Listening to your favorite album was, until recently, something that required some effort. A vinyl record, a cassette tape, a CD — when they stopped playing, you had to get up and physically change whatever was playing. The rise of MP3 players marked the beginning of the digital revolution, and streaming services knocked analog listening into the realm of collectors and baby boomers.
The music available on streaming platforms is effectively endless. Depending on which one you use, there can be up to 45 million tracks to sift through, good for about 2.5MM hours of audio. (That’s 285 years for those keeping track.) So what happens when you get to the end of an album on your favorite streaming service? Well, the music keeps on going. And the songs that follow are the result of a complex set of algorithms that takes what the streaming platform knows about you and your peers and turns it into recommended songs.
How recommendation algorithms work is a bit of an opaque process. There are collaborative filtering models, which take into account how you behave and contrasts it with how other users behave, as well as audio processing models that take the actual sonic data in a song file — think pitch, tempo, bass levels, etc. — and cross-reference them with similar songs to create recommendations that speak to specific tastes. (The first generation of song suggestion systems came from teams of actual human listeners who would assign descriptive tags to songs, which would then be sorted into related playlists.) The major streaming platforms all use a hybrid approach to build a constellation of recommendation algorithms that can often border on the eerie in how well they know your fondness for certain kinds of music. But that’s the magic of the algorithm. Sometimes it knows your taste better than you do.
The issue with automated recommendations, though, is that it takes some of the humanity out of discovery and can influence what we listen to in hazy ways. Algorithms aren’t truly objective, after all. Embedded in the long strings of equations and code are the conscious and unconscious biases of its authors. In the case of music recommendations — and for the purposes of this article, we’re talking specifically about the algorithms employed by major music streaming platforms rather than, say, video recommendations — that can have the odd effect of forcing artists to change their sound to better fit a cloudy set of technical parameters. Recommendation algorithms also have the habit of creating musical echo chambers, because they’re reinforcing tastes rather than broadening them.
“It’s a fool’s errand to try to reverse-engineer something that’s infinitely complex and unknowable,” said Armando Young, a New York-based musician. Young’s music is ephemerally layered, and he uses his background as a musical polyglot to create soundscapes that are as beautiful as they are difficult to categorize. His inability to fit into a box means that when you ask for recommendations of related songs you end up with a small coterie of artists Young has worked or recorded with, a likely byproduct of natural language processing (NLP) driven recommendations. (More about that here.) “That obfuscation makes me want to find ways, outside of the shadows of these streaming services, to reach listeners, capture their attention,” he said, when asked about whether or not algorithms were fair to artists like him. “Along the way, you hope to inch your way into a preferable position in those recommendation feeds.”
Several companies are looking to level the playing field for artists when it comes to the fairness of discovery. Bandcamp, a 12-year-old music platform based in Oakland, CA, has made connecting artists and consumers their central mission by making their platform human. Their successful editorial operation, Bandcamp Daily, gives users a look into new and notable releases that fly under the radar of many outlets, and artists can create personalized pages that list supporters. Humanized discovery is at the heart of Bandcamp’s purpose.
While their business model of selling digital and physical copies of music directly to consumers isn’t novel, their commitment to transparency in compensation is something that has set them apart in the music industry. Bandcamp patrons have made nearly half a billion dollars worth of purchases on the platform since it started, and it’s become an invaluable tool for artists who are trying to make a living creating music for a dedicated fanbase.
Stem is also doing its part to give artists a comprehensive toolkit to address fairness in a world dominated by streaming platforms and the will of the algorithm. Stem was founded in 2015 by Milana Rabkin Lewis, a former digital talent agent who saw artists struggling to understand their monetary worth in an industry defined by its opacity.
Lewis doesn’t see algorithms as the enemy, but she does believe that because artists are often left in the dark when it comes to granular data, they can be leaving money on the table. “Streaming platforms launched dashboards to help artists understand how the content performed, the streaming metrics, the volume of consumption — that’s all great,” she said. “But it doesn’t do anything to address an artist’s understanding of what their value is.” Stem lets artists connect that data from streaming to actual financial data to help artists understand how much financial leverage they have in the industry, powerful information when you think about an artist trying to negotiate a record deal based on their streaming numbers.
Whether they’re fair or not, algorithms in music are here to stay. They’ve become standard practice for major streaming platforms that are dead set on taking listener data and transforming it into suggestions that keep you listening. Whether that’s created a generation of uncritical listeners or not is up for debate, and the lack of human touch has artists like Young yearning for the days when a passed record meant something more than another click. “I have discovered a lot of music that I really enjoy through the recommended listening portals,” he said. “But I don’t think they’ve ever given me music that I truly cherish. The recommended playlists will always have value, but did your high school crush ever burn you a mixtape? That’s worth a lifetime of algorithmically derived playlists.”