As algorithms — such as those Facebook and Google use to curate and sort content — come under scrutiny for their hidden, immense power, they’ve become stand-ins for all the evils of the tech industry. But there are good algorithms still out there, working their magic in mysterious ways. Take, for instance, whatever the hell they’re doing at Spotify.
This week, Spotify introduced a new algorithmically generated playlist called “Time Capsule,” which the company describes as “a personalized playlist with songs to take you back in time to your teenage years.” Mine is accurate, in the sense that it contains a bunch of songs I like, and a lot of bands that are not necessarily popular now but were about 15 years ago. The first three tracks are from Dashboard Confessional, Jimmy Eat World and Simple Plan — there’s a decent amount of pop-punk. The playlist ends with Wheatus’s “Teenage Dirtbag,” which seems a little too on-the-nose to have been procedurally generated.
I’m not the only one who’s been impressed by my Time Capsule playlist — Twitter is full of wide-eyed people shocked at how accurate Spotify was at describing their teenage music choices. The vibe is very “David Blaine just levitated in front of us” — the playlist acting as a fun and shocking magic trick.
And like a magic trick, the seams show once you look at it closely. Many of the tracks are ones I’ve listened to on Spotify multiple times. I never listened to Taking Back Sunday in high school, but I assume the band appears because My Chemical Romance does. “The Way We Get By” and “I Turn My Camera On” are probably the two most popular Spoon songs in the band’s discography and those are the two that made the cut. There is a tangible logic behind most of the entries — some balance along the axes of personal affinity and overall popularity.
Overall, the playlist is very good not because Spotify is good at recommending new things, but because it’s decent at imitating past behavior. It’s like the algorithm used my account, which is connected to Facebook and thus makes my date of birth known to the service, to calculate my teenage years, and then just picked songs that I’d listened to that came out in that time frame. Then it used some fuzzy matching to throw some oddballs into the mix, so that it doesn’t just seem like a rehash of my listening habits. (This assumed strategy appears to have worked. “Are these songs I listened to?” someone asked me while scanning their own list.)
Whether that’s actually how it works, I cannot say. Matching algorithms are usually kept under lock and key, considered a trade secret. But the seams showing on Spotify’s recommendation process has a sort of whimsy to it. It produces the exact right reaction that an algorithm should produce — a sense of a computerized service knowing and understanding its user, without being eerie or uncanny about it. Its mechanisms appear transparent: There’s nothing in my Time Capsule playlist that I can’t explain. And like any drill-down of cultural preference, there is a mix of the embarrassing and the braggy that makes it almost impossible to not share and compare with others.
Maybe most important, it’s just for me, and it’s just a list of songs I like. Compare that with matching algorithms like Facebook’s People You May Know, which has the capacity to recommend secret family members with no mutual connections to a user, or shopping algorithms that know you’re pregnant based on the items that you’re buying or browsing, or algorithms that can analyze a user’s interests to assume a user’s sexual orientation. The trouble with algorithms, Spotify shows, isn’t the algorithms. It’s companies that wrap them around our lives in ways we don’t understand and can’t control. That’s the scary encroaching future. Algorithms that do nothing more complicated than serve up pop-punk songs are exactly what a tech utopia promises.
