Form and Analysis

In high school, I studied piano and music theory. I remember building four-part harmonies and counterpoint, learning the rules of chord progression, the scales, even a bit of twelve-tone theory for spice.

When  I look at a site like Pandora, I'm a little awed. The Music Genome Project has created this site and is populating it with music profiles, "DNA"s of music. Their people sit around categorizing music not by traditional "form and analysis," but other, less academic methods. These are invisible, transparent to the end user, but you get a "profile" and they match music to your tastes.

I think the old traditional way of analyzing music formed a method, a vocabulary whereby those educated in it could find the music they liked, or, in the language they used a hundred years ago, the music that was "eternal" or "great" or "timeless." These new systems democratize, reach not for common denominators of greatness, but for an infinitely versatile sliding scale that codifies and uses your own tastes in the service of exposing...and marketing, of course.

I'm of mixed minds about it. After all, they don't teach you how to use or apply the scale themselves, they don't educate. You click a button and they provide "music you might like" based on their algorithms, their black-box analysis of the music in your own iTunes library or on your hard drive. They'll only suggest music in their own catalogs, of course, which is a marketeer's delight, and that also worries me. This tool can be used for evil--coopting the Internet's astounding heterodoxy with normalizing traits: it doesn't matter what you like, SOMETHING mainstream in our catalog will match your "profile" close enough, you should listen to this, then buy it from us. This is already happening on Pandora--you can only skip (read "browse") a few titles, then you're stuck listening to what they gave you. Hmm.

But the Internet routes around even these sorts of things, because you can't put a lid on the idea. Other sites with other algorithms will pop up, ones which match in different ways. I expect there'll eventually be "meta-recommendation" sites that rate and explore the idiosyncrasies of the profiling sites, or that places like Digg.com will serve as watchdogs.

In all, though, it's part of how market forces collide with Internet possibilities to generate low-level AIs, and that fascinates me to no end. What does it mean that a computer, through non-aesthetic analysis (after all, a computer can't LIKE a piece of music) nevertheless uses trend-analyzing algorithms to refine what I like, whether I know I'll like it or not?

If it doesn't work, of course, I can always fall back on the old-fashioned method. Because, see, I love choral music in minor keys that ends with plagel cadences.

--Peter

Comments

[Peter Wells]

Thanks Aubrey!

We'll try to post more now. :)

--Peter

[Aubrey]

Fantastic blog! Fantastic site!