Pandora, Genius, and the Guillermo Klein problem

The Internet strengthens niches—a political cliché that's especially true when it comes to music. For most of the 20th century, radio forced listeners into a grudging democracy of taste—we would hear songs we knew we liked, songs we didn’t like, and new songs that intrigued us. Sometimes we'd discover great new music. Most of the time we complained that the DJ sucked.

That's over now. First the iPod gave us the ability to hoard hundred-hour playlists. Then the Internet—from the Wild West of Napster to Myspace and streaming radio—endowed listeners with the power to shape the world to their tastes. Death to the autocratic DJ! Viva you! We finally had the freedom to listen exclusively to the music we wanted to hear.

Yet even the most unadventurous listener finally tires of his lifelong favorites. With radio effectively gone, where's a music consumer to turn for something novel? Ladies and gentlemen, I present Pandora and iTunes’ Genius. Pandora is a personalizable internet radio service with a library of 750,000 songs; Genius is a “smart” playlist function that is limited to a user’s iTunes library. Both balance enabling discovery with nurturing personal taste: you choose one song, and Pandora and Genius will curate hours of programming that you're guaranteed to love.

A techno-utopian might imagine a mind-expanding journey of discovery—the familiar boldly ventures into the unknown and suddenly the divisions collapse between genres, eras, and continents. A techno-skeptic would realize that Pandora and Genius could became hallways lined with facing mirrors. A man gazes at his infinitely replacted image convinced he's seeing the diversity of the world.

Last week, WBGO’s Josh Jackson (who hosts the radio show The Checkout) interviewed Sameer Gupta, a percussionist and Pandora “song decoder,” about how Pandora’s Music Genome Project creates personalized stations. Gupta described analyzing a song based on its structure, instrumentation, and subjective qualities (the emotional intensity of a solo, for instance) and then feeding that information into a radio-station-creating algorithm. Jackson, a genial and rarely confrontational interviewer, allowed Gupta to explain the process before voicing his doubts.

“Here’s where you lose the scope,” Jackson said, “let’s take that premise that I love Bill Evans and I click ‘I like Bill Evans’ and Herbie Hancock pops up—very similar in nature in terms of instrumentation and the way that they’re improvising to some degree. So, okay, I love Herbie Hancock too, [it doesn’t give] me another Herbie Hancock which could potentially be like Headhunters. I don’t get to explore the range of Herbie Hancock so much as get to explore the range of the acoustic piano.”

Pandora, Jackson argues, can’t make the surprising choices, inspired leaps, and productive dissonances that are the hallmark of good human DJs and bold human artists. Listening to a random sampling of the work of a curious, prolific musician like Hancock would introduce listeners to smoldering 60s jazz like Maiden Voyage, fusion like Headhunters, electronic hip-hop like Future Shock, and neo-traditionalism like Directions in Music. Listening to a Pandora station based on Herbie Hancock’s "Dolphin Dance" would keep a listener in the realm of "instrumental post-bop jazz with long piano solos."

Pandora claims that it builds its stations around musical properties alone—ignoring the cultural baggage that often dictates taste—but the evidence suggests otherwise. As Rob Walker noted in the New York Times Magazine, “the [Pandora] genome, quietly, doesn’t really screen out sociocultural information. For instance, its algorithms are tweaked by genre, and the inclusion of genes for ‘influence’ (‘swing’ or ‘gospel,’ for example) brings in factors that aren’t strictly about sound.” In other words, Pandora’s cultural assumptions about music can inform its programming as much as, say, the repeated use of minor-7 chords.

I first became aware of Pandora's penchant for emphasizing extra-musical qualities while listening to a station I’d created around the pianist and composer Guillermo Klein. Klein, an Argentine who has lived for the last two decades in New York and Barcelona (and whom I interviewed last year for All About Jazz), plays music that’s tough to pigeonhole. It's ostensibly large-ensemble jazz, but it has several elements that set it apart from the tradition: composition trumps improvisation (some pieces have no improvised sections at all); South American rhythms drive the music instead of swing; and Klein sings with a rough, untrained voice that's more Bob Dylan than Johnny Hartman. Given the genre-defying nature of his music, Guillermo Klein's Pandora station should include modern small-group jazz (The Bad Plus), untraditional big bands (The Pedro Giraudo Jazz Orchestra), historical influences (Olivier Messiaen, Duke Ellington, Astor Piazzolla), and even bands from different genres that share Klein's sense of harmony and rhythm.

Instead of exploring Klein's unique synthesis of jazz and non-jazz idioms, Pandora creats a Guillermo Klein station based on one non-musical fact: Klein was born in Argentina. Both my Guillermo Klein station and my station based on “Yeso”—a song off Klein’s most recent album Filtros—play almost exclusively Latin American music. When my “Yeso” station played Monna Bell’s “Culpable”—a poppy swoon that sounds almost nothing like “Yeso”—Pandora said it selected the song because it “features female lead vocals, minor key tonality, acoustic instrumentation, bowed strings, and a string ensemble.” Not only is that justification a little baffling—“Yeso” features neither female lead vocals nor a string ensemble—it obscures the most obvious reason the song was picked: Klein, like Monna Bell, has been tagged as “Latin." Pandora’s sound decoders seem to have privileged heritage over artistic qualities, helping to create a station in which Klein’s brooding compositions incongruously segue into the poppy Bell and the raucous Cuban drummer Horacio “El Negro” Hernandez.

To compare the way in which Pandora and Genius select music, I chose the same song, “Yeso,” as the starting point for a new iTunes playlist. Comparing the two programs in this way is wildly unscientific—Pandora’s library is 750,000 songs, Genius’ library consists of the 7,500 songs on my computer—but even with that significant caveat, Genius distinguishes itself as the superior analyzer. My Genius “Yeso” playlist includes Rudresh Mahanthappa’s "Introspection," John Scofield’s "Down D," and Charles Mingus’ “Prayer for Passive Resistance"—unexpected choices that suddenly sounded surprisingly Kleinian. Genius doesn’t recognize most of the Argentine jazz on my computer, but it's nonetheless significant that on my 25-track “Yeso” playlist, Guillermo Klein is the only Latin American musician.

Genius is not without limitations. All its “Yeso” selections are clearly identified as jazz, suggesting a rigidly genre-based approach. Genius also can’t select anything that I don’t already own, which severely limits how much music it can really help me discover. But as a tool for both introducing music and revealing connections between artists, Genius seems to have a far more fruitful approach than its Internet radio cousin.

At the risk of sepia-toned sentimentality, I'll still take a good radio disk jockey, thank you. When my friend Gregory Kress DJ’ed jazz shows on the University of Michigan’s WCBN station, he’d frequently make the kind programming leaps of which both Pandora and Genius seem incapable. At WCBN, free-form programming was encouraged, and free-form DJs often brought jazz into shows that were dominated by college rock. When WCBN listeners tuned into a good free-form show, they’d hear their Arcade Fire and Clap Your Hands Say Yeah, but they might also be forced to sit through some Albert Ayler.

The benefits of forced exposure go beyond radio. Darcy James Argue recently lamented the absence of opening acts in jazz, another part of our larger cultural trend toward offering listeners only what they already know they like.

When you go see a rock show—any rock show, doesn't matter whether it's a CMJ showcase, or a bloated stadium tour, or just your average night out—you're going to hear bands you probably haven't heard of before you get to hear the headliner. Often those bands will surprise you. Why doesn't this happen in our own little corner? We've talked before about the benefits of cross-pollinating across genres, but even within the jazz world, why don't we see Gretchen Parlato on the road opening for Esperanza Spalding, or Marcus Strickland out with Dave Douglas, or Andrew D'Angelo out with Tim Berne? Instead, this only seems to happen within the headliner's band (e.g., when Marcus plays in Dave's groups). But what about, say, Aaron Parks's own quartet opening for Kurt Rosenwinkel's, or Donny McCaslin's band opening for Maria Schneider's—not just as a one-off, but actually touring both groups together? You'd think the overlap in personnel would make those combinations irresistible.

Some of my greatest experiences as a listener have come when I’ve happened upon an act I’ve never heard before. Hearing the Randy Weston Trio at the Caramoor Jazz Festival in 2001 and the David S. Ware Quartet at the Newport Jazz Festival in 2002 gave me the sensation not only of loving the music but of falling in love with the music—that dizzying head-rush of excitement.

Driving across Maine a few summers ago, I was flipping through radio stations when I heard a few bars of “Señorita” from Béla Fleck and Chick Corea’s duo album The Enchantment. I'd never heard it before. I was intrigued. I stopped scanning. By the height of Corea’s solo—about two-thirds of the way through the song—I shouted out a gleeful “gaaaaaahhhhhhhh!!!!!” to no one but the empty car. When I arrived in Portland, I drove directly to a record store.

Pandora and Genius might technically be capable of giving listeners this joy of discovery, but both (especially Pandora) have decided to play to a narrow niche. That might be good business for now, but it’s ultimately limiting, tiring, and boring. I’m looking forward to the day when someone starts a personalizable music service that can give me a Guillermo Klein radio station with Olivier Messiaen, The Bad Plus, Dirty Projectors, and forty other musicians with connections to Klein I’ve never considered. The Internet's vastness should broaden and deepen our knowledge. We'd be fools to continue avoiding its unknown.

Comments

[Jacob A]

As someone who frequently uses Pandora, I wholeheartedly agree. I have several saved "stations" on Pandora, and although each time I go to the station, the playlist is slightly different, I find that I am listening to the same five or six artists each time. I love Pandora, but I think that in a lot of ways it has made me even lazier when it comes to music selection and has replaced the enjoyable task of exploring for music with the luxury of having it served on a silver platter, free of charge. Yet another compelling argument for the human interaction of radio over the technological aids of massive databases.

[Eric Benson]

While no Internet radio station will ever replace actively exploring music—going to concerts, analyzing recordings, studying artists—I actually think Pandora does a decent job at introducing listeners to new artist within the very narrow scope of its stations. My big lament is that the stations have been programmed to make such conservative choices, and, in cases like Klein's, have been programmed to make choices based on the wrong criteria. Many human DJs are no better than Pandora, but the best showcase a deep intelligence, sensitive ear, and coherent logic behind their selections. With all of the information that the Music Genome Project gathers about each song, I think Pandora could come much closer to this ideal. Maybe Pandora could introduce two station options: "regular," the current limited and often repetitive playlist, and "voyager," which would make bold-yet-calculated musical leaps (basically iTunes' Genius with a little more daring).

[Andrew Durkin]

This is a very interesting piece. I'd be curious to know about the constituency of music listeners that Pandora is attempting to serve (I haven't had a chance to listen to the interview, so I don't know if this was covered there). But I think they are assuming that most people are pretty happy hearing "more of the same." Why else would they tout their service as "stations that play only music you like"? You can't really be adventurous in your listening without the risk of failure, and if they're counting on you liking everything they come up with, well, where's the risk?

That in fact is why I have always been suspicious of services like Pandora, Genius, the Amazon recommendations system, etc. I do use them from time to time, but I kind of resent the way they seem designed to make the process of discovering music easier. That seems like a pitfall to me, because personally I have always enjoyed the hard work involved in the process of discovering new music through my own active research: methodically triangulating numerous sources (friends, libraries, reviews, DJs, etc.) and coming across some real gems in the process. While it's true that that approach is more time-consuming, involves a greater degree of risk, and a can occasionally yield a dud, it still feels like when I am actively participating in the my own music search, instead of having possibilities handed to me by an algorithm, the end result is a feeling of greater connection with the music I end up loving.

Of course, on the other hand, I understand the motivation behind services like Pandora, which I think would not be springing up if there was not exponentially more music in the world these days. It's a response to what Alvin Toffler called "overchoice" -- the dizzying array of new releases in any given week, month, year. So I get it, but I still avoid it.

[Eric Benson]

I really like the idea of Pandora as a response to "overchoice," because it exposes the paradox of multiplicity: as people are presented with more and more options, their selections become less and less adventurous. I completely agree that Pandora is after a musically risk-averse crowd, but as a filter for our overchoice I think it's a failure. A really interesting overchoice selector would be able to hop from genre to genre and make the kind of surprising, intuitive connections I talked about in the post. Pandora falls victim to the same overchoice paradox as most human listeners, sticking more closely to a particular niche when the plethora of music becomes too intimidating. There's certainly no replacement for the hard work of musical discovery, but as someone who often gets lost outside the borders of jazz, I'd love to see an Internet radio service that could show me ways to use the music I love to access the music I don't know.

[il maestro]

Here's one for you - and I say this in total seriousness - I often find that iTunes SHUFFLE fashions some of the most interesting and surprisingly brilliant playlists out there. I really wonder how it goes from song to song, but often I find that it will play 10 songs in a row that have a similar beginning - say starting on a soft, high note - and each one will come from a totally different genre. Sometimes it seems to have a really canny sense of humor, building up to a huge climax in a Mahler symphony and then replacing the promised arrival with some light piece of 1930's saloon music. I swear to god I'm not being facetious here - I often sit back and listen with rapt attention to what insights it will bring next. Just a thought.