Your Musical Tastes, Automated
by Thomas Brett
Virginia Heffernan has an article on the popular Internet radio service Pandora in today’s New York Times. Pandora allows users to set up personalized radio stations that generate playlists based on users’ musical preferences. All it takes is an initial song, and Pandora analyzes* (see footnote below) the music based on quantifiable qualities such as instrumentation, timbre, melodic contours and harmonic structure, and even the presence or absence of singing and lyrics. As users agree or disagree with Pandora’s recommendations, the playlist refines its focus, seemingly becoming ever more acute. What we have here is a quasi-intelligent technology that, as Heffernan observes, “refuses to group songs on the basis of their being good, bad, cool or otherwise enshrouded in cultural auras. Pandora explodes the aura. It turns music into math.”
What is interesting about Pandora is that it kind of does a good job at predicting the kinds of musics we might like based on just a few examples that we already like. (You can try it for yourself…) This idea of automating human taste can be unnerving, since we like to think that our aesthetic interests–including our musical tastes–are mysteriously organic, and that they allow us to “feel free and let us revel in our subjectivity” as Heffernan puts it.
To put the Pandora technology in perspective, Heffernan makes an analogy to the great chess master Garry Kasparov’s 1997 loss to Deep Blue, an IBM computer. The point here is that Kasparov and Deep Blue weren’t really playing the same game; or rather, chess means different things to a human and to a computer: an intuitive sense of a game versus abstract number crunching. Heffernan goes on to note there “is a spiritual exhaustion that descends when what is traditionally an experience with another mind (a musician, a chess player, a conductor, a D.J.) turns out to be an encounter with a machine.”
Does Pandora really make our listening tastes feel predictable though? Perhaps it merely sheds light on the kinds of musics that speak to us. At any rate, the next step we need to take ourselves is ask why: Why this kind of music and not that? Why these kinds of sounds, and not those?
You can read the article here.
*Actually, this is a little misleading. Pandora hires real people to analyze hundreds of thousands of songs, one by one, in terms of their quantifiable musical qualities. Once analyzed, each piece of music is “tagged” with its relevant data points and then added to the Pandora music database–referred to by Pandora founder Tim Westergren as “The Music Genome Project.” For a more in-depth exploration of Pandora, see Rob Walker’s article, “The Song Decoders” here.