Music Curators and Algorithms: Beyond the Play Button
Once upon a time, record stores dominated the music retail market. However, in the late 1990s, the advent of file-sharing services such as Napster sparked the hope for free, democratic, and unmediated circulation of music. All it took was a simple software installation, a few keywords, and users could download their favorite tracks directly onto their computers. Although far from the convenience of modern-day technology, these services were revolutionary at the time; suddenly, listeners gained access to a virtually endless music repository which allowed them to exchange files with fellow internet users. At the same time, these digital communities were shielded from the influences of music journalists, radio programmers, and other experts. Music became free in its broadest sense: devoid of cost and, at the same time, of the restrictions of industry gatekeepers.
However, dreams of music-sharing anarchy were short-lived; within a few years, streaming platforms stormed onto the scene, putting an end to the network-based revolution. Spotify was launched in 2006, and Amazon Music followed a few months later, with Apple Music and many others smaller companies appearing in the subsequent years; these platforms gradually imposed themselves as the new intermediaries of music. Today, streaming services hold an unmatched power on the industry, with music streaming representing 65% of the global total music industry revenue. As a result, digital music listening gradually shifted from database-centered search services designed for informed and active listeners, to platforms that conveniently showcased a selected collection of music in their front-window. Instead of simply providing access to music, the role of streaming platforms quickly became that of producing soundtracks for its users’ daily lives. A quick experiment proves this point: any Spotify user can open their app and scroll through the countless playlists created for them. Do you have dinner plans with friends? Simply press play. Need motivation for a workout session? The platform knows how to hype you up. In this regard, streaming platforms should not be considered as mere music distributors, but rather as producers of a musical experience – an individually tailored and curated one.
This difference is not trivial, because conceiving of streaming companies simply as mediators does not acknowledge the actual power they hold. As Daniel Ek, founder of Spotify, once stated, “over 30% of consumption on Spotify is […] a direct result of recommendations made by the platform’s own algorithms and curation teams… [which] puts Spotify in control of the demand curve”. Fittingly, a 2020 study found that securing the first position in New Music Friday – one of Spotify’s most popular playlists – is worth roughly 14 million streams, which translate to streaming revenues ranging from $84,000 to $117,000. Since platform-curated playlists reign supreme in popularity within their ecosystems, by determining which song will feature in which playlist, streaming services have the upper hand on determining the circulation and commercial destiny of music.
While the disillusion of an independent musical experience makes it easy to point fingers at big corporations, it isimportant to remember that the decisions that shape it are ultimately taken by a few powerful individuals. In the early 2010s, as streaming emerged, several platforms started hiring experts to curate their music recommendation systems; ever since then, the main job of curators at streaming companies has been that of creating and updating playlists. It is estimated that, globally, there are only a few hundreds of these editors working at major streaming companies – an elite of influential music specialists. These people hold the fate of countless musicians in their hands; not by chance, a 2017 article on The Wall Street Journal dubbed the global head of hip-hop at Spotify as the “most influential gatekeeper in the music business”, whose curatorial choices for Rap Caviar – a Spotify playlist counting more than 15 million listeners as of October 2023 – “set[s] the agenda for the hip-hop industry”. Despite their power, curators are not the only editorial authority within streaming music companies, as they need to rely on their most loyal collaborator – the algorithm. Music recommendation algorithms analyze and break down songs in discrete quantitative data, such as number of plays, number of skips, and number of saves.This data is collected on two levels – global and individual. At the global level, data are used to evaluate the performance of songs, playlists, and artists, as well as to predict trends, and to inform editorial choices carried out by curators. At the individual level, usage data are collected and used to craft tailored user experiences in the form of algorithmically personalized playlists and recommendations.
If the fact that our musical experience is in the hands of a handful of people does not sound sufficiently concerning, perhaps the awareness that the algorithms upholding these power structures can perpetuate the racial bias held by their designers should do the trick. It is no coincidence, but rather a symptom of a systematic reoccurrence, that automated systems might reflect the short-sightedness of their programmers, as frequent instances of poorly trained face recognition systems unable to recognize Black people remind us. Within music streaming platforms, reiteration of racism might be more subtle, but equally impactful.
“If you’re a singer and you’re black, you’re an R&B artist. Period.”, once claimed Afro-American artist Frank Ocean. Ever since its inception, popular Western music struggled with a racialized conceptualization of music genre, and streaming services seem to have successfully embraced this tradition. Spotify’s genre tag system is a great example of how human biases can slip through lines of code, with the algorithm systematically assigning different quantities of labels to different genres. For instance, rock music is generally described by more genre tags than hip-hop is– glam rock, punk rock, garage rock and many other tags can describe different stylistic variations of the former, whereas hip-hop songs are more likely to be assigned generic macroscopic-level labels such as rap. By grouping many different stylistic characteristics into one big generic category, these labels paint a reductive picture of the genre, often reinforcing the oversimplified popular perception of hip hop artists as indistinctly Black and male. To put Eco’s notion of cultural coding in simple terms, we name what we recognize, and we recognize what we value. Therefore, these patterns simply reflect the biases in attribution of value that those in charge of designing genre classification systems – music streaming companies’ employees – hold.
But the shortcomings of these algorithms do not stop to problematic implicit assumptions: by creating a loosely defined network of stylistic labels, undercoding certain genres results in poorly performing music recommendation systems. For example, a hip-hop track that presents stylistic influences from other musical traditions – such as jazz or gospel – and which gets classified simply as rap will not be linked to their networks. If the jazz and gospel tags are not part of its classification, the algorithm will not recommend other works from these genres, and users will be presented with other generically defined rap tracks. Thus, this algorithmic limitation carries important real-life consequences by generating lower circulation of marginalized musical genres, which constrains their economic value potential and affectsthe livelihoods of their artists.
We do not know exactly what the future holds for the world of mass music consumption, but it seems that algorithms are here to stay. As of now, data-driven analyses of listening behavior are revealing new patterns within streaming communities; for one, they might be detecting a departure from rigid genre boundaries towards a nuanced, identity-focused musical landscape. It seems that sometimes identity labels such as sad girl aesthetic – which group artists of different music genres, but who share common cultural codes – might be more fitting to describe communities of listeners better than traditional ones. In other words, in an era of unprecedented genre-blending, it might be of little use to rely solely on traditional genre classifications rooted in musical similarity, as they become less effective in capturing the shared cultural connections within communities of listeners.
To the extent that the relentless commercialization of music is reducing the notion of genre to that of collective listening patterns, it becomes important to wonder how platforms are shaping our perception of music. Just as one day in 2020 thousands of Spotify users were surprised to discover that, according to their usage statistics, their favorite genre was Escape Room – a label arbitrarily created by one of the platform’s employees to describe a cluster of listeners. The ways in which these new practices of consumption will shape our understanding of music remain uncertain. As of now, imagining that our next musical obsession might depend on a curator’s mood does not sound too far-fetched. Nevertheless, what is a concrete reality rather than a possibility is that automated processes harbor the potential of reproducing social inequalities. Overall, the dark side of algorithms and the concentration of power over many music listeners in a few hands might raise some concern and should warrant a critical evaluation of our listening habits. The seemingly unstoppable advance of the datafication and platformization of our lives continues, and it shows no signs ofsparing our musical experiences.