How many releases does it take to break into the algorithm?
It’s a sad indictment of where art has found itself that the word ‘algorithm’ is so freely banded as a success metric. It’s even worse, when artificial intelligence is spoken about as some sort of musical overlord that needs constant pleasing to obtain said success: “The algorithm really likes this one”.
But what does the algorithm even mean? As a Spotify user, I marvel at the ease with which the platform identifies my taste, recommends similar music, and then bundles it into hyper-focused genre playlists. Post-punk, aggrotech, microhouse—the subgenres are endless. While you may mistake it for carving out an exciting cultural niche, it’s actually a much wider ploy to understand your listening habits.
“At the beginning, the editorial playlists were made by people” former Spotify’s data alchemist Glenn McDonald told The New Music Business Podcast. “The breakthrough idea which was called algotorial was to have a human editor pick a pool of tracks, tracks that made sense with the premise of a playlist” he continued, “but instead of them being sequenced by a human, the algorithm would take that pool of tracks and your listening and use basically the same technique as discover weekly, but instead of pulling from the whole universe to surprise you, would pull from this 500”.
So what does that mean more generally, that creativity in music is essentially governed by a select pool of tracks that are responsible for defining a genre? Perhaps the more pressing question is, what qualities would be measured when assessing a track’s similarity to one within the pool?
This is where BART comes in, Spotify’s artificial intelligence tool whose full name is Bandits for Recommendations as Treatments. Using three main functions, BART draws tonal comparisons between songs and recommends them to users. The first of the three functions is, natural language processing, which analyses the language, lyrics and content of a song. The second is Raw Audio Analyzation, which aims to detect the “mood” of a song’s audio, which is then characterised in somewhat reductive terms: upbeat, chill, heavy, minimal, or downcast. Lastly, BART uses collaborative filtering, which compares new songs to a listener’s current habits using established mood characteristics and lyrical content.
Now, on a platform that is reported to see 120,000 new songs released daily, it’s understandable that these songs are processed with some artificial intelligence. But ultimately, the most important question is, where does this leave the artist? How does a young bedroom artist who could have quite feasibly just penned a modern ‘Strawberry Fields Forever’ get that listened to, and is that one song alone enough? Apparently not.
In a recent episode of the Burstimo podcast, it was revealed by a Virgin Music employee that the average amount of songs released by an artist before they “break” was 32. For context, that’s three lots of Magical Mystery Tour, the album responsible for the release of ‘Strawberry Fields Forever’. Not only is it troubling that labels have now enumerated the prospect of “breaking,” but the bar has been set at a simply unhealthy level for emerging artists. What often happens in the follow-up to an artist being granted the privilege of being featured on a platform editorial playlist is an extension of gratitude, an Instagram post from the artist thanking the faceless streaming platforms for their exposure.
What is at threat isn’t just the artist either. As Kyle Chayka, explains in his book Filterworld: How Algorithms Flatten Culture: “When a human interprets a piece of art, it adds value rather than takes it away. An algorithm has no capacity to interpret”.
With artists at the mercy of a 32-song algorithm rule and the success of it relying on artificial intelligence’s ability to interpret its tonal mood, where does that leave the fan, the critic, and the journalist in the chain of artistic interpretation? Traditionally they played a necessary role in not only recommending music but also challenging your taste and widening your palette to spark essential conversations you would have on the grounds of venues, pubs or school classrooms. So, as a singularity within the arts hurtles towards the accepted norm, it begs the question: does it take 32 songs for an artist to break the industry, or 32 songs for the industry to break the artist?
