Robot Recommendations

Your computer wants to know what you’re listening to

Volume CIX, No. 1January, 2009

Mikael Elsila

How would you like your computer to recommend music for you? That’s the latest trend, but is it good for consumers or musicians?

Here are three quick examples of how computers are thinking for you. Warning: these will only make sense if you are reasonably tech-savvy.

1. In iTunes 8.0, If I select a song — for instance, “Dear Prudence” by the Beatles — iTunes can create a playlist for me of other music in my own collection that should groove with it. In this example, I get some Simon & Garfunkel, Joni Mitchell and the Rolling Stones to go along with the Beatles. No surprises there. But iTunes also suggests that Queen, David Bowie, Deee-Lite, Natalie Merchant, the Cardigans and the Moldy Peaches should work, too. Hmmm. 

2. The online Web radio site also has a similar form of artificial intelligence. If you go to the site and tell it one piece of music that you like, it will recommend others to you. Pandora calls this the “Music Genome Project.” The site claims that Pandora staff are listening to thousands of musical pieces and classifying each of them according to different attributes that they call “genes.” Examples of “genes” might be: the style of the music (classical, jazz, blues, etc.), instrumentation, timbre, and even something as esoteric and detailed as how much distortion there is on the electric guitar part (if there is one). Since every piece of music has different “genes” and since these genes can be compared, the site is able to figure out how different pieces of music may be related and compatible to listeners. The site also lets listeners give feedback, which further refines the process.

3. Finally, most Web users know that the online store will recommend books and music to you based on past purchases. This too is a form of artificial intelligence.

So now, the question: is any of this good for musicians or listeners? I say yes. For musicians, it gives added buzz, P.R. and recognition. For listeners, it’s a treat because it opens up new musical doors.

However, a computer is just a computer. A computer doesn’t really know you the way a friend does, when a friend recommends music for you.

Also, there are gaps. There might be some great music lurking around the corner that you would never know about if you just trusted the computer for everything.

Finally, just because you tell the computer you like the Beatles doesn’t mean you’ll like the Moldy Peaches. There is more to music than just “genes” or attributes. How we first hear a song creates emotional imprinting. Whenever I hear Paul Simon’s “Graceland” album, I’ll remember the cross-country drive with my family when I was 15, and therefore the album means much more to me than the music itself. Computers can’t get that deep into our heads and figure that out. Not yet.