Computers Deciding what Music We Like

http://blog.wired.com/music/2008/10/mufin.html

The latest music recommendation engine, Mufin, will tell us what music we like using a serious of relational algorithms. Mufin developed by "the father of mp3s" Karlheinz Brandenburg doesn't pay attention to the artists you like. It decides purely from computations made off of over 40 aspects of music, like percussion, style, tempo, sound color, instruments, volume, dynamics and loudness. This often produces bizarre results. Most music recommendations will give you a band that is within the same time period and genre. Mufin's recommendations disregard all of this. Other music recommending services have much different ways of going about recommendations: Pandora radio consults music experts, and Apple's genius and CBS's Last.fm uses software that notices when artists and albums are in the same collection as each other. Mufin will completely disregard all human conception of musical association of whats hip and what came out in a certain time period. It focuses on attributes which can be assigned numerical values.

The author of the article tried the service and found strange results:
"The site told us that the Beatles' "Sgt. Pepper's Lonely Hearts Club Band" sounded like "Dirty Love" by Tim Feehan, Edo Zanki's "Wie Ein Feuer" and Rafael, Joe und Die Partysingers' "Freudenspender."
The author did say that the suggested song shared characteristics such as tempo, vocal range, and chord pace, but the songs often had different emotional messages that the computer could not pick up on. All though he does admit that only Mufin can pick up on certain things:
""It's really difficult to put together similar artists for David Bowie, because he has such a great diversity of music that he has been doing," explained Djekic. "This is why Mufin is purely song-based. You can have a song that is similar to another, but an album that is similar to another? Or an artist that is similar? It's really kind of difficult." By ignoring that data, Djekic said, Mufin's recommendations are more accurate."
In this way Mufin offers its user something new. The ability to find a song that is similar regardless of genre and time period. It presents the same abstract mood.