Ask an Expert: Music Edition

Our team was able to get musicologist and digital humanist Dr. Jada Watson on board for a written interview centered around the effects of machine learning on music. 

Q: In your work, have you seen the influence or impact of artificial intelligence (AI) in the music industry?

The music industry has a long history of innovative use of technology in all realms of creative and promotional work, and that is certainly true in this algorithmic era we’re in right now.

There are several ways in which automated generation and AI is used in the music industry. On the creative side of the industry, some music makers have turned to automated services for the mixing or mastering process. Current technology allows for automated workflows (mixing and mastering), sample-driven or computer-generated instruments, and even melody- or chord-generating technologies. This technology can even be used to analyze Big Data to determine musical trends and create AI-generated music. This Forbes article relays a story of Alex da Kid’s use of IBM Watson to analyse songs and cultural data to determine a theme for what would become a computer-generated song called “Not Easy.”

Algorithms underpin most (if not all) of the digital tools used in music distribution, including the programming software used by radio programmers and the recommender systems of digital service providers (YouTube, Spotify, Pandora, Apple, etc.). Not only do these tools play a critical role in distribution (especially important for unsigned artists looking to distribute music and build an audience), but they are also used for various forms of artist discovery — by labels looking for their “next big act” and by audiences looking to find new music.

Q: How do you think AI impacts how music is sold or promoted?

AI is radically transforming how the industry works and this has both positive and negative consequences. 

AI plays a critical role in the promotion/distribution of music. Digital service providers (DSPs) like YouTube, Spotify, Pandora, and now TikTok (to name a few) make it possible for artists to launch their career without the aid of the traditional label-system. These is a long history of artists following a DIY-method of self-producing and releasing music, which (in the past) involved the artist pressing albums and distributing them on their own. But with the aid of DSPs, artists can cut out this middle step in distribution, upload their music to Spotify (for example) and then push their music to audiences via social media. Artists can then use the tools available to them to build their audience, attract industry attention, and remain completely independent of the label-system. For many, this is critical, because it means that they can retain control/ownership of their music (the masters) and creative control of their career. For others, these tools allow them to build an audience and attract industry attention in the hopes of landing a recording contract. In many ways, this work is critical and can be leveraged in the process, because it means that they have already developed an audience interested in their music that would be (presumably) ready to purchase/stream their music once signed. Whether you’re looking for independence or aiming for label support, then, these tools are critical (especially in this pandemic world when touring isn’t possible or as easy) for artist promotion. For example, most DSPs offer a pre-save function that does not just help to build anticipation for an upcoming single or album release, but it has been shown to significantly boost first day streams, which can in turn up an artist’s changes of their song being playlisted. This is a crucial marketing tool offered by the service, then, one that has the potential to shape a song/album release in powerful ways.

However, this really just accounts for their use value, and doesn’t take into consideration how AI might impact an artist’s career, which can have both positive and negative consequences.

For example, the more attention an artist can get through follows, likes, and play listing, the more likely they are to cut through. Functions like “related artists” or “sounds like” or discovery/browse options, which use metadata and listening habits to recommend artists based on a user’s listening preferences plays a big role in discovery of new music and new artists. For an artist, landing music on a Spotify curated playlist (for example), can be the opportunity that changes the trajectory of an artist’s career. Ottawa-based singer fanclubwallet’s career is an excellent example of what can happen when the singer’s “Car Crash in G Major” was added to Spotify’s “New Music Friday” playlist. A much-followed playlist, the song racked up 1.5 million streams and the attention of an agent and label. My student, Kate Thornley, is currently doing her DH Capstone project on playlisting, looking at the impact of Spotify-curated versus fan-curated playlists on the trajectory of Canadian artists’ careers. Her findings have so far been quite fascinating, showing the power of fan-curated playlists, with fanclubwallet as just one of the artists she follows.

But like every realm of society, DSPs have played a critical and damaging role in exacerbating systemic inequities within the industry. On the one hand, we see studies that show that (like the headline of this RollingStone article), “90 percent of the streams go to the top 1 percent of artists.” Thus, despite the power and potential for discovery, streaming services have contributed to the widening of the gap between top artists and everyone else. It’s been reported that 60,000 tracks are uploaded daily to Spotify. With so much music being ingested in the system, it makes complete sense that discovery can be really challenging for new artists, and that the tool is truly benefitting artists with the resources behind them to push them music to the fore.

In many ways, then, this idea that streaming would create equal opportunity is only true to a point — and that point is a free platform for uploading and sharing music.

What happens beyond the upload/share of music, is dependent on artist having both label support (money, digital promotions teams, and marketing strategies) and audiences (i.e., an existing fanbase).

Coupled with this imbalance is the fact that the broader industry is a deeply and harmfully oppressive system in which Black, Indigenous, and artists of colour, women, 2SLGBTQIA+ and disabled artists are pushed to varying degrees to the margins of the industry. This industry was founded in the 1920s along a musical colour line that has been carefully maintained for the last 100 years through data — through radio airplay, popularity charts, recognition systems, and now streaming services — that has dictated participation in the industry (in general), but also genre categories (specifically). The culture of streaming services, then, reflects the culture of the industry around it and so we see significant inequities emerging within the AI that underpins these services.

For example, country music remains one of the most inequitable genre categories in the music industry. This is an industry founded on structural racism, sexism, heterosexism, and ableism and even 100 years after its birth, we are still seeing significant levels of exclusion of white women, BIPOC and 2SLGBTQIA+ artists, and the absence of disabled artists. Spotify’s system, then, replicates these inequities through its recommender system because it reflects the culture around it. I did a short experiment on this back in Fall 2019, after country singer Martina McBride found that the service was only recommending songs to her by white, male artists. In short, it took 121 songs before the system recommended a song by a female artist and all 40 songs (9%) recommended (out of 420) by a female artist were backended in the algorithm. For the general user then, AI has a significant and damaging impact on their experience – and some might not even realize it. If they are passively listening to country music and letting the recommender system do the navigating for them, AI is curating a playlist dominated by white, cis/het men. This causes all kinds of assumptions about who belongs to or contributes to country music, leading those unaware of the inequities in the system to believe that white women, BIPOC or 2SLGBTQIA+ artists are not interested in country music.

While there is need for a very real conversation about the impact of AI on representation in the industry, this cannot be pinned to DSPs alone. Algorithms discriminate based on the data they lack — based on the absence of (in this case) songs by BIPOC artists, women, 2SLGBTQIA+ artists and disabled artists. And this points back to the systemic issues of the genre and industry culture that the DSP services: the industry has maintained these inequities, but then AI exacerbates them.

I would just briefly touch on the financial impact of this question. Much has been written about how little artists and songwriters are compensated in this digital age and it is a major cause for concern. Artists make less than a penny-per-stream in the current model, with $0.0033 split between song creators. But there is still a lot of misinformation floating around about royalties — or rather, the information is accurate, but the fingers are not always pointed in the right direction. From what I have learned Spotify (for example) pays 75% of their revenue to the content creators (songwriters, publishers, and recording owners). This is then divided rather inequitably: 15.1% of that 75% goes to songwriters and publishers and the largest portion (59.9%) goes to the sound recording owners. This conversation is constantly evolving, complicated, and often hard to keep tabs on, but it is worth noting that the current economic model doesn’t benefit everyone equally, and that it is increasingly more challenging for artists outside of that top one percent (especially songwriters) to make a living on their craft.

Q: Do you think that AI can benefit artists?

I think it’s probably clear that I have very real concerns about the impact of AI on the health and culture of the industry (as I say above, algorithms underpinning many digital service providers perpetuate discriminatory gender and racial issues in the industry), there are ways in which it can benefit artists.

Perhaps now, more than ever, artists can make and control their own music and image and that DSPs play a big role in creating this equal-opportunity-of-entry platform. Tech-savvy artists, artists that understand the algorithm, understand social media, and that have the time to devote to self-promotion have the potential here to build new career paths outside of the mainstream and (as a result) maintain greater control over their careers.

I am also excited about the fact that Spotify has the potential to reshape and perhaps even dismantle the oppressive genre categories that have controlled the boundaries of the industry for a century. These categories have historically been exclusionary to different groups based on racial classification rather than musical style. And it’s not just genre boundaries, but also linguistic and geographic boundaries. The industry has, for one hundred years, been controlled by a white Western (largely US) market, and Spotify is slowly tearing down those walls and creating a space where artists and fans can connect on a global level.