Every Monday, a small cultural artifact appears in the inboxes of millions of Spotify users across the globe. Discover Weekly is what Spotify’s team counts on to distinguish its service from the handful of other offerings on the market. It is an algorithmically generated playlist tuned to each user’s individual taste, and the algorithm is frighteningly effective at keeping listeners engaged with the system. Last May, Spotify reported that over 5 billion tracks had been played off Discover Weekly alone.
Despite the ubiquity of music recommendation algorithms in the modern listener’s discovery process, researchers have given little consideration to the engineers behind them. Yet, the engineers who design these algorithms play an important cultural role in shaping musical trends, defining new genres, and altering the way people treat music. This is where Tufts professor Nick Seaver comes into the picture. Professor Seaver has devoted the past six years of his career to an ethnographic study of the developers who create music discovery algorithms.
To effectively study this group, Seaver has had to deconstruct the stereotypical image of engineers as culturally inept. “There’s this common trope – often in popular media, but also among academics in humanities and social sciences – that what happens when engineers encounter cultural stuff is that they sort of mess it up. They reduce everything to numbers, they assume that people behave totally rationally, they supposedly do all of these things that [academics in humanities] know better than.” The stereotype is rooted in a perceived disconnect between the cultural and the technological, a spectrum in which music is firmly planted in the middle.
As a cultural phenomenon, music is universally human but distinctly different between societies, often to the point of sounding alien to unfamiliar listeners. As a technological phenomenon, music has always evolved with new technological developments. The sonic palate available to a modern musician is nearly infinite, thanks to the ubiquity of synthesizers and digital audio workstations that can incorporate both analog and digital instruments. More recently, it is the technology underlying the means of production and distribution that has evolved – anyone can record music on their laptop, put it on Spotify, and find popularity without radio airplay. As with all technological upheavals, this disruption meets resistance from both record labels and music critics. The industry sees their business model disrupted by streaming services like Spotify. Likewise, critics are afraid that algorithms will not only make their job defunct, but further will fail to expand listeners’ tastes. They are too good at selecting songs that regurgitate the user’s existing palate. Music journalists like Ben Ratliff at The Guardian exacerbate this fear, pumping out thinkpieces with titles like, “Slave To The Algorithm?” Are critics like Ratliff right? Are the engineers who create these algorithms destroying musical taste?
Music discovery in the pre-digital era was a laborious process. Music aficionados would dig in the backs of record stores in search of obscure records. Others would turn to their favorite radio host or music critic to do the work for them. Even in the present day, the process remains the same; music blogs like Pitchfork sift through mountains of mediocrity to award their highly regarded distinction, “Best New Music” to a select few songs per week. A music selection algorithm takes the effort, and thus the creativity out of music discovery. To critics who have based their career on selecting songs that they feel are most culturally important, an algorithm taking the place of their job is the destruction of a culture. When a music journalist claims an algorithm will destroy musical taste, they base their claim on some definition of what culture is, or should be.
While the journalistic approach of the moment seems to hinge on the assumption that the writer’s definition of taste is the only correct definition, and any change to that detracts from the cultural value of music, professor Seaver takes a different approach. “Rather than going in and saying ‘I know what culture is. I’m a cultural anthropologist, therefore I’m an expert and I’m going to tell you how you’re wrong,’ I take from the outset [that] my job is to go in and study how a particular group of engineers imagine what culture is, what technology is, what taste is, and how they all interact.”
Seaver’s work strays from the presumption that musical taste is static. “Taste is partly a technological phenomenon,” he says, in that the selection of music available to comprise a listener’s taste is constantly evolving. Two centuries ago, musical taste was defined by what live music was available locally to the listener, if they could afford it. Fifty years ago it was developed by digging through stacks of vinyl in a record store, an expensive endeavor both in time and money. Furthermore, taste as a signal of social status has also evolved. “In the past,” explains Seaver, “to be a high status person was to like the music of high status people – so you like Opera, you like Classical music, and you don’t like the cheesy, cheesy pop music. That used to be the way taste worked. Now, there’s a lot of research about omnivorousness.” Today, virtually all varieties of music are available to stream, for cheaper than a monthly utility bill. “Back ten, twenty years ago you would hear [people say they prefer] ‘everything but Rap and Country,’ which gives you a side of what’s going on. You like everything except for two genres, either associated with black people or with poor people.” Taste is clearly changing, and there’s no indication that it will stop today. The future could hold songs that automatically customize to a listener’s preferences. Modern musical taste is increasingly shaped by aesthetic movements like Vaporwave and Health Goth. According to The Fader writer Adam Harper, these aesthetic movements are replacing ‘scenes’ and genres.
One critique of music recommendation algorithms is that big corporations like Spotify are wielding influence against listeners’ taste. However, Seaver notes that “a more important critique is that these corporations are changing what taste is. If we’re going come out of this decade with Spotify and maybe one other streaming company streaming all of our music, the hidden infrastructure they use is definitely going to shape what taste is on a large scale.” Consumer feedback, however, can also reduce the algorithmic bias in music recommendation. “In the little interstices of the ordinary day when you’re an engineer and you’re just tweaking something, you’re going to test [the algorithm] on yourself,” says Seaver. This once produced pretty extreme bias in how Spotify, for example, sub-categorized genres. “There used to be four different genre buttons for Metal – Heavy Metal, Death Metal, Black Metal, and so on – and only one for Electronic. For both Electronic and Metal music, the fans are known for being very aggressive taxonomizers. The fact that one of these had a lot of categories and one had only one – that’s not in the algorithm per se, but that’s a very good example of how much attention you pay to something is a function of how much you know.” With feedback, engineers are able to work on a wealth of listener data from their service. The larger the services get, the more effective they become for all users.
Seaver has also found that engineers of music recommendation algorithms are cognizant of the cultural significance of their work in shaping musical taste. Many actually entered the industry out of a desire to enable a middle class of musicians. In Seaver’s experience, “usually what people say when they are working in this space is, ‘I started as a musician, and I was wondering how anyone was ever going to find my music. I started studying computer science and I decided that I was going to work in this space to help people discover obscure music.’” Indeed, many new features on streaming platforms are targeted at promoting lesser-known artists, and although payment of artists is currently a contentious issue in streaming, there is some hope that in the future artists will be able to make a living while still making relatively unpopular music. Professor Seaver explains that many existing streaming services, notably Pandora, were founded with the hope that, “to become a successful musician, it doesn’t mean that you have to either stratosphere up and become super popular or live in misery forever.” Artists making a decent living is good for streaming services, too.
While it is scary to contemplate the implications of algorithms redefining taste, it is important to avoid the paranoia that dominates most public discourse on the topic to date. While algorithms are not quite yet able to recreate the ‘back of the record store’ epiphany that music junkies crave, perhaps in the future they will do something even better. Professor Seaver’s work highlights the people behind these algorithms, and their awareness of their enormous cultural impact. These people are just as dedicated to cultivating a thriving music scene as the journalists who decry the centralization of recommendation systems; but while the latter are hung up on their own preconceptions of taste, the former are busy redefining it. Musical taste is constantly evolving, so it’s best to embrace the change with open arms.