This Is What You'll Pay For
The Menace of Generative Music
About a year ago, an artist called "Jaime Brooks" uploaded music to Spotify for the first time. Their debut release was a forty-six second instrumental composition called "Martin Luther King Conservatory." It would be misleading to describe the piece as "music." In practical terms, it sounds like short clips of many different violin performances chopped up and re-arranged at random. There's no arc, no sense of movement, and no tension or release. Listening to it feels like waiting for a video to buffer, or trying to read a glitchy social media feed that keeps yanking posts away the second your eyes begin to settle on them. You start to crave something more focused and intentional, like the sound of a CD skipping or a kitten walking haphazardly across a set of piano keys.
During Jaime's first year on the platform, the project managed to attract a respectable following of nearly seven hundred monthly listeners. Artist pages on Spotify feature a section called "Discovered On," which can help give us some idea of how Jaime was able to accomplish this with such functionally unlistenable output. When you reach the end of a playlist or album on Spotify, the recommendation algorithm kicks in and begins playing songs it thinks you'll tolerate based on what you were just listening to. An artist page's "Discovered On" section shows us which albums or playlists users were listening to when that particular artist's music was served up to them. When I first discovered Jaime's artist page, the “Discovered On” box informed me that "Martin Luther King Conservatory" was recommended to listeners of classical music albums. Specifically, classical music albums intended to be listened to by babies.
Please consider subscribing if you are able.
It's probably just a coincidence that "Jaime Brooks" has the same name I do. The project's artwork, song titles, and audio all have the feel of "generative" assets created with machine learning tools like DALL-E and ChatGPT. It's commonplace for these products to be described as "artificial intelligence," which I think is stupid. Obfuscating the real-world shortcomings of new products by framing them as science fiction made real has long been standard operating procedure for the tech industry. We have nothing to gain by going along with it.
I would describe ChatGPT as a "generator." It generates content. Have you tried to look up a recipe on contemporary search engines at any point in the last several years? If so, you probably had to scroll through a longform essay's worth of filler text in order to reach the information you actually wanted. Traditional recipes are too short and useful to generate much ad revenue, so it's actually the filler text that makes money whenever anyone visits the page. That's content. Long before generators are ever used to make "art," they will be used to make digital obstacle courses that exist for no other purpose than to trap you in a vulnerable position where your time can be converted into income for digital landlords.
The case of "Jaime Brooks" perfectly illustrates how these scams will manifest in the world of music. By generating audio that is harmonically similar to classical music, "Jaime" was able to trick Spotify into recommending it to classical listeners. By deploying familiar sounds in a randomized, nonsensical fashion, the resulting compositions were able to bypass copyright filters designed to block infringement of existing works. By targeting a style of music that often appears in sleep-oriented playlists, "Jaime" gained access to the least discerning users on the entire platform. Unconscious adults are very unlikely to notice they're not listening to Mozart anymore and get up to hit fast forward, especially if the interruptions are brief and unpredictable. Babies don’t even have the option! This is how generators are going to change the internet in the coming years: by infesting the darkest, most neglected corners of our collective online experience with the digital equivalent of roaches, bed bugs, fruit flies, and mold.
When I was a teenager, I routinely paid more for a single album on CD than a whole month of unlimited access to the entire history of major label recorded music costs today, even after adjusting for inflation. Spotify's all-access subscription product is an unbeatable value for money compared to the rampant price-gouging of the CD era. It's kind of an anachronism; one of the few tech products on the market today that still feels like a welcome step forward, the way Google Reader or Vine once did. A relic of Obama-era techno-optimism, of a culture that was still willing to believe in half-baked scams like the "hyperloop" concept that the CEO of a car company cooked up to divert resources away from high-speed rail. To subscribers, Spotify continues to feel right no matter how wrong the rest of the internet becomes.
The reality is that Spotify is unprofitable. It always has been. Their single biggest operating expense is the cost of licensing music from the major labels, which continues to be the platform's biggest draw. Spotify is the market leader in streaming with about one hundred and ninety-five million paying subscribers globally, and they lose money every time one of their users hits play on a Taylor Swift album. They can't make money giving you what you want, because what you want is too expensive. Serving up music to customers is a tedious, money-losing business that Spotify only ever bothered with in the hope that it would teach them how to get us all hooked on something cheaper. The trajectory of the platform's shift from user-created playlists to in-house editorial playlists to personalized, algorithmic "Discover" playlists makes this clear. The business model only works if Spotify's users can be persuaded to see listening as a passive, solitary activity rather than an active, social one.
Podcasts were supposed to help them do this. Spotify used to address investor concerns about the viability of their business model by comparing themselves to Netflix, a company that was successfully able to woo subscribers away from expensive licensed fare with their own original first-party content. For this to work, podcasts would have had to overshadow major label music on Spotify the way "House of Cards" and "Stranger Things" came to dominate users' consumption habits on Netflix. That did not happen. All these years and acquisitions later, podcasts currently account for only seven percent of total listening hours on the platform. Spotify doesn't compare themselves to Netflix anymore. Instead, they've taken to invoking YouTube, a business that revolves around selling advertising.
When I came across Jaime Brooks on Spotify, I worried about how the platform's recommendation "algorithm" would hold up as generators become more popular and sophisticated. If it already struggles to understand that a forty-six second track by a brand new artist doesn't belong anywhere near a "Mozart For Babies" compilation, I wondered, how will it contend with what's coming? This was naïve of me. I was still thinking in terms of Netflix's business model, where a flood of low-quality generative assets clogging up recommendation queues would likely trigger a wave of cancelled subscriptions. YouTube's business model, on the other hand, cultivates lower expectations. While your average Spotify subscriber probably spends most of their time on the platform listening to the same music over and over again, YouTube is full of manipulative, low-quality media that succeeds in making money whenever it can successfully bait a user into sitting through a few minutes of it. Jaime would fit in well there.
Technically, the American record business peaked twenty-three years ago. The success of the all-access subscription streaming product certainly reversed the trend, but with caveats. Giving customers unfettered access to the major labels' catalog for a flat monthly fee was a great way of luring them away from the flourishing digital piracy ecosystem, but the egalitarian nature of the business model means everyone pays the same price regardless of their means or level of consumption. That means there are no "whales" on Spotify. There is no way for labels to extort more money from even the platform's most active and engaged users. Accordingly, current industry revenues remain thirty-seven percent below their all-time highs at the end of the nineties. Recorded music is simply worth less now than it was then.
You typically hear about this issue from musicians, who have organized to protest against low streaming payouts as part of initiatives like the Union of Musicians and Allied Workers' "Justice At Spotify" campaign. These efforts have largely been met with indifference, both from Spotify's users and from the company itself. A more equitable system of remuneration for artists would likely mean paying even more to license music than Spotify already does, resulting in an even steeper uphill climb to profitability. Subscribers, who mostly seem to use the service to listen to old music anyway, have been content to accept the radical changes that the streaming model has imposed on the contemporary musical landscape for fear that the cost of a fairer system would be passed on to the consumer.
Lately, though, signs of strain have become more widespread. The vinyl market was the one successful holdout against the collapse of physical music formats. Prestigious indie labels staffed by true believers in the format drove a postmillenial revival of consumer interest in vinyl that saw records and USB turntables being stocked nationwide by retailers like Whole Foods and Urban Outfitters. Their enthusiasm for packaging and sound quality became globally contagious against the tumultuous backdrop of the collapsing CD market. Vinyl sales saw explosive growth year after year, until the major labels could not help but re-enter the frame. Record stores were soon flooded with overpriced, arbitrarily chosen reissues from deep within the majors’ sprawling catalogs. Last year, the vinyl market grew only four percent after seeing double-digit growth during the preceding years. If Taylor Swift hadn't made vinyl the centerpiece of the rollout for her Midnights album, the vinyl market may actually have shrunk for the first time in twenty years. Analysts and record store owners are now predicting a market crash.
Digital revenue sources like streaming now account for ninety percent of the total market for recorded music in America, but even that prize is now threatened. Low-cost distributors like DistroKid allow independent uploaders to compete directly with the major labels for market share, and they can do this for much less than it would have cost to get physical products on store shelves during the CD era. Market share is not merely an academic concern for the major labels, either. Every premium Spotify subscriber pays the same amount of money regardless of how much media they consume, which means that the majors' share of the overall streams determines how much they get paid. Sony Music chairman Rob Stringer recently complained that "if there are eighty thousand tracks a day being uploaded on major DSPs, then [major label] market share is going to to be diluted by default" during a financial presentation last June.
Music Business Weekly founder Tim Ingham was much more blunt in his assessment. "To put it in much simpler terms," he wrote. "unpopular artists on Spotify, in the grand scheme of things, are commercially irrelevant as individuals. But combined, they are a powerful force." In September, Universal Music Group chairman Lucian Grainge proclaimed that the number of new tracks being submitted to digital service providers every day had increased to one hundred thousand. The total share of streams claimed by the major labels and the Merlin Network is down by more than ten percent since 2017. In an effort to reverse the decline, the majors drastically ramped up signings during the pandemic. A recent investigation by Billboard found that the initiative had failed to create a corresponding increase in the number of new stars, and may in fact have harmed the prospects of potentially viable artists by forcing them on overworked product teams.
Uploaders with less than a thousand monthly listeners account for ninety-three percent of the total number of artists on Spotify. This figure includes hobbyists and daydreamers laboring in obscurity, yes, but also generative scam projects like "Jaime Brooks." The tactic of infiltrating mood and sleep-based playlists to generate royalties is so well-established by now that evidence suggests even Spotify themselves have utilized it. Every such loophole that entrepreneurial scammers identify from here on out will be exploited exponentially faster and more cheaply than previously available tools have ever allowed for. One hundred thousand new uploads a day will seem like a quaint, manageable figure in retrospect.
Generative narrators will spit out entire audiobooks in less time than it would take a human being to record a single podcast episode. Generative mood music will proliferate faster than even the most prolific and unfiltered ambient musician could ever keep up with. If commercially irrelevant individual artists are already such a "powerful force" in aggregate, the impact of generators on streaming will be significant regardless of whether or not generative assets ever become part of the tiny percentage of tracks on the platform that subscribers actually enjoy listening to. There will simply be too many of them to ignore.
The unavoidable outcome of all this is the further devaluation of recorded music. The major labels and the streaming services may have conflicting goals, but they are collaborators, not enemies. They have already conspired to devalue recorded music once before, when the labels embraced the all-access subscription model as a means of protecting their intellectual property from the existential threat of digital piracy. The threat posed by generators will inspire a similar gambit. Industry stakeholders like Ingham and Grainge are already floating the idea of a two-tiered royalty system that would value major label music differently than the unpopular "flotsam and jetsam" that has come to dominate streaming platforms. This would allow the majors to keep making the same amount of money they’re currently getting from streaming platforms even after the coming tide of low-quality generative assets makes it impossible for traditional labels to maintain their current share of the market. The labels will make more because the rest of us will be making less by default.
What would it even mean for recorded music to be worth less than it is now? How will our relationship with music change once the platforms we all use to engage with it start putting every single new work by an independent artist in the same low-royalty garbage bracket as the billions of low-effort generative assets that are about to overwhelm the entire internet? How much more of the shrinking territory between passive consumers and the tiny minority of successful major label signings will be obliterated by this change? How many new “recordings” can be brought into existence before most people start to see the medium itself as a landfill rather than a library? What would it be like to live through the decline of an institution as ubiquitous as the record industry? Are we about to find out?
In 1906, the composer John Philip Sousa published an op-ed called "The Menace of Mechanical Music" in Appleton's magazine. He composed music for marching bands, which provided one of the most electrifying musical experiences you could have in the days before amplification. Even today, I think HBCU-style marching bands are the highest-level musical performers in the entire country. Though Sousa hits like "The Stars And Stripes Forever" have been rinsed too thoroughly to retain any of the power they might have had during his day, HBCU-style bands often do thrilling arrangements of contemporary songs that can do a lot to illustrate the appeal of this style of music. The influence flows in the other direction, too: Atlanta R&B mastermind and frequent Beyoncé collaborator Terius "The-Dream" Nash recently cited HBCU bands like Florida A&M's famous Marching 100 as his childhood inspiration for becoming a musician. "We studied those VCR tapes," he said. "and in third grade, I started playing the trumpet."
Anyone alive at the turn of the century wouldn't have needed to have the appeal of marching bands explained to them. Playing instruments was a much bigger part of people's everyday lives back then, and recorded music was still a niche, low-quality product. Magnetic recording tape wasn't available, so recording sessions were organized around huge mechanical devices that engraved the music directly onto a disc as it was being performed. Copyright was not a factor at that time, so royalties didn't exist. The practical shortcomings of the tech were innumerable, and Sousa addressed many of them in his piece, but it was the broader implications of mass adoption that troubled him most deeply.
Before there was a record industry, music was something you experienced in the moment. No one ever showed up at a concert and felt cheated because the songs didn't sound enough like the recorded versions. No one ever worried about whether or not a particular performance of a song was "definitive," because they had no way of ever hearing it again afterwards. If you liked a song, that meant you were just happy to hear it generally. The notion of paying thousands of dollars for the privilege of piling into a stadium to hear one particular person perform their specific version of the song would have seemed absurd. If you could sing or play an instrument, your friends and loved ones' favorite version of a particular song might have been yours.
"There are more pianos, violins, guitars, mandolins, and banjos amongst the working classes of America than in all the world," Sousa wrote. "and the presence of these instruments in the homes has given employment to enormous numbers of teachers who have patiently taught the children and inoculated a love for music throughout the various communities." He believed that everything truly great about American music was only possible through the enthusiasm of amateurs. If musical instruments were replaced in the home by "music-reproducing machines" like record players, he feared that the "tide of amateurism" that drove popular enthusiasm for playing and understanding music would "recede," leaving only "the mechanical device and the professional executant." Essentially, he was terrified of the world we are currently living in. "When a mother can turn on the phonograph with the same ease that she applies to the electric light, will she croon her baby to slumber with sweet lullabys," he wrote. "or will the infant be put to sleep by machinery?"
Today, the world’s most powerful corporations are aggressively investing in the idea that generators are going to fundamentally change our collective relationship with media. Sousa’s op-ed is a thoroughly satisfying exercise in shit-slinging à la Steve Albini’s “The Problem With Music,” but what really makes it worth reading today is the perspective it gives us on a similar moment in history, when the disruptive new technology of records and phonographs was about to change music forever. Sousa’s perspective is fascinating because we know he was gearing up for a fight he couldn’t win. He was such an entrenched member of the American cultural establishment that his marches are still taught in public schools over a century later, but every single change he advocated against in his manifesto happened anyway. There was nothing he could do to stop any of it. We should study him because his experience can teach us about the scale of what we now stand to lose.
When I was a kid, my Mom’s love for music was fanatical. I don’t remember her ever looking happier than she did while she was dancing around the house singing along with George Strait hits like “Adalida” and “Check Yes or No” on her stereo. Whenever she did this, she would apologize to whoever was around for the noise she was making. Often, she would exaggerate the idiosyncracies of her voice or sing deliberately off-key when she noticed me listening as a type of self-deprecation. Despite the obvious joy she derived from singing, she acted like she couldn’t imagine any circumstance under which anyone would prefer to hear the sound of her voice over a recording of a professional’s.
Records seemed like the only force on Earth that had the power to lighten the burden of being a working single mother supporting two kids on her own. Those brief pockets of musical release were an oasis apart from the hours I spent alone while she was working three jobs to support us, and from the stress of helping her comb through the tips she brought home hoping to find enough quarters to make rent that month. Accordingly, I started treating records the way other people treat religious texts and sacraments. I studied them extensively. I used all the money I made at my first terrible minimum-wage job on CDs. I read about people like Trent Reznor that could make entire albums by themselves, and then I used pirated production software to teach myself to do the same. It seemed important to.
By no stretch of the imagination could I ever be accurately described as a skeptic of the medium of recorded music. Records have been good to me. I credit other people’s records with saving my life, and my own with introducing me to most of my friends. Records are the means through which I reached out into the void and asked if there was anyone out there who understood what I was going through. The world answered by grabbing my hand and pulling me up out of the hole I was in. I don’t know what kind of person I would be without records.
Regardless, the era of recorded music started fairly recently. Many of us may have grown up around elders who could remember the way things were before. When I look at the moment we’re in now, and consider the possibility that we may be about to witness that kind of paradigm shift for ourselves, my feelings are complicated. I feel humbled by the opportunities I was afforded that will never be available to my younger peers. I feel incredible grattitude for the places I’ve been, the people I’ve met, and the happy memories I’ve forged that wouldn’t have been possible at any other moment in time. At the same time, I just keep thinking about those days in the kitchen with my Mom. I always just took her apologies at face value, and quietly revered the recordings she was singing over the way I thought she would have wanted me to. Looking back, I notice an emptiness in those memories. A void left by my silence. As much good as records have done for me, I’m not sure I wouldn’t trade all of it for the chance to go back and try singing along with her instead.
Thanks for reading.
Deeply provocative post. I was struck by the thought that Spotify could adopt an Amazon model fairly easily — viz., use their own analytics to track which songs by real artists are at any moment attracting streams and have generators produce simulacrums that enter the algorithms or playlists with a little extra heat. Since these generated songs would be "owned" by Spotify, the company wouldn't have to pay out for those streams that would, invariably, eat away at listens that might otherwise accrue to real musicians.
This move would be utterly evil, but they have little incentive NOT to be evil. The major labels might even be on board with such a practice, as it might increase the perceived value of their catalogs full of music from real human beings.