Thanks to danah I've been checking out Pandora, a music recommendation engine that can be loosely compared to something like last.fm, but is ultimately doing something fairly different. Pandora emerged out of the Music Genome Project, which brought a number of musicians together to actually analyze the musical qualities of songs and build a ginormous database based on this analysis. To use the service, you input a favorite artist or artists and create a "station" based on qualities of those artists that will be used to select new music for you. The engine continues to learn over time as you tell it "yes, I like this" or "why is this crap entering my ears" and so on.

This is a very different approach from listening to a tag station on last.fm. This would be the "authority-driven" flip side to the emergent, bottom-up tag-based solution. My initial impressions are that it does a fairly good job of finding something I want to listen to — even if it sometimes makes me scratch my head over what exactly relates this track to the original input. Then again, I think that's probably a good thing — I want to be stretched beyond those conventional associations where artists tend to get lumped together. It's definitely a useful service, and I'm keen to see where they take it, especially as they add more social features to the service (station sharing and a rating system are a couple of the items on the docket so far).

Tech Crunch reports the service will go live officially next week after an invitation-only beta. If you've been beta testing, your free trial will extend until September 28, after which they're planning on charging a $36 yearly fee for the service. Would I subscribe? Probably. Then again, I'm the person who thinks nothing of subscribing to eMusic.com, RealRhapsody, Napster to Go and Yahoo! Music Unlimited… simultaneously.