This is what makes Pandora unique from Last.fm, or Ping, or other "social" music discovery sites. Instead of recommending songs according who likes them, it learns what kind of music you like. It looks across genres and artists to the actual musical features of each song.
One thing that I really like about the service is finding artists that seem to lead to lots of other great songs. It isn't always who I think it is. For example on a lark I put in "New Radicals" because I liked one of their songs and I was so impressed with how well that defined a particular sound. When I put in big names like the Beatles or Rolling Stones I don't think I left those stations defined for very long. I wonder is there is some kind of structure to the what Pandora has calculated that could show which bands or songs seem to be most influential.
P.S. My friends are always complementing me on how awesome my music is. I try not to take too much credit for it.
I want to see the future of radio be one of full choice with great suggestions, not pandora's restrictions. I think Pandora has built amazing suggestions, but it's not perfect and I want to push for the next generation of music experience.
FWIW, you've caused one rabid Pandora fan to give it a look.
It is my understanding that they have hired musicians to listen to each of the songs they stream then rate each one according to dozens of factors. The factors include things like "acoustic guitar", "strong melody", "high studio production values", "gritty vocals", etc etc. Then they are able to do some kind of cluster analysis on that dataset. So that if you put in "Stone Roses" you hear music from that band but then others that have a similar sound.
I find about 2 or 3 new bands that I really like every week. I have actually started buying CDs again.