Thirty-three-year-old Josh Rosenfeld isn’t just the prototypical indie-rock kind of guy—white, lanky, hoodie-wearing—he’s an über–indie rocker. As the co-founder of Barsuk Records, he’s helped launch the careers of Death Cab for Cutie, Nada Surf, and Menomena. He’d probably be the first person you’d go to for recommendations about new music—if he weren’t so frazzled about the whole thing himself.
“Right now, as we talk, I have 30 CDs sitting on my desk that have been sent to me, in many cases by people whose taste I trust. But I just don’t have time. There’s way too much music to absorb,” says Rosenfeld. “There’s no way in hell I could do it.”
If Rosenfeld can’t keep up, what chance do the rest of us have? The digitization of music has caused its availability to explode, as entire albums can be recorded, uploaded, and transmitted around the world in a single day—to say nothing of individual songs. The result? Anyone online has a seemingly unlimited supply of new music.
“That’s great in a way,” says Rosenfeld, “because it’s very democratizing, but it’s also a disaster from the perspective of the paralysis that ensues from that much freedom of choice.”
So if you wanted to learn about new music, where would you even start?
The Internet, which has largely been responsible for this torrent, is now attempting to solve the problem it helped create. The perfect music-recommendation technology is equally tantalizing for music listeners and makers. If the computer could figure out just what kind of music we’d each like, labels and bands would know who would be most receptive to their music, and listeners could blissfully discover one new favorite band after the next. Labels certainly need something to happen: Industry data-gatherer SoundScan reports that so far in 2007, overall album sales, which include digital downloads, are 10 percent behind 2006, with 118 million albums bought, versus last year’s 131 million.
There has got to be a way to cut through the noise. And naturally, it started with a Belgian computer scientist.
In 1989, Pattie Maes moved to the U.S. from Belgium, her home country, with her Ph.D. in computer science, and wanted to find new music that she would enjoy. She transferred from the artificial intelligence department at MIT to the university’s Media Lab, where she helped develop Firefly. Originally known as Ringo, it sought to automate word-of-mouth music recommendations. Early users e-ranked 125 artists on a scale of 1 to 7; algorithms calculated the compatibility of each of the 1,000 users based on the number of artists they had in common. Users were then e-mailed a list of recommended albums from their highly compatible mates. As its popularity exploded, the increasing number of users made the results smarter. Firefly made Maes a millionaire, and earned her a place on Newsweek‘s 100 most important people to watch, the World Economic Forum’s 100 people to listen to, and People’s 50 Most Beautiful People in the world.
Firefly Inc. then went where so many promising startups have gone to die: It was acquired by Microsoft in 1998, and integrated into its less-than-world-changing Passport Web ID product.
The take-home lessons from Firefly—that the intelligence of any user-based system is directly proportional to the number of users, and that you can use math formulas to find art people might actually like—remain the simplest and most frustrating touchstones in the world of recommendation technology. The same concepts are used today to get a sense of what you’ll buy on Amazon.com or rent on Netflix. What it’s saying is that your tastes are not created in a vacuum, and that there are other people who like what you like. Somewhere is a computer with several gigabytes of music on its hard drive; it has every single album you own—plus two more. Who wouldn’t want to know what albums those were?
One of the newest entries into the science of sound selection, iLike, was launched a year ago in Seattle by 34-year-old twin brothers Hadi and Ali Partovi. Ali, who handles the “sales and marketing side of things” from the San Francisco office, is a recreational guitar player who helped found GarageBand.com, a site for promoting unsigned bands. Hadi was most recently with Microsoft. Both brothers hold master’s degrees in computer science from Harvard. Their new project melds the online friends-making mechanics of MySpace with the obscure-band infrastructure of GarageBand.
“Our goal is to facilitate social music discovery,” says Hadi Partovi.
The iLike offices on Capitol Hill bespeak a typical dot-com startup: Just past the perky receptionist is a tranquil sea of white walls, nubby carpeting, and twentysomething computer engineers whose desks are filled with dual oversize monitors and cans of Mountain Dew. Inside the conference room, a dry-erase board is covered with numbers, grids, and T-shirts bearing the iLike logo.
Also typical is a seeming disregard for profit. This place is so Web 2.0 that even revenue is a distant goal. “We are focused on growing users and usage,” says Hadi Partovi. “We will get to revenue as the next step.” Right now, the only source of income is “affiliate fees” when users purchase music from iTunes or Amazon through iLike. But somebody has faith in the company’s profit-making potential: Ticketmaster invested $13.3 million for a 25 percent share of iLike last September. Not surprisingly, one of the next services on the site will be alerting users to when their favorite bands are touring, and providing links to Ticketmaster.com.
“I think if we can get to a 10 million user base, we’ll be a real player in terms of how music gets found and discovered,” says Hadi Partovi. If so, he’s got a good ways to go: Right now, iLike has about a half-million users—and a ton of competition. There are at least a dozen sites trying to do roughly the same thing. And no wonder: The Berkman Center for Internet and Society at Harvard estimates that by 2010, 25 percent of all music purchases will be guided by “consumer taste-sharing applications,” or music-recommendation Web sites.
LET’S BEGIN: HOT OR NOT?
iLike starts out the same way as did Firefly and most other music-recommendation Web sites. Once you join, pages of bands and musicians are flashed before your eyes, begging you to say whether you like them or not. Of these, you select your 12 favorite artists, and the site begins building your profile of musical tastes.
Next you download a sidebar that attaches to your copy of iTunes and registers your entire musical library, including your play counts. Then it throws up your profile page (but not without first getting a photo!) that displays your top 12, what bands and songs you’ve listened to most, what songs you’ve most recently played, and what you’re listening to now. After analyzing your iTunes library, iLike will also sign you up for a few groups on its sister site, GarageBand; I was immediately signed up for the Indie Pop, Indie Rock, and Experimental Rock groups, and a list of “hot tracks” from unsigned bands in those genres was delivered to my inbox each week.
In addition to user pages, iLike has pages for each band, listing their most-played songs, all the iLike members who’ve professed an affinity for that band, and which members have listened to that band more than any other member. (Big shout-out to Austin, the man who’s listened to Slayer more than anyone else, and to Dean D, who manages to be the top listener of both Weird Al Yankovic and Avril Lavigne.) Each band page also lists a few “related artists,” based on the bands that most commonly show up next to each other in listeners’ libraries. For example, Pantera, Metallica, Slipknot, Megadeth, and Sepultura are recommended to those who visit Slayer’s page. (Obscure or newer bands, who don’t show up on as many people’s iTunes, get less love; Peter Bjorn and John are related to no one.)
So, do all Slayer fans really like Pantera? These “related if you like” recommendations are “a good shorthand to start talking about music,” agrees Ruben Mendez, leaning over the counter at Capitol Hill’s Sonic Boom Records. Mendez, a tall, soft-spoken clerk who’s been at Sonic Boom for six years, agrees that it’s a bit simplistic. “But there’s so much music, you need to start somewhere.”
iLike works best if you remain interactive and befriend other users. The system will generate a list of “musically compatible people,” or people with similar libraries and top-12 rankings; then it’ll send you recommendations and updates based on what those soulmates have recently added to their libraries. The more friends you make, the more info you get.
Did you just listen to a song and love it? You can go to your profile, click on a button, and now that song is recommended to every one of your friends—or just one. Did you just copy a Paul Simon CD into your library? iLike will tell all your friends. Do you have a question for the top fan of Frog Eyes? Send it. (You can listen to 30-second clips of most songs on the site; for the unsigned GarageBand participants, full MP3s are available.)
My profile begins to take shape, and soon I’m getting pegged into that Radiohead/Beck/Beastie Boys group. Eventually, R.E.M. appears in the list as one of the bands for me to rank. But what if I like early R.E.M., but not late R.E.M.? How would I know that the R.E.M. I like is the same R.E.M. that someone else likes?
“The [mistake] there is that most people are not artist-oriented, they’re song-oriented,” says Daniel Levitin, author of This Is Your Brain on Music: The Science of an Obsession. Levitin is the creator of one of the first music-recommendation Web sites (MoodLogic), a former music producer, and currently a professor of psychology and electronic communication at Montreal’s McGill University. “You can have an artist with a very wide range of sounds, and a lot of people are interested in particular songs; they’re not interested in everything the artist makes,” says Levitin. You might like early Beatles music without being a fan of “Revolution 9.” Sites like iLike assume, perhaps correctly, that the days when a band’s sound could radically evolve over time are over.
And what about my own evolution? The playlist count on my iTunes only goes back to when I bought my computer a year and a half ago. Last.fm, a London-based site, is one of the only recommendation engines that charts your listening history. After all, without a sense of your progression of tastes, iLike has no idea what music you’ve already run into the ground and pray will never be recommended to you again.
I didn’t want to start off comparing my tastes with anonymous others’, so I e-mailed my friends and urged them to join. Because he had to, my boyfriend said yes and went through the process of rating various bands; soon after, I signed on, clicked on his profile photo, and waited as iLike computed our compatibility.
LOW.
What? Wait: Don’t we actually love the same music? Wasn’t our freakishly near-constant overlap of tastes, our love of music and going to shows, one of the ways we initially bonded? How the hell did this happen?
I saw that, in rating bands, he’d erred on the side of clicking “iLike” more than I had; it was enough, for him, to enjoy some of the band’s music, whereas I was intent on choosing bands if I enjoyed most of their music.
Do I want to profess that I like Justin Timberlake, even if I only know his latest album? This binary business is too simplistic. And where the hell is the “iHate” button for Good Charlotte?
SIDEBAR: ANALYZE THIS
As with most online commerce, there’s a trade-off between privacy and efficiency; the more you allow the software to track your listening habits, the more thorough and accurate a profile iLike creates. Whenever I play a song on my iTunes, it shows up on my iLike profile page, within seconds.
The day after I downloaded my iLike sidebar, I awoke to the following e-mail: “You’re listening to the Rushmore soundtrack. Good choice!” In Hadi Partovi’s mind, this should elicit a kind of bond-building cyber-high-five, but it’s just creepy—despite the fact that it came from a friend I’ve known for nearly 15 years.
“Well, this generation doesn’t think anything of putting things up on the Internet,” says Partovi, after I mention that broadcasting my history took me by surprise. “Songs are the least of it.” It’s official: At 26, I’m a dinosaur.
“Well, do I open up Windows Media Player when I’m listening to some crap music?” jokes Adrienne Massanari, a 31-year-old doctoral candidate in communications at the University of Washington and co-editor of Critical Cyber Studies.
Massanari is as animated as her bright-red hair with dark roots and wears Converse sneakers and a black blazer over what she calls her “supergeek T-shirt” (math joke). She’s a fan of Menomena, Arcade Fire, Broken Social Scene, Iron & Wine, and Sufjan Stevens. She’s also one of my digital taste soulmates: In iLike’s eyes, we’re highly compatible.
“The interesting thing about iLike,” she says, “is that it becomes a place to show off your musical tastes. It’s very much about presentation of self, where you say, ‘I’m listening to this indie band that no one has heard of.'”
After e-mailing my friends, urging them to join iLike, one wrote back: “Is this one of those Web sites that tells the whole world you’ve been listening to Hall and Oates for three days straight?” (He didn’t join.)
In fact, iLike—and other sites like MyStrands and last.fm—lets you hide bands you play from your profile entirely; it’s as if that Britney song never happened.
The 10 most-hidden artists on iLike read like a list of the world’s guiltiest pleasures: Britney Spears, Kelly Clarkson, Michael Jackson, Avril Lavigne, Black Eyed Peas, Christina Aguilera, Backstreet Boys, Justin Timberlake, Hilary Duff, and the Pussycat Dolls. The problem is that if you choose to hide certain artists you listen to, they won’t count for your compatibility ratings or preferences at all. You and your taste soulmate might be hiding the same thing.
Another problem I encounter is iLike’s failure to distinguish what music you like most from what music you play most. Some of my most-played bands on the computer aren’t actually my favorite bands, they’re just bands with short songs, or instrumental acts I listen to while reading and writing. As a result, iLike lists me as the top fan of both Caribou and the Octopus Project. But I’m attracted to their lack of distracting words—not just to bands with random animal names.
Needless to say, my taste soulmate Massanari has the same experience. “I have very specific music that I listen to when I’m writing,” she says. “I can either listen to things that have no words—so I listen to a lot of weird electronica that’s pretty ambient—or I can listen to world music, because it’s in a foreign language.”
A few of my iLike “taste soulmates” are users with whom I only share a few favorite bands…with lots of undesirable extended family. Since I love the Canadian band Broken Social Scene, I’m deemed compatible with other BSS fans, as well as with fans of BSS offshoots, including Feist, Stars, and Jason Collett. I like Feist, but Stars I find cheesy and melodramatic, and Jason Collett puts me to sleep. BSS needs all of its 10 players to produce that undefinable 15 percent, that unquantifiable “n” factor, that distinguishes the music I like from the music I don’t. It’s where the shorthand of RIYL, or “related if you like,” starts to show its weakness, when unquantifiable characteristics—the chemistry between two singers; holding a special place in your heart for ELO because of your first kiss—make or break a band.
The problem with automatic recommendations, Levitin says, “is that none of them really take into account you as an individual. They’re all looking at correlation matrices of what other people like, but they’re ignoring your individual taste.” There are probably plenty who hear Jason Collett, jump for joy, and walk away after hearing Broken Social Scene. But these are not people I’m friends with.
SHAKY COMPARING
iLike’s recommendations are all based on what’s called “collaborative filtering,” a well-worn method of predicting consumer preferences based on the past behavior of people who’ve displayed similar tastes and interests. It’s how Amazon decides what books and products to push at you. But at least one content-specific site eschews this strategy in favor of a more detailed musical analysis. It takes 20 to 30 minutes for the staff at Oakland, Calif.–based “music discovery service” Pandora—most of whom are professional musicians—to analyze each song in over 40 categories, such as time signature, instrumentation, and tonality.
So, what if I liked the soft, indie sounds of Fruit Bats (a Seattle band on Sub Pop) and wanted to find new music? Pandora would use that data to calculate the typical Fruit Bats sound before playing other, similar songs.
“I’m kind of nervous,” says Eric Johnson, lead singer of the Fruit Bats, as I tell him that I’m inputting music by his band into a few recommendation Web sites. Pandora identifies the Fruit Bats as having “folk influences, a subtle use of vocal harmony, and major key tonality,” before sifting through its database and playing a song that’s been identified as similar, without regard to genre. (Pandora is best known for this total disregard for genre specificity.)
The first song that pops up is the title track to Hedwig and the Angry Inch. I start to laugh, and so does Johnson: “So that’s an off-Broadway musical, right?” He listens for a few seconds before issuing his decree: “No.” They do sound somewhat similar—sort of—but the stigma of a musical may be too much for Johnson.
Johnson recalls for me how he used to discover music. “I had this girlfriend whose older brother was a super-duper, the ultimate; he was a 5,000-LPs kind of guy, and he was like my Internet. I would call him and be like, ‘OK, I like that one thing that you gave me, so what else would I like?’ and he’d give me five other things. He introduced me to so much, but he was one man with an Internet-like database of music.”
Next, I plug in the Fruit Bats song “Silent Life” into another content-specific Web site, Vancouver, B.C.–based Audiobaba. A song by the band Great White pops up, “Maybe Someday.”
Johnson perks up: “I love Great White. When I was a kid, I went to their concert in seventh grade. Maybe they had some sort of formative impact on me that you can’t hear, but this computer can,” he says. “Maybe there’s some hidden influence that this program can pick up.”
By contrast, on iLike, the completely random, blind nature of the process means that listeners are in charge of rating music that’s well outside their area of expertise. Hadi Partovi walks me through the process of listening to and tagging the unsigned bands that appear on the iTunes sidebar: You hear a song, rate it, and input what other musicians it sounds like. After listening to one fairly generic band that’s a bit heavy-hitting, Partovi says he likes the production quality and thinks it sounds like Jason Mraz…who also happens to be Partovi’s second-most-listened-to artist.
“It certainly isn’t perfect,” says Partovi, responding to my comment that you can only talk about music in terms of what music you know.
NO INTERNATIONAL CHARACTERS
For the most part, on iLike, I’ve been finding music that either sounds similar to music I already own or fits within what Amazon would call my “purchase circle.” When I play Modest Mouse, the iLike sidebar suggests I also try the Shins, Bright Eyes, Radiohead, Death Cab for Cutie, and the Pixies. I already own albums by all of these groups, but, geez, I’m not taste soulmates with every hipster. Am I?
“The same stuff keeps popping up on those sites,” says Barsuk’s Rosenfeld, correctly. The “tyranny of the majority” criticism of collaborative filtering is true. It makes sense from an economic standpoint—it’s safer to err on the side of popularity and recommend the better-selling album—but doesn’t take into account my need to find nonmainstream indie music.
What if I want to branch out and listen to music that sounds nothing like what I already own? Pandora won’t help me here—it actively searches for music that’s similar to what I know I love. Besides, I might not even know what kinds of music I’d like.
“That’s it! That’s exactly the problem,” says Levitin. “If you’re trying to do what computer scientists call bootstrapping, and bootstrap some list based on what you already know, it’s going to tend to be myopic. I think that’s a good start, but to end there is a mistake.”
“What if I really like rockabilly music?” says Massanari. “It’s overwhelming to think how much music in other genres I’m missing.” Some recommendation Web sites, like Goombah, Audiobaba, and last.fm, have “adventurousness” knobs that help you veer from the tried-and-true tyranny of the majority. But even then, I have to go to “obscure” mode before a list of artists comes up that I’m not already familiar with—and my list of suggestions dwindles in proportion to their album sales.
How to escape my indie bubble? I decide to try sliding in through a different genre. The actual user demographics indicate that iLike users are “mostly teens at this point,” admits Partovi, thus explaining the glut of pop music on the site. Jazz aficionados have yet to weigh in. Playing John Coltrane evokes suggestions of Charles Mingus, Miles Davis—and Tchaikovsky. While I can readily envision those instrumental tracks in one of my “Music for Reading” playlists, when presented without context, it seems bizarre and sloppy.
And lovers of Syrian pop music are also nowhere to be found. The millions of Omar Souleyman fans out there will be disappointed to know that when you play his hit “Alkhatiba Zaffouha” on your iTunes, iLike will leave you at a dead end. Not only does it have no recommendations, it doesn’t even recognize the artist.
“Right now, for music with certain international characters in the name, it doesn’t register properly,” wrote Partovi via e-mail. “I’m not sure if that’s your issue, but it might be.”
The recommending savvy of most other services is rendered moot when confronted with world music. After playing Indian Soundscapes, MyStrands told me: “Sorry, we could not match this track with others in our catalog. Please enjoy this general recommendation: The Fugees, ‘Ready or Not.'”
The iLike sidebar had nothing to recommend for Indian Soundscapes. And of that social network of teens, none had copped to enjoying songs like “Chowpatty Monkey Trainers.” My foray into world music is short-lived.
In general, the bands I don’t recognize that do pop up on my iLike sidebar are the unsigned acts, courtesy of GarageBand. But they sound little like what I’m listening to, and nothing like what I want. Random bands from Audiobaba, MusicIP, Pandora, and MyStrands have been recommended to me, but even when I focus my efforts—zoom in on a single band or Web site—trying to comb through the slosh drives me crazy. It’s just another random, unanchored band with an OK-sounding single.
“It’s one song from a band I’ve never heard of,” agrees Massanari (of course). “There’s no information, no context….It doesn’t mean as much as a recommendation. Nobody took the time to say why I should listen to it.” If I want an explanation, I’d get better recommendations from John in the Morning.
How do professionals, like Barsuk’s Rosenfeld, know what to look for? That answer requires a bit of a story, starting in the early ’90s. Rosenfeld says he and Barsuk co-founder Christopher Possanza were sending out demos for their band, This Busy Monster.
“Nobody was getting back to us. I remember having this feeling that indie rock in particular was this super-cliquish, exclusive insiders’ club, and we didn’t know the right people to get inside. We would play shows with bands who were good, and they would get signed to Sub Pop, and it always seemed like they knew someone. I have come to realize how that happens, and that it’s not cliquish, exactly. How I hear new music, and how we at Barsuk hear new music that we wind up really liking, and deciding who we want to work with, is almost always through bands that we work with…because somebody in Death Cab liked them or somebody in Nada Surf sent us a couple songs—someone we know, whose opinion we trust,” he says.
While Rosenfeld’s explanation of how he finds new music sounds oddly like the method he once railed against, he’s got a point.
“I just talk to my friends,” shrugs Sonic Boom’s Mendez.
Or, in the case of iLike—”friends.” “It’s pretty neat—if you have a friend who has better music tastes than you or is more active at discovering music, you can draft off of them,” Partovi says. He estimates that because he’s “quite active,” between his 500 online friends, he’s received recommendations for 200 individual songs within the past few weeks alone. But I don’t want 200 songs—and I certainly don’t want 500 friends. I just want a few that I love.
BACK IN THE SADDLE
Unbeknownst to my iTunes, I spend an entire weekend surrounded by music. I attend a friend’s CD-release show, pick a few hours’ worth of music for a party, and burn my boyfriend four CDs, including Wolf Parade’s Apologies to the Queen Mary. I listen to it for the millionth time, fall in love with it all over again, and buy the band’s earlier EP. (Lesson learned: I could have started finding new music just by looking at what I already own.)
I get on iLike to find similar artists, but discover that I own them all. I research the band on last.fm and learn that Wolf Parade’s singer/keyboardist, Spencer Krug, has a side project, Sunset Rubdown. It’s not one of the bands that’s related to Wolf Parade, and it’s too obscure to come up on the iLike sidebar at all; you have to know what you’re looking for. One of Sunset Rubdown’s top listeners on iLike, Ray Y from New York, may very well be my taste soulmate—our libraries are nearly identical. I see what else he has, and find Animal Collective—my umpteenth such recommendation of the day. And Beirut, which I’ve heard plenty about but have never listened to, is one of his favorites. Why has no one made me listen to these bands before?
Even though it’s still squarely in hipster haven, I have a list: Beirut! Animal Collective! Sunset Rubdown! I head to Sonic Boom on 15th, and find Ruben Mendez behind the counter, listening to country music and ringing up customers, all well-groomed men in their 30s buying Modest Mouse. I’d been e-mailing Mendez for weeks to get his opinion of iLike. He sneezes into his sleeve and explains that he’s still recovering from South by Southwest.
“I think that might be the last time I go. It was just too much—too many bands, too much drinking,” he says, shaking his head. What have I been listening to? he asks. I tell him, and notice how much more aware I am of what I’ve been listening to, and how it’s related to what I own. More important, I know what I want. But I don’t tell him. I ask him what’s good. He leans over the counter and folds his hands.
“You know what’s really great that just came out yesterday? The new Panda Bear—it’s from that guy from Animal Collective.” Animal Collective! How did he know? But no Web site mentioned Panda Bear.
I listen to the song that Mendez recommends and spend the next 12 minutes in bliss. I ask about Sunset Rubdown and Beirut, and am directed to their new releases. Mendez and I have highly compatible taste in music, I decide. He could have told me about Panda Bear on iLike, but so many of the Ruben Mendezes of the world are not on iLike.
“I’m sorry I never got a chance to go to that Web site,” he says, cashing me out. “I’ve just been too busy.”
