “If all I knew about you was the last five books you read, I probably wouldn’t know much,” he says. “But if I knew the last five songs you listened to on a streaming service, I’d probably know a lot about you.”

  Films, he suggested, are less predictive. There are fewer of them and fewer consumption opportunities. Genres matter, but there is not the same hairsplitting as with music. “They’re more directly social things,” he said. “Your wife will make you watch a movie.” Music is what people do on their own: in the car, with their headphones, via their playlists and customized stations. Preferences for it are strongly personal, and people will talk about “my music” in a way they do not about “my movies.” When people broadcast the bands they like on a social network like Facebook, research indicates they will not necessary influence someone else to like that band. They may in fact do the opposite.

  —

  In an age in which, as the Echo Nest engineer Paul Lamere described it, you can carry “almost all of recorded music in your pocket,” the question of what to play next has grown increasingly complex. Many of the people who sign up for trials on music-streaming services, Whitman said, never actually listen to anything. “They see a blank search box. What do you do?” Some people, McDonald suggested, might “listen to that Dave Matthews album, the CD of which is in a box somewhere they haven’t unpacked from the last move.” They are happy for forty-two minutes.

  And then what? Call it “Search Fright.” You sign up for a service that has everything you could ever want to listen to, and suddenly the prospect of listening to any one thing becomes overwhelming. The goal of music “discovery,” as it is called, is to steer listeners through the morass, navigating within the boundaries of the acceptable and through the shoals of disaffection. “How would you distinguish between the ten million songs you’re never going to like, either because they’re terrible or because they’re something that has no context for you,” McDonald said, “and one of the ten million songs that might be your favorite thing, if only you knew it existed?”

  Located on the other side of the computer screen, the Echo Nest faces the “cold start problem” that bedevils all recommendation enterprises: What is the first song I should play for this person whom I do not really know much about? Figuring out what kind of listener you are, the Echo Nest believes—rather than simply knowing what you listened to—is the key to keeping you engaged. It models attributes like “mainstreamness”—how far out do your tastes go compared with those of other services’ listeners? Is Radiohead thrillingly experimental for you or about as popular a band as you will listen to?

  The Echo Nest began as an effort to understand, through data and machine learning, the vast world of music by merging its two central qualities: how it sounds, and how we talk about it. A few years prior, Whitman had been recording “intelligent dance music” (“the only genre,” he joked, “happy enough to compliment itself in its name”) under the name Blitter. Like many musicians, he was finding it hard to successfully do it “at scale.” That is geek talk for no one listened to it. As he recalls, the audience “was out there but hard to find.” How could those fans be discovered and connected? Returning to grad school, he began doing work in natural language processing and thinking back to his original problem. “All these people are writing about music on the Web. There must be some way to automatically figure out what they’re saying about it.”

  The way we talk about music is, it turns out, fairly predictable. “We see people talking about its context related to everything else they know,” he said. “That’s exactly the kind of text you want.” Musicological detail is relatively unimportant; knowing the key or pitch of a song does not help guide listeners to the next song, Whitman suggested. You want to know where a band is from, what its influences are.

  The Echo Nest’s other co-founder, Tristan Jehan, meanwhile, was toiling in the world of “Music Information Retrieval,” a wide-ranging discipline that seeks to turn music into data so it can be better understood. Trying to assign an emotional valence to songs can baffle machines. Is a propulsively major-key, but vaguely somber, song like New Order’s “Ceremony” happy or sad? Computers struggle with distinguishing the harpsichord from the guitar. “It’s a plucked string at the end of the day,” said Jehan, a stylish Frenchman who, with his long, lank hair, looks more like one who performs, rather than analyzes, music. “The difference is how you play it.”

  Computers are also not very good—sonically, at least—at understanding the human classificatory system known as genre. In a sprawling project called “Every Noise at Once,” McDonald was using the Echo Nest’s semantic engines to map the world’s corpus of musical genres—everything from “Romanian pop” to “Finnish hip-hop” to “Polish reggae.” Curiously, he does not rely at all on what the genres sound like to identify them as genres (where a computer might struggle, humans can recognize a genre faster than we can say the word “genre”).

  Genres, to paraphrase the music critic Simon Frith, are as much social distinctions as musical distinctions. To human ears, there may indeed be Polish reggae; McDonald described a “Polish polka-folk melody to some of it.” And the lyrics are in Polish. To a computer, however, the distinction is murkier. There are reggae bands from Bulgaria to Omaha that would sound, in terms of audio signal, fair
ly similar. “But ‘Polish reggae’ is clearly a thing,” McDonald said, “and bands from Bulgaria to Omaha aren’t part of it, no matter what they sound like.” The Echo Nest’s computers help tell us something about music: We say we like the way it sounds, but, often as not, what we really like is what it means. And something else: Knowing what to call something helps us like it.

  —

  Lamere gave the example of Miley Cyrus, whom his then-fifteen-year-old daughter was into a few years back. Acoustically, he suggested, you could line Cyrus up with “a few indie singer-songwriters.” On paper, they sounded pretty similar. But you would not want to play, on a music service, one of those indie singer-songwriters after you had played Miley Cyrus. “The cultural impedance mismatch would just be too bad,” he said.

  What he was talking about, of course, in the wonky language of a software engineer, was perhaps the greatest machine learning challenge of all: human taste. It is humans who decide that Miley Cyrus is not appropriate to play among a group of other similar-sounding singer-songwriters. It is humans who decide what genre an artist belongs in or whether something is a genre; those genres are endlessly shifting.

  The singer Lucinda Williams tells the story that when she was shopping around an early demo tape, she was turned down by the record labels. “At Sony records here in L.A., it’s too country for rock, and so we sent it to Nashville.” In Nashville, they said, “It’s too rock for country.” As it happens, her album was eventually released by an English record company known more for punk, and it became a touchstone in the emerging “alt-country” genre, “whatever that is,” in the famous refrain of the movement’s chronicling magazine, No Depression. Her song “Passionate Kisses” eventually charted in Nashville, but only when sung by Mary Chapin Carpenter, in a version that the Echo Nest’s computers would probably have trouble distinguishing.

  Whitman confessed that while the company’s algorithms had gotten pretty good at automatically making sense of music itself—based on more than a trillion data points covering over thirty-five million songs and over 2.5 million artists—they had less of a grip “on understanding how listeners relate to that music.” So when I visited, the company was testing its “Taste Profile” technology. At its furthest horizon, this is about using music to understand people’s other affinities. In one Bourdieu-style exercise, the Echo Nest correlated people’s listening preferences in the United States with their political affiliations. As Whitman asked, “Can we tell if someone is a Republican just from his or her iTunes collection?” There were some obvious findings: Republicans more often liked country; Democrats more often liked rap.

  But other correlations were more unexpected. Pink Floyd, it turns out, is one of the bands most liked primarily by Republicans (even if the band’s members seem to be rather liberal in outlook). Whitman speculated this was mostly about the changing demographics of an aging fan base. But Pink Floyd itself changed with age, musically, and so Whitman was able to identify a split in which fans of the earlier, more psychedelic, Syd Barrett–helmed Pink Floyd tilted more Democratic. Data mining revealed other tendencies: Democrats liked more music genres (ten) than Republicans (seven); and liking for the Beatles pretty much predicted nothing in the way of political preference.

  Curiously, the least predictive of all musical genres when it came to political affiliation was metal. Loud and rebellious apparently cuts all kinds of ways. “Think about all the different ways you can get into metal,” Whitman said. The Every Noise at Once genre map lists nearly a dozen variants of “black metal” alone. Radio-friendly headbanging anthems on the one end to “church-burning metal,” as Whitman jokingly called it, at the other. They all seem united by a fierce, if free-floating, sense of “independence,” the English music sociologist Adrian North had suggested to me. Lurking down in the subgenre level of metal—“symphonic black metal,” “neo-trad metal,” “death core”—might lie more robust political correlations. Maybe there were untapped delegations of “mathcore libertarians” or “synth-pop social Dems.” Music could tell you a lot about people—once you figured out what the music was.

  —

  Bourdieu had proposed that one reason music was so predictive of people’s class, historically, was that it was hard to acquire, for example, the ability to play a “noble instrument.” Easier, less costly “cultural capital” could be found in galleries or theaters. This argument collapsed with the arrival of the phonograph. “One can hear famous pieces of music,” the composer Claude Debussy noted, not kindly, “as easily as one can buy a glass of beer.”

  Now the cost of reproduction has dwindled to virtually nothing. There is so much to listen to on Spotify that, as the Forgotify Web site illustrated, circa 2013, some four million of the service’s twenty-million-odd tracks had never been played (whatever the virtues of Desperation Squad’s “I Need a Girl [with a Car],” the world was deaf). What happens to taste in an age when most people have equal access to much of the music that has ever been recorded? As the sociologist Richard Peterson writes, “The appreciation of classical music, rock, techno, and country can hardly be expected to retain their status-making value if they are increasingly commodified and easy to acquire.” Is there anything less scarce these days than access to music?

  Of course, Bourdieu had always hinted that what you did not listen to said as much about you as what you did. Your love of opera precluded a liking of country and western. But in the early 1990s, Peterson and his colleague Albert Simkus, poring over Census Bureau data on the arts, discovered an interesting trend: From 1982 to 1992, so-called highbrows began listening to—and liking—more kinds of music, including “lowbrow” genres like country and blues.

  They called it “omnivorousness.” It was not as if all relations between music and class were vanishing. Listeners of classical were still likely to be older, wealthier, better educated. And listeners of less prestigious genres were not suddenly buying box seats at the opera. Rather, a new kind of “distinction” had emerged that was less about symbolic exclusion than about a wide-ranging, inclusionary appreciation. This might have seemed as though Bourdieu’s old categories were crumbling. Or were they? The mass media, and the Internet, were making all kinds of culture available. As the music critic Nitsuh Abebe put it, “There’s just too much music in the world to be all that sure the stuff crossing your path is officially more worthwhile than the rest.”

  The old highbrows, now under the flag of the omnivore, were redeploying their cultural capital, going wider rather than deeper, redrawing taste hierarchies with horizontal, rather than vertical, boundary lines. Being a snob could actually be socially counterproductive, lessening one’s ability to move across different social networks. The culture of the MP3 playlist—where nothing was physically owned—was less about having the right music than having the most eclectic; less about rejecting music genres outright than having “interesting” reasons for adding them to the mix (as Bourdieu put it, “liking the same things differently”).

  Bourdieu had always emphasized that the way you consumed things was as representative of your taste as what you consumed. The omnivore, as some have argued, was just taking the old highbrow strategy—the ability, wrote Bourdieu, to confidently “constitute aesthetically objects that are ordinary or even common”—and bringing it to heretofore excluded musical genres. In places like the personals section of The New York Review of Books, a watering hole of upper-middle-class educated taste if ever there was one, omnivorousness is as reliably encountered as the love for nature walks and France. To take an example at random, from the issue (pardon my cultural capital) at my side: “Affinity for great food, independent film, intriguing travel, chamber music, jazz, rock…” Subtext: My tastes are as wildly adventurous as I am, but they are still tastes. One struggles to imagine a cardiganed and horn-rimmed reader of the 1950s announcing that he liked Bill Haley as much as hard bop and Brahms.

  It was not, suggests the sociologist Omar Lizardo, that omnivores really loved all that new
music but that they could maintain a weak, wide-ranging appreciation, a number of small pots on a low boil. After all, liking things takes time. Not only do people consume music, notes the sociologist Noah Mark, but “musical forms consume people.” The more you like one genre, the less time and energy you have to like others. At the Echo Nest, Whitman found flickers of omnivorousness: “We modeled listeners where we have a rich taste profile, and then compared their behavior for a week to station profiles for the top 12 radio formats. In a given week, the average on-demand listener went across 5.6 listening formats.” Of course, there is probably self-selection going on here: The most omnivorous people are going to want the huge variety offered by online services. Most people still seem to want the hits: By one analysis, 1 percent of artists accounted for 77 percent of all income from recorded music.

  Even omnivores, however, need to leave room for their dislikes. When the sociologist Bethany Bryson looked through those same Census Bureau taste data in 1996 but focused on what people said they disliked, heavy metal and rap (per the bureau’s rather crude classifications) were least favored by the most “tolerant” subjects. This should not have surprised: After all, those were the genres least liked by people in the entire survey.*5