I remember how after first discovering the early incarnation of Napster as a college student in the late 90s, I pretty quickly came to realize how limited my musical taste was: Plopped in front of an infinite jukebox, it had taken surprisingly little time to track down pretty much everything I was interested in but didn’t have already. So, like a lot of others, I decided to branch out by taking advantage of Amazon‘s social filtering software to figure out what else was favored by people whose taste was similar to mine by looking at the albums purchased by people who’d also bought albums I liked. It worked pretty well—some of the bands I first discovered that way remain among my favorites. And then LastFM came along and automated the process, using social filtering to help create customized online radio stations automatically. ( iTunes’ “ministore” recommendations use both social filtering and simple automatic queries like “does this artist have anything forthcoming?”)

Now, a tip from amigo Matt Klokel of Fantom Comics points me to Pandora, which also allows you to create customized radio stations, but using a very different principle: A Music Genome Project analyzed thousands of songs and artists using a classification system making use of hundreds of musical “genes”. It then takes one or more artists or tracks you give it and puts together stations that play genetically similar music. Feed in Eminem and Ella Fitzgerald, out comes Nellie McKay, presumably.

Now, these guys are doing different things: One uses community preferences to figure out what music is connected by clusters of taste, another to find music that sounds similar. Those can be different for all sorts of reasons—because scene boundaries keep separated music that, objectively speaking, might sound relatively similar, or because of other characteristics having to do with history or people rather than the sound of a piece in itself. Mike Doughty’s solo guitar stuff doesn’t sound a whole lot like Soul Coughing (except, of course, for Doughty’s voice) but a social filter will link them because plenty of old S.C. fans stuck with Doughty after the band’s breakup. Or, more generally: Social filtering will tell you that people who like X and Y tend to like Z, but not why.

So I wonder which will be more apt to do a better job of introducing you to new and interesting music you hadn’t heard. We’re all supposed to be singing the praises of decentralized, dynamic stuff like social filtering over expert classification. But I wonder whether it mightn’t be Pandora that pans out better—partly for some of the reasons adumbrated above having to do with quirks of social filtering (though those are a feature, not a bug, if you mostly just want to sound clued in when mixing with scenesters) and partly because so much of our cultural consumption is already socially filtered, the marginal added benefit of having a computer do it drops off relatively soon.