And the Band Played On
The Algorithmic Push Toward the Past
Pavement, a band who haven’t recorded any music since 1999, released a new video this week – to a song from 1997. In recent years, the song has become a streaming hit for them.
Everyone will tell you that streaming is booming – and it is. But the music data company MRC (formerly known as SoundScan) dropped a heavy qualification in their 2021 summary report, released in January:
“For the first time since MRC Data began tracking streaming data, streaming of new music has declined in volume year-over-year. Which means Catalog has gained a significant share of total listening in 2021, increasing nearly 5 points from 2020 to 70% of total album consumption.”
Did you catch that? Streaming is indeed going up, just like you thought. But the streaming of new music is going down. So far down that in 2021 only three of every ten streamed songs were new releases.
If streaming is replacing radio, it’s oldies stations up and down the dial.
Why would this be? MRC themselves posit a theory in their report, but it doesn’t really seem based on data: “This is an acceleration of a trend that picked up steam during the first waves of COVID-19 lockdowns, as music fans turned to old favorites for nostalgia listening.” Well… maybe? I don’t see how MRC could know why people decide to listen to one song over another. Are we even so sure ourselves?
But leave the why out of it, and you can still ask how so many came to listen to older tracks. That’s easy, given the way most choices are made for streaming these days: algorithms.
Algorithms don’t feel COVID anxiety. Algorithms have no nostalgia. Yet streaming algorithms would seem to have a distinct preference for older music.
I have a hunch how that works, based on watching my own tracks interactions with Spotify’s algorithm.
In 2018, I noticed that one Galaxie 500 song – “Strange” – was streaming on Spotify far more than any others. There was no obvious explanation. The song hadn’t been included on a popular playlist; it hadn’t been used in a film or commercial; it hadn’t been covered by another band, or mentioned by some celebrity influencer. As far as we could tell, there hadn’t been any change in its public profile for literally decades. What’s more, it hadn’t been one of our singles, or “emphasis tracks” for radio back in the day. There is no music video. It simply wasn’t the logical track to try first, if you wanted to check out what Galaxie 500 sounded like.
And yet, the numbers were clear. “Strange” was far and away our most “popular” track on Spotify – streamed twice as much as any of our other songs, and ten times more than the tracks that surround it on the album where it originally appeared, On Fire.
I mentioned this on my blog at the time (Tumblr! talk about nostalgia…). And surprisingly, Spotify’s “data alchemist,” Glenn McDonald, noticed. Glenn took it upon himself to look into the odd behavior of “Strange,” and called me up.
As I reported in a second blog post at the time, Glenn pinpointed the rise of “Strange” to a particular day in January 2017 – the same day Spotify initiated a change in their app preferences, making “autoplay” a default rather than a selection by users. Autoplay means that after hearing whatever you have chosen to stream, Spotify doesn’t turn itself off but keeps playing music – letting its algorithm select what you hear next.
The algorithm chose “Strange.”
To be clear, all Galaxie 500 tracks are “catalogue” – the band broke up in 1991, before many Spotify listeners were born. So any of our tracks that the algorithm latched on to would contribute to the 70% oldies streaming on the platform. Just like Pavement’s “Harness Your Hopes.”
Nevertheless, I think it could be the same bias that plucked “Strange” out of our catalogue, and causes streaming algorithms to select catalogue music in general over new. That bias is for songs that sound… like other songs.
When I spoke to Glenn in 2018, he explained that with autoplay switched on, the algorithm aims to select music that matches in some manner the music that just finished playing. How it makes that match is a trade secret. But the simple goal is a resemblance – a familiarity - to whatever the user had initially chosen to hear.
“Strange,” evidently, fit that role better than any other Galaxie 500 song. Or you might say: “Strange” resembles songs by other bands more than any other Galaxie 500 song. Twice as much as any of our other tracks. Ten times more than some. The algorithm would seem to have identified “Strange” as our least peculiar song – the one most likely to sound like whatever else you had played.
Now, zoom out to all possible musics streaming for all possible listeners. What music will most likely resemble what they just chose to hear? I suspect the answer is, way more often than not: something old.
Older music is, by definition, like other music. How could it not be? In fact, what we were just listening to sounds a bit like it. And so we let the music stream on… and on… and on… so long as the resemblance to what we already know continues.
This, I’m afraid, is the great algorithmic push toward the familiar. The echo chamber of resemblance. The online bias against difference.
Which, it seems, also contains a disturbing bias for the past. Look away from music and toward information platforms driven by similar algorithms, like Facebook, and that same bias could be casually amplifying a “lost cause” like the Confederacy, or the “former glory” celebrated by Fascistic nationalists (“make America/Russia great again”).
Because what is the new, if not the unfamiliar. Or someone unfamiliar.
Listening to: Stromae’s Multitude
Cooking: a recipe I’ve never tried