First of all it's not being shoved in your ear. You can choose to listen and they offer curated playlists etc. as alternative discovery mechanisms. I use Netflix everyday - and usually end up watching repeats of TV episodes because I have no idea what to watch. I use Spotify all day and listen to stuff I know because their radio and playlists are shit. Beasts was the one app I found that actually offered good suggestions or albums and playlists and I spent most of my time on their reconditions screen. Algorithms don't work for this stuff. Maybe they will and we can keep working on it but for now good human curation + algorithms is the key.

Music discovery is becoming HARD because the entire A&R scouting system and top 40 radio model is falling apart. I do not want to listen to 200 garage bands to see if any of them is the next Nirvana; I'm quite happy if Apple is going to pay Zane Lowe to do that for me.

Exactly. Great point.

Except that those recommendations are really poor. And it's not clear to me at all that there's any way to scale the incoming dataset large enough to get useful results from it. Netflix might have good algorithms but they are handicapped by the shape of their catalog. Music has a much, much larger selection space and currently nobody has a sufficiently deep usage history to make "if you listened to x, you'll likely listen to y" work.

Anywhere but smack dab in the center of the distribution algorithmic music recommendation engines fail completely (save for Pandora, which has a different small catalog problem.)

I agree, especially about Netflix and Pandora's poor recommendations as a product of their limited catalog. I don't however think that humans are any better in this regard. Humans also tend to recommend near the center of a distribution of tastes, and when they stray from that, it's out of a deliberate move to find something new/different. Algorithms can do the same (and many do), but many of those suffer from other problems - such as poor interfaces or limited catalogs as mentioned.

Just because no one has done it really well yet, that doesn't mean we should all go back to human-curated playlists in my opinion.

Edit to add: Perhaps part of my reluctance to rely on DJ's and human-curated playlists is a lack of trust. When a person recommends something to the masses, my gut instinct is to question motives (who is being promoted and why). When an algorithm recommends something to me as part of an unlimited service, I tend to feel like it's been tailored to me for no reason other than to maximize my enjoyment (though admittedly, the truth is that algorithms may be built with just as much - if not more - questionable bias as any human).

I get my music recommendations from music blogs or my local record store, and I always always always find them more relevant than anything an algorithm poops out. Finding a way to scale that up would be interesting, but I also don't really think that there's a lot of money in it. Google could do this, really, or Facebook, I suppose, because they have the huge amount of activity data that could be used to take the place of "Hey, I like this Galactic Cannibal record, what else should I be listening to?"