At Medium's "OneZero" blog, Dave Gershgorn writes about how the combination of a powerful algorithm and extensive data sets means the Spotify streaming music site "knows exactly what you want to listen to," according to the headline.
Except, of course, it doesn't. At least the different times I've tested the waters with it, anyway. If you get a hankering for an outlier number among your fave bands or artists, then it's not long before the algorithm runs out of things that resemble what you picked and it flails around through things that might be in the vein of the tune you've chosen, if you can somehow manage to squint your ears. It may include songs where your entered performer worked with someone else, and start to veer away from the entry you made because it finds more of the other performer. And it doesn't know that some of those duets or band appearances are really not what you want to listen to: I enjoy a lot of Keely Smith's music but I've got little patience for the vocal mugging and goofoffery of her onetime husband Louis Prima so I don't really enjoy listening to their joint work. Spotify don't know that and it takes it awhile to learn unless I spell it out in some way.
Sometimes it leaves weird gaps. Playlists that include the mid-80s cowpunk outfit Lone Justice ought to lead a listener to singer Maria McKee's solo work, since she has performed that way far longer than she ever did with the band. But it doesn't. Enter either Lone Justice or McKee into the search engine and get the one you enter, but no recognition the other has any work at all. Not to mention the omission of Lone Justice's second album entirely. Sure, they show up as other entries under the "related artists" button, but that's not the algorithm finding them -- that's me.
And of course there are the bands which aren't, for whatever reason, on Spotify. Old musicians, niche acts, favorite small-scale performers who don't rise to the service's notice.
I don't really have anything against Spotify or other streamers other than how they pay artists the next best thing to nothin' for their work. But the idea that its algorithm -- or any algorithm -- knows "exactly" what I want to listen to is silly. After all, I don't always know, so how will a program?
2 comments:
I'm back on the Spotify for a bit, but I've found that when I ask it to play music that sounds like a band or artist I like, I get musical stylings from four or five other acts. And if I choose another band or artist I like, I tend to get the same four or five other bands.
Letting autoplay run on YouTube is marginally better.
When I lived in Chicago, I was a devoted WXRT listener because of the mix of established and experimental (they've changed these days, I understand). I called it, "the kind of music that you wanted to hear but didn't know you wanted to hear it until they played it." These days any kind of fully random playlist just shovels out endless amounts of ephemera speckled with a nugget you might want to pull out and listen to a couple more times, and putting restrictions on the playlist eventually exposes the catalog limitations.
Post a Comment