Image from Dall-E 2
Music has been an integral part of human life since, well, as long as we can remember as a species. It brings forth emotions in us in a way that we can’t quite understand. How we came to enjoy and employ music in our daily lives is still a mystery to scientists.
Regardless of the origin of music, reason for its existence and scientific underpinnings, it is a phenomenon that is very much present in our modern lives and can be found in the deepest recesses of our social lives.
Now, within a capitalistic western society, and much like many other ventures – creative or not – the consumption of music has become industrialized. We talk of the ‘music industry’ when we want to discuss how a piece of music can become so popular that it has been streamed over 3 billion times (Ed Sheeran – Shape of You), and to be honest, nobody is really capable of explaining in its fullest how this comes to be (and those who say so are most likely overly confident in an otherwise erroneous answer). The music industry is an extremely complex entity.
Motivation for the research
I have myself been engaged with this complex entity for the past two years and have been trying (and still try) to understand how the whole puzzle fits together. I have come a good way, partly due to the method I have employed, which has been to identify the individuals and organizations who hold the power, who are the ‘gatekeepers’ of the music industry. The identification process has been threefold: (1) I have written down the names of people and organizations I have come across through projects I have worked own; (2) I have conducted desk research and (3) I have been participating in courses and industry presentations. The list, whilst non-exhaustive, includes radios and radio hosts, large record companies and their board of directors, bookers for festivals and venues, and streaming platforms.
The motivation for identifying these actors and how they are working together or against each other, has been to study how they reproduce certain types of negative bias, the most identifiable of those being a gender bias. In 2022, the top 10 most streamed artists in Denmark featured only male artists (Hjort, 2022). On a global scale, only one non-male artist, Taylor Swift, can be found on the top 5 list of most streamed artists (Sprangler, 2022).
Within this list of gatekeepers, I mentioned streaming platforms, which, of the main actors in the music industry, are the one’s which I believe should warrant a higher level of scrutiny. Compared to other actors in this industry, they have an impact on the vast majority of musicians. Almost all artists opt into having their music on streaming platforms (exact numbers would be hard to find, but of the musicians I know, all but one group have their music on streaming platforms), whereas other actors, even major labels, only deal with a smaller portion of musicians. On the other end of the spectrum, the vast majority of music listeners use a streaming platform to access their music, be it YouTube Music, Spotify or Apple Play. Accordingly, how music is consumed has changed, from listening to album on vinyl – which granted, has made some sort of comeback, but still is a rather niche phenomenon – to human and algorithmically curated playlist listening on streaming platforms.
Specifically the market leader, Spotify, which as of Q2 2021 held over 30% of the market share (Forde, 2022). As a company that primarily qualifies itself as a technology company, their impact on how people listen to music and which artists get promoted by their services justifies a deeper look into the inner workings of their company and the platform that they provide to listeners and artists.
Research into AI technology
My inquiry into Spotify will be focusing on the systems of recommendations that Spotify has put into place for their users. The focus on recommender systems comes from my own experience with Spotify’s platform – which features many playlists and functionalities that draw from the power of recommender systems to sift through millions of songs in order to propose the most fitting music – which I myself use to a very high degree whenever I listen to music (especially Discover Weekly). I can imagine that I am not alone in this, and many other people rely on algorithmically curated playlist to deliver them tailored playlist that they will most likely enjoy.
Executives at Spotify know this and have therefore invested lots of resources into developing algorithms that can better cater to the tastes of individual users, using cutting-edge research into AI topics such as Machine Learning, Recommender Systems and User Modelling. The thinking seems to be that algorithms can do the job of radio hosts and tastemakers better, and with less risk of collusions and bias.
However, these technologies do not come without their own share of problems, specifically when it comes to bias. A quick search on scholar.google.com for the terms ‘machine learning bias’ since 2022 returns a total of 37.200 results. For ‘recommender system bias’ and ‘user modelling bias’, the total is 17.300 and 17.200 respectively.
Finding out the extent to which AI technologies are prone to bias made me feel like I was on the path to something very interesting. What I was not aware of, was that the next hurdle – studying recommender systems within Spotify – would prove to be the most difficult step to overcome.
How do we open the black box?
As other researchers have found out, conducting research from within on how Spotify operates its platform is nearly impossible, if not downright impossible. Even though I was provided access to the entirety of the source code, I wouldn’t be able to decipher anything from it, lest be able to tell how this might cause any form of bias.
Luckily for me, the lab manager at ETHOS lab was quick to (verbally) recommend to me to read some of the works of Nick Seaver, which led me to the book called “Spotify Teardown: Inside the Black Box of Streaming Music” by Eriksson et al. (2019), and “Computing Taste: Algorithms and the Makers of Music Recommendation” by Nick Seaver (2022). The former introduces the reader to a couple of researchers faced with the same issue that I am encountering, namely not being able to access the back-end of Spotify to perform some data collection and analysis. Their responses to this hurdle was to conduct a series of ‘experimental interventions’, which has allowed them to peer into the inner workings of Spotify. The latter, “Computing Taste”, deals with the same topic, through Seaver’s own ethnographic research of music recommender companies.
My own method of investigation will be heavily inspired by the stance that these researchers have taken.
Methodological framework
The main method I am looking to employ is the walkthrough method by Light et al. This method employs ethnographic methods and mobilizes concepts from STS to conduct a critical analysis of an application (Light et. al., 2016). My thinking is that even though I cannot access the back-end of Spotify’s services, I can still peer into their application, and provide some analysis that will open the black box, even just a little bit. As Light et. al. mention: “Another way of understanding the influence of non-human actors is through consideration of a technology’s materiality and the affordances it extends” (p.886).
What’s next?
Ideally, I would like to employ a second method, which could somehow allow me to reverse engineer Spotify’s algorithm in order to understand what kind of bias they reproduce, as well as the degree of bias being reproduced. Hopefully, further research into and maybe even some serendipitous luck might provide me with such a method.
References
Hjort, M. (2022). Her er årets mest streamede sange på Spotify – Tobias Rahim og Gilli dominerer. SOUNDVENUE.COM. Retrieved December 16, 2022, from https://soundvenue.com/musik/2022/11/her-er-aarets-mest-streamede-sange-paa-spotify-tobias-rahim-og-gilli-dominerer-500386
Light et. al., (2016). The walkthrough method: An approach to the study of apps. New Media & Society, 20(3), pp.881–900. doi:10.1177/1461444816675438.
Spangler, T. (2022). Spotify Launches Wrapped 2022: Bad Bunny, Taylor Swift Are Most-Streamed Artists of the Year. [online] Variety. Available at: https://variety.com/2022/music/news/spotify-wrapped-bad-bunny-taylor-swift-1235444491/ [Accessed 16 Dec. 2022].
Forde, E. (2022). Spotify Comfortably Remains The Biggest Streaming Service Despite Its Market Share Being Eaten Into. [online] Forbes. Available at: https://www.forbes.com/sites/eamonnforde/2022/01/19/spotify-comfortably-remains-the-biggest-streaming-service-despite-its-market-share-being-eaten-into/?sh=3181df5c3474 [Accessed 16 Dec. 2022].