AI Pop Stars or Just Sparks? The Future of AI-Generated Musicians
PART I: DESCRIPTION OF THE CURRENT STATE OF THE MUSIC INDUSTRY
Few media industries are as all-consuming as the music industry. Considering that the vast majority of Americans listen to music on a regular basis and many of the country’s most popular celebrities are musicians, interacting with the industry is all but impossible to avoid. This shows in its revenue; according to the Recording Industry Association of America, which is the United States’ foremost authority on national music revenue, the industry made $15 billion in 2021 from recorded music alone. Of this, 83% is attributable to streaming services, followed by another 11% in physical album sales such as vinyl records and CDs, and 4% in digital downloads. Although recorded music accounts for most of the industry’s annual revenue, it is only part of the process. The industry also relies on touring, merchandise sales, and brand deals between artists and non-music companies.
In analyzing the current state of the music industry, we must examine the process by which artists create their products. Since the 1950s and 1960s, the music industry has recruited talent by using A&R representatives. These representatives work for record labels, and their job is to find upcoming musical talent through a variety of means and pitch them to the record label. Once a musician signs a contract with a label, they are committed to creating a certain number of albums, performing at a certain number of concerts and tours, and releasing their image and likeness for merchandising. Record labels will take a large portion of the money their artists make.
Over time, the means by which A&R representatives have found this talent has changed significantly. Prior to the advent of the internet, A&R representatives would go to house shows, bars, and open mic nights to find talent. They would also seek out musicians who would send self-made recordings directly to them. Producing a recording like this was a challenge, however, as recording time in a studio is expensive and one would need access to instruments and professional-grade equipment. Because of the Supertrend of technological development, especially as it pertains to the internet and social media, artists can now create music from their own bedrooms. Even by using free programs such as Apple’s GarageBand, aspiring musicians can create these recordings completely independently and post them on social media for A&R representatives to find. This has saturated the market with potential talent and shifted the way musicians are signed to record labels.
According to this process, we can analyze the industry’s landscape using Larry Kramer’s concept of the C-Scape like this: Consumers stream and purchase music within the genres that they enjoy; typically, people stick to a handful of specific niches. Based on demand from consumers, record labels will sign musicians who align with the genres that fans are seeking. This is the element of curation. Convergence occurs when the labels integrate new technology and developments into their curation processes. These musicians produce content that then returns to the consumers, who demand more, and the cycle continues.
The convergence step has recently begun to integrate a new technology: Artificial Intelligence, or AI, into the A&R recruitment process. A handful of musicians have begun to use generative AI– a classification of AI programs that compiles existing material and learns from it to create a new, unique product– in this case, original music. However, these programs are difficult to use because they are new and complex, so their use has been limited in scope thus far. Another, even more complex way that AI could be used, though, is to create a completely unique AI musician. As of 2022, there have been a few musicians who are effectively virtual characters that release music generated by an AI. One such musician is Holly Herndon, a human who created an AI clone of herself to augment her work and performances. If AI is combined into the A&R process, it could remove a large part of the human element of music, as labels could simply create their own musician without having to sign a human being who might become sick, break their contract, or be subject to any number of real-world variables. AI musicians can generate perfectly-on-trend music on demand without getting tired and without the ability to have creative differences with their label, which might help the label increase revenue.
There are quite a number of factors that will keep humans involved in creating and performing music for the foreseeable future, though. Record labels have already signed hundreds to thousands of human beings who have legal obligations to complete a certain number of albums, as previously mentioned. In addition, human beings want to connect with other human beings, and there is already a litany of human celebrities with massive fanbases that we cannot expect to disappear in favor of AI celebrities. Simply, the music industry has been powered by live humans for the past 80 years, so it is extremely unlikely that it will turn on a dime. Despite this, AI musicians have begun to enter the market, and will likely remain. In the next section, we will explore how the trend of AI-generated musicians might augment the existing music industry.
PART II: PREDICTIONS ON AI-GENERATED MUSICIANS AND THE MUSIC INDUSTRY
In the coming years, I predict we will see significant growth in the number of AI musicians in the market. However, they will likely not remain in the mainstream and will most likely fall into a dedicated niche fandom that will still be lucrative, but not as lucrative as Top 100 artists, who will continue to be human.
We can return to the “consumer” element of C-Scape to help illuminate this prediction, as the music industry is primarily driven by genres and niches. Some consumers are already interested in the concept of AI-generated celebrities, as proven by AI social media influencers and VTubers, which are animated or AI YouTube personalities. There is already a market for content like this, although it is relatively small to date, and it is not a large logical leap to assume that someone who enjoys an AI influencer would also enjoy AI-generated music. There is also an existing fandom based upon AI-generated and AI-augmented musicians that exist today; Holly Herndon has about 33,000 monthly listeners on Spotify. Therefore, we will see some consumers, likely those who already have an interest in AI and similar technologies, gravitate toward these musicians and this music. The content, of course, will be different from the content we are used to seeing from human musicians. Among the AI-generated or AI-augmented musicians that exist today, the vast majority make music that falls into the hip-hop, rap, and electronic genres. These genres all have a history of readily adopting new technologies, so their adoption of AI makes sense. However, genres that rely on more old-fashioned methods of musicianship such as country and rock and roll may be slower to adopt AI technology because of the value those genres typically place on legacy and tradition. In addition, we must consider the demography of these genres, as younger people are more likely to be willing to adopt new technology and are more likely to listen to genres like pop and rap. As a result, the content we can expect to see from AI-generated musicians in the next 5 to 10 years will presumably remain in younger, more technologically-savvy genres such as rap, dance, and pop.
This demography will in turn influence the way the AI musicians’ representative characters look. Each AI-generated musician will likely have a human-like form, at least to begin with, as humans tend to want to attach a face to a voice or a piece of art. If AI musicians do not have an animated physical form, it may be challenging for them to build a following for this reason. AI musicians will, then, likely appear as young people between the ages of 18 and 25 who wear hip and on-trend clothing, makeup, and hairstyles, but who have a distinguishably non-human feature about them. Because of the concept of the “uncanny valley” and the human brain’s powerful ability to identify things that are not human but appear to be, an AI musician who looks too human-like may be considered creepy and could drive away a potential fanbase. It would be more effective to make the musician look like a video game character or an animated person in order to prevent this.
In the best possible scenario for the future of AI-generated artists, at least one musician’s music enters the Billboard Top 100, giving it routine radio play and popularizing the concept of AI musicians. This would allow the musician and the team behind it to continue to produce music and would likely mean they become signed to a record label if they are not already. Another way this popularity could occur is if an AI-generated musician’s song goes viral on a social media platform like TikTok– there have been many record deals secured this way in the past five years. In this case, regardless of the method by which the musician enters the mainstream, the musician would set a precedent that AI-generated music can be valuable and worthy of listenership. At this point, multiple record labels would begin to sign other AI-generated musicians in the hopes of competing with the original musician. In this ideal scenario, these musicians would remain popular and generate large sums of revenue for their companies through consistently charting hits, and possibly even Grammy awards and other major awards.
In the most likely scenario, AI musicians will not achieve lasting mainstream success in the way described above. If they do achieve mainstream success at all, it will likely be fleeting and last no longer than 5 years. This is simply because it is extremely challenging to remain in the public eye for that long, especially considering the speed at which the trend cycle turns over in pop culture. Unless there is a major change in this trend cycle, AI musicians will likely fall into a niche genre with a relatively small but dedicated fanbase. This might be more lucrative for them in the long run, as demography and the feeling of social belonging from consumers are critical to a successful business model. Though they likely will not have the popularity of human musicians like Beyonce or Taylor Swift, it is important to note that very few humans achieve that level of success either. In this scenario, record labels will continue to sign and produce AI musicians, but at a lower rate because there is less demand from consumers. Labels may still create dedicated AI departments and acquire generative AI technologies to supplement this, but these would be to appeal to the AI musicians’ fandom rather than the mainstream. Again, this can still be lucrative if done properly, but it is crucial that the musician and label study the fanbase to understand what they are seeking and how best to appeal to that.
In the worst-case scenario, AI musicians’ music will never catch on in the mainstream, nor will it develop a strong, dedicated fanbase. Reasons for this may be controversies regarding plagiarism, moral or ethical questions about offensive material, or the uncanny valley effect that makes prospective listeners uncomfortable. These issues are already circulating in AI-generated art spaces; many visual artists dislike generative AI programs such as DALL-E because they feel it is plagiarizing their work and threatening their livelihoods. If human musicians begin to feel the same way, they may be able to rally their fans to boycott AI-generated music, at which point it would be difficult for AI musicians to be successful. Without a genre or fandom, there is no way for any musician, human or AI, to generate revenue, and so the musician will be dropped from their record contract or never signed, to begin with. This would relegate AI in music to be augmentative rather than unique to an artist or character.
PART III: PRESCRIPTION FOR THE FUTURE OF AI-GENERATED MUSICIANS
Record labels who aim to capitalize on this trend should, first and foremost, hire a team of staff who understands how AI operates. This team will be primarily responsible for programming the AI artists of the future and optimizing their output. Grouped in this account should be the artists’ PR team and representatives so that they can all communicate effectively about the product.
In the best and most likely case scenarios, it might also be a rewarding return on investment to acquire an existing music-generative AI program so that the record label will have the rights to not only the music the artist produces but also the technology behind it. If a label were to have this technology in-house, it could attract more prospective AI artists to sign with that label. This would also allow for the creation of new AI artists in the future who would already be signed to that label, with relatively little external legwork compared to traditional A&R processes.
As previously mentioned, there are a number of potential blocking forces that record labels should be aware of. First, labels must consider that human musicians will not disappear and may even come to resent AI musicians as vapid knockoffs of their work. Likewise, A&R representatives could feel shafted by this technology as well, considering their job is to find human talent. This may be problematic for the success of these AI musicians if the existing human celebrities conduct an effective boycott of the technology. In addition, there are many ethical and legal concerns with the implementation of this technology. This is already true of existing AI musicians; AI rapper FKN MEKA was recently dropped by his record label because he emulated racist stereotypes about Black people despite not being human (notably, the programmers behind the character are white). Because these AI artists are not humans, they themselves cannot distinguish between offensive and non-offensive material, so it will be the responsibility of the artist’s management to do this on their behalf.
Furthermore, the question of how the AI learns to create music and whether or not this can constitute plagiarism may be a cause for concern. The AI might end up plagiarizing an existing song by an artist within the same genre as the AI, which could lead to an expensive intellectual property lawsuit. In order to avoid this, the artist’s dedicated team should screen each song prior to release and, if possible, even create a cross-checking program that can identify whether the song in question is similar to another work. Another tool could be a deep learning algorithm, which is more complex and creative than a combinatory type of algorithm. If the artist and their team utilized a deep learning program, they are less likely to create an existing chord progression or lyric, which are the primary causes of a plagiarism suit.
The largest and most likely threat to the success of AI musicians is human musicians who dislike them. Therefore, it would be in a label’s best interest to educate its existing artist base and A&R representatives on how generative AIs like this work. If humans understand the process, they are less likely to feel that they are being replaced by machines and are therefore less likely to publicly speak out against AI-generated music, which would leave fans more open to the possibility of enjoying the music and supporting AI artists. Of course, it is still possible that, even with the understanding of what AI-generated music is and how it operates, musicians may still call for a boycott. Considering the art industry is seeing a similar discussion right now, it is actually quite likely that some musicians will be disgruntled; for the record labels, it is just a matter of mitigating that response.
If this technology proves successful, a new record label that exclusively signs AI musicians may emerge. It would be particularly lucrative for a label like this to own its own musical AI technology because it would be less complicated and possibly less expensive to produce that content in-house than outsourcing it to another company. A label like this would begin as an indie label that would effectively act as a start-up, but it could gain a loyal enough fanbase to propel it to prominence, as mentioned in Part II. With this, we would also likely see established record labels create AI departments to compete; in order to lead this trend, it would be smart for record labels to begin forming these departments as soon as possible. The departments can and should remain small in the event that AI music does not win favor with the public, but having members of the team who are experienced in this technology will be useful to them in the long term because human musicians have also begun utilizing AI technology in their work recently, on top of the emergent trend of AI celebrities.
Regardless of whether AI-generated musicians become popular or carve out space to create their own genre, record labels should aim to gain knowledge of generative and deep-learning AI programs. These programs are still relatively new, and they will continue to develop at a rapid pace. If the music industry wants to influence other industries, it should become intimately familiar with this technology and continue to find ways to implement it in its products. If it does this properly, it can increase its revenue and position itself on the cutting edge of modern technology.
