A Facebook survey
When online music streaming first became a thing (Napster and Pandora launched at the turn of the millenium, YouTube and Spotify a few years later in 2005/2006 respectively), these services promoted the possibilities they proposed for user-initiated, search-centred listening, positioning themselves as the ‘saviours’ of a music industry whose stability was threatened by illegal digital downloading and file-sharing.
Since then, most popular streaming services have done a full-on U-turn, angling their marketing to emphasise music discovery and curated, personalised listening recommendations. The shift is illustrated in the evolution of Spotify’s primary interface, whose initial feature was an empty search box: a symbol for endless music access as part of the company’s push to be associated with liberated, democratic listening. By 2013, the open portal for music listening choice had been replaced by dozens of pre-curated playlists, described (in that sickly-sweet tone they have) to be ‘made for you’. In the era when music is seemingly available to us in abundance, personalised recommendation is cast as the solution to the paradox of choice that listeners are now faced with.
Although each popular streaming service offers a different version of personalised music recommendation, these listening suggestions are usually curated through algorithmic processing. One of the biggest influences that personalised recommendation systems have is in what tracks actually ‘find’ us, what we discover, enjoy and remember, leading to repeated plays and further avenues of music listening. As music taste is most often seen to be expressed in acts of consumption, these algorithms potentially have the capacity to shape listeners’ music tastes; a pretty important role considering how significant music can be in the very formation of personal identity.
But are these coded systems really replacing traditional cultural gatekeepers such as radio DJs and music critics in their intermediation of musical culture? And what happens when these cryptic codes become the new tastemakers? Should we be worried about the effects these allegedly impenetrable algorithmic ‘black boxes’ have on our listening habits and, in turn, the production of music? I decided it was time to do some investigation into these oh-so-mysterious algorithms. Naturally, it took the form of a wee survey amongst 100 Facebook friends, to explore their processes of music discovery.
The first step of the survey was to establish whether this group of 100 participants (mostly aged ‘18 to 24’ or ‘45 to 54’ – thanks for sharing the link, parents’ friends) are users of streaming services. Perhaps inevitably, the majority of survey participants (61.62%) selected online streaming as their primary listening medium, with radio nabbing second place. This is interesting: as arguably the original audio ‘streaming service’, radio is another medium that emphasises curated music selection.
I was working with a group of streamers, that much was clear. But are these participants obedient to their chosen platforms? Are they the kind of listeners that let Spotify do the choosing, leaving their music choices to the whim of the mood or activity based playlist; to the ever-discerning algorithm? Not necessarily, the results suggested. The next question asked about favoured music discovery methods, and the responses were as wide ranging as you’d imagine. TV and film soundtracks, watching support bands and scouring festival lineups all came up numerous times, as did social media groups and the judgements of music critics. But above all the most popular answers were surprisingly old hat. Among these 100 participants the most popular means of discovering music was through suggestions from friends and/or family, followed closely by our old pal, the radio. Two explicitly human music discovery tools.
So it seems that consumers still seek out and value listening recommendations from other sources. As much as streaming platforms hold personalised recommendation at the core of their services, it’s still possible to simply use them for their easy access to an ‘unlimited’ catalogue of music to be gleaned at the consumer’s discretion. And this point where human agency rubs up against listener agency is the sweet spot for understanding contemporary music taste formation.
Humans are not vessels who mindlessly drink up whatever tunes their ‘Discover Weekly’ or ‘recommended for you’ serves them, and the next of my survey questions aimed to ascertain the ways people approach algorithmic music suggestions. The majority of participants claimed to find the suggestions presented to them to be ‘accurate’, in that they were congruent with their existing tastes. Yet most participants also described their music tastes as unchanged by the music they had discovered through streaming personalisation.
Music recommendation systems are ‘personalised’ in that they are usually based on, among other things, previous listening. This can result in a so-called ‘feedback loop’ effect, with algorithms acting merely as ‘taste-reflectors’ rather than ‘taste-shapers’, potentially narrowing breadth of listening and restricting contact with new styles. Many survey answers pointed towards this sort of phenomenon, with participants
claiming that they tend to be served more of what they already know and listen to. One participant had even noticed that their ‘motivation to challenge [theirself] with new music’ had been ‘stifled’ by following their recommended streaming.
But the effects of algorithmic listening loops may be more troubling than our music tastes becoming a bit boring. Just as humans are not merely empty listening vessels, algorithms do not present their recommendations completely neutrally. In how they sort and hierarchize information, in what they choose to include and exclude, they exert power.
A 2017 study by the Swedish academics Maria Eriksson and Anna Johansson investigated how gender and gendered tastes may be constituted through algorithmic recommendations given by Spotify, the most widely used streaming service. They found that 80% of recommended artists across numerous genres were made up entirely of men. Maybe this comes as no surprise – the history of popular music is hardly one of egalitarian diversity. Perhaps the algorithm is just reflecting an inequity that already exists (which in itself would be a less than desirable means of driving music discovery). But based on the overwhelming extent to which musicians not identifying as men were excluded from the recommendations recorded in Eriksson and Johansson’s study, it would be fair to suggest that the platform is, in fact, contributing to the entrenching of inequalities in their listening suggestions.
In a similar study by writer Liz Pelly into Spotify’s non-personalised, highly popular playlists such as ‘RapCaviar’ and ‘Today’s Top Hits’, she also suggested that the platform is playing a role in ‘upholding and even exacerbating gender inequality in music’. She found that this handful of playlists, which between them have billions of listeners worldwide, were staggeringly dominated by men. The worst was the playlist ‘Hot Country’, where 92% of tracks featured no women or non-binary musicians at all. Although Pelly’s investigation concerned primarily Spotify-curated, non-personalised playlists, the feedback loop effect of the algorithm means that tracks featured on the platform’s most popular playlists inevitably end up on a greater number of user-generated and algorithmically-curated playlists, such as the ‘personalised’ recommendation playlist ‘Discover Weekly’. These tracks will then become yet more popular as a result; again feeding back into the cycle. By reflecting existing gender biases in listener’s music tastes, the workings of the Spotify algorithm means that these disparities become further entrenched, in effect codifying bigotry.
As our survey suggested, a significant degree of user agency comes into the ways that listeners use online streaming services, and consumers continue to put in the effort to discover new music in a variety of ways. Yet the fact remains that these Spotify-curated playlist have millions of regular listeners worldwide. Exposure matters in terms of diversity—as fewer diverse musicians are listened to and discovered, fewer diverse musicians will see music as a viable option for them. Even the listeners most assertive in their music discovery will find it hard to find diverse music if it ceases to exist.
Statistically, it is younger people who use streaming services to listen to music – the consumers with the most impressionable and malleable music tastes. Although humans undoubtedly continue to play a vital role in music discovery, whether as radio DJs or as keen muso friends, the influence of these algorithmic tastemakers is only going to grow. As the effects that they have on the music industries become more pervasive, it will (in true algorithmic spirit) become increasingly difficult to break the perpetually revolving loop.
The very nature of an algorithm means that is is not fixed in nature, but constantly emergent and unfolding. So let’s use our consumer agency to shape what kinds of curators we want our algorithms to be. Let’s feed them the listening that we want to be fed back out, in terms of genre, but also in diversity and curiosity. Listeners, let’s lean forward in our discovery, and not become lean-back captives of our algorithmic curators.
Words, Kate Walker