Media Theory for the 21st Century

February 20, 2008

Database and Recommendation

Filed under: Discussion,General — mkontopoulos @ 8:40 am
Tags: , ,

This doesn’t have much to do with aesthetics, but last night, I started playing around with Pandora, which is an website used for creating custom radio stations based on personal taste. Since I’ve been really into Curtis Mayfield lately, I started by typing in his name. Pandora pulled up a song by Curtis, and then proceeded to select songs that were similar to the first, based on a series of several hundred predetermined criteria. By rating several songs in a row, Pandora was able to fairly quickly and accurately, configure a radio station based on my impulsive desire to listen to soul music.
If you read a bit deeper into the Pandora website, you’ll find that their archive is based on what is called the “Music Genome Project”. Launched in 2000 by Will Glaser, Jon Kraft, and Tim Westergren, the intent was to “capture the essence of music at the fundamental level”. Ultimately, these three music aficionados developed over 400 attributes with witch to algorithmically analyze a database of songs. I am aware that the algorithmic prediction of taste is not new, as seen by Amazon’s or Netflix’s recommendation resources. What is striking about Pandora, however, is how it bases its selection criteria not only on user encouragement, but also largely on analysis of the song itself for things like “bebop qualities”,  “heavy use of funk samples”, or something as specific as “Subtle use of Harmonica”. I find it’s success in pleasing my musical needs to be equal parts advantageous and creepy.
The obvious connection point here is to the Human Genome Project. In considering human DNA as a database, as Vesna points to early in her text, we open up a whole new way of thinking about our bodies, especially in learning that we are 99.9% genetically similar. I’m similarly drawn to the Music Genome Project, for the way it makes a database out of a form of expression as ancient and primal as music, breaking down the differences between styles into such a (relatively) small amount of criteria; not to mention the questions this raises about disembodiment and removal of subjectivity in the age of algorithms predicting our  desires…
Christian Paul writes: “What distinguishes digital database from their analog predecessors is their inherent possibility for the retrieval and filtering of data in multiple ways.” (96). With, what Zach so eloquently described as “utopian flare”, Paul seems to praise the digital databases “limitless possibilities”. To parrot Zach’s discussion of exclusion, I’d like to suggest that the exclusion of certain songs in Pandora versus the positive rating of others does, in fact, construct a narrative of personal taste, based on an enormous range of choices that is algorithmically limited over time, per station. Here, I identify Pandora very strongly with Lindsay’s discussion of Machinima and the range of gaming options as being a sort of database, while individual play represents a type of narrative.
What is excluded from many of these texts, however, is consideration of issues raised by narrative/database and the politics of algorithmic recommendation/prediction. I find it disturbing, the extent to which software now attempt to finish my thoughts and predict my desires. In some cases, this can be useful, like with Pandora, but in most others it takes on a more political agenda, like Gmail attempting to push sponsored links based on text in my personal email, or Amazon attempting to sell me products based on my casual searches. To what extent do we forfeit agency when we leave more control to the predictions of algorithms?


Blog at