HOWEVER, since selections and recommendations are algorithmically generated, and clearly drive towards sponsored artists (and sometimes what appear to be algorithmic black holes: looking at you Tycho), the variety of the recommendations I get tends to get be pretty damn stale. It takes substantial effort on my part to introduce variety into the system and when allowed to run on its own, it always seems to regress towards a very static set of recommendations.
I think the analog variant of this, a record store resolves this by at least forcing you to walk past a bunch of music you might not have other wise listened to. Same thing is true with the book store. You might not be into the classics, but you'll walk past them everytime to get to sci-fi. Eventually you might be like, eh, sure, lets try out some Dostoevsky.
I think a possible resolution to this is to stop treating recommendation algorithms as a purely a product recommendation system or as some black box secret sauce. Give people the ability to work with and manually adjust the algorithms that are built to offer them products/ make recommendations. Allow users to work in partnership with the algorithms to get a more varied (or less varied) range of recommendations.