AI Can’t Make Music – The Atlantic
![AI Can’t Make Music – The Atlantic AI Can’t Make Music – The Atlantic](https://i1.wp.com/cdn.theatlantic.com/thumbor/SU1GSDh_eEUyB5QcVjBNhmF5Uf0=/0x43:2000x1085/1200x625/media/img/mt/2024/07/AI_music_2_horz/original.jpg?w=780&resize=780,470&ssl=1)
The first concert I bought tickets to after the pandemic subsided was a performance of the British singer-songwriter Birdy, held last April in Belgium. I’ve listened to Birdy more than to any other artist; her voice has pulled me through the hardest and happiest stretches of my life. I know every lyric to nearly every song in her discography, but that night Birdy’s voice had the same effect as the first time I’d listened to her, through beat-up headphones connected to an iPod over a decade ago—a physical shudder, as if a hand had reached across time and grazed me, somehow, just beneath the skin.
Countless people around the world have their own version of this ineffable connection, with Taylor Swift, perhaps, or the Beatles, Bob Marley, or Metallica. My feelings about Birdy’s music were powerful enough to propel me across the Atlantic, just as tens of thousands of people flocked to the Sphere to see Phish earlier this year, or some 400,000 went to Woodstock in 1969. And now tech companies are imagining a new way to cage this magic in silicon, disrupting not only the monetization and distribution of music, as they have before, but the very act of its creation.
Generative AI has been unleashed on the music industry. YouTube has launched multiple AI-generated music experiments, TikTok an AI-powered song-writing assistant, and Meta an AI audio tool. Several AI start-ups, most notably Suno and Udio, offer programs that promise to conjure a piece of music in response to any prompt: Type R&B ballad about heartbreak or lo-fi coffee-shop study tune into Suno’s or Udio’s AI, and it will spit back convincing, if somewhat uninspired, clips complete with lyrics and a synthetic voice. “We want more people to create music, and not just consume music,” David Ding, the CEO and a co-founder of Udio, told me. You may have already heard one of these synthetic tunes. Last year, an AI-generated “Drake” song went viral on Spotify, TikTok, and YouTube before being taken down; this spring, an AI-generated beat orbiting the Kendrick Lamar–Drake feud was streamed millions of times.
Twenty-five years after Napster, with all that’s come since then, musicians should be accustomed to technology reordering their livelihood. Many have expressed concern over the current moment, signing a letter in April warning that AI could “degrade the value of our work and prevent us from being fairly compensated for it.” (Stars including Katy Perry, Nicki Minaj, and Jon Bon Jovi were among the signatories.) In June, major record labels sued Suno and Udio, alleging that their AI products had been trained on copyrighted music without permission.
Some of these fears are misplaced. Anyone who expects that a program can create music and replace human artistry is wrong: I doubt that many people would line up for Lollapalooza to watch SZA type a prompt into a laptop, or to see a robot croon. Still, generative AI does pose a certain kind of threat to musicians—just as it does to visual artists and authors. What is becoming clear now is that the coming war is not really one between human and machine creativity; the two will forever be incommensurable. Rather, it is a struggle over how art and human labor are valued—and who has the power to make that appraisal.
“There’s a lot more to making a song than it sounding good,” Rodney Alejandro, a musician and the chair of the Berklee College of Music’s songwriting department, told me. Truly successful music, he said, depends on an artist’s particular voice and life experience, rooted in their body, coursing through the composition and performance, and reaching a community of listeners. While AI models are starting to replicate musical patterns, it is the breaking of rules that tends to produce era-defining songs. Algorithms “are great at fulfilling expectations but not good at subverting them, but that’s what often makes the best music,” Eric Drott, a music-theory professor at the University of Texas at Austin, told me. Even the promise of personalized music—a song about your breakup—negates the cultural valence of every heartbroken person crying to the same tune. As the musician and technologist Mat Dryhurst has put it, “Pop music is a promise that you aren’t listening alone.”
It might be more accurate to say that these programs make and arrange noise, but not music—closer to an electric guitar or Auto-Tune than a creative partner. Musicians have always experimented with technology, even algorithms. Beginning in the 1700s, classical composers, possibly even Mozart, created sets of musical bars that could be randomly combined into various compositions by rolling dice; two centuries later, John Cage used the I-Ching, an ancient Chinese text, to randomly compose songs. Computer-modulated “generative music” was popularized three decades ago by Brian Eno. Phonographs, turntables, and streaming have all transformed how music sounds, is made, and becomes popular. Visual artists have experimented with new technologies and automation for a similarly long time. Radio didn’t break music, and photography didn’t break painting. “From the perspective of art, [AI] is absolutely a boring question,” Amanda Wasielewski, an art-history professor at Uppsala University, in Sweden, told me. To say ChatGPT will force humans to invent new languages, or abandon language altogether, would be absurd. Audio-generation models pose no more of an existential challenge to the nature of music.
Within this framework, it’s easy to see how they might be useful tools. AI could help an artist who struggles with a certain instrument, isn’t good at mixing and mastering, or needs help revising a lyric. Andrew Sanchez, the COO and a co-founder of Udio, told me that artists use AI to both provide “the germ of an idea” and workshop their own musical ideas, “using the AI to kind of bring something new.” This is how Dryhurst and his collaborator and partner, Holly Herndon, perhaps the world’s foremost AI artists and musicians, seem to use the technology. They’ve been experimenting with AI in their joint work for nearly a decade, using custom and corporate models to explore voice clones and push the limits of AI-generated sounds and images: synthetic voices, ways to “spawn” works in the style of other willing artists, AI models that respond to user prompts in unsettling ways. AI provides the opportunity, Herndon told me, to generate “infinite media” from a seed idea.
But even as Herndon sees AI’s potential to transform the art and music ecosystem, “art is not just the media,” she said. “It’s the complex web of relationships and the discourse and the contexts that it’s made in.” Consider the prototypical example of visual art that observers scorn: a Jackson Pollock drip painting. I could do that, detractors say—but what’s relevant is that Pollock actually did. The enormous paintings are as much the tracks of Pollock’s dance around the canvas, laid across the floor as he worked, as they are delightful visual images. They matter as much because of the art world they emerged from and exist in as because of how they look.
What is actually terrifying and disruptive about AI technology has little to do with aesthetics or creativity. The issue is artists’ lives and livelihoods. “It’s actually about labor,” Nick Seaver, an anthropology professor at Tufts and the author of Computing Taste: Algorithms and the Makers of Music Recommendation, told me. “It’s not really about the nature of music.” There is “not a chance in hell” that the next Taylor Swift hit will be AI-generated, he said, but “it’s very plausible” that the next commercial jingle you hear will be.
The music industry has adapted to, and blossomed after, technological threats in the past. But there is “a lot of pain and a lot of dislocation and a lot of immiseration that happens along the way,” Drott told me. Musical recordings eventually allowed more people to access music and enabled new venues of creative expression, expanding the market of listeners and creating entirely new sorts of jobs for sound, recording, and mastering engineers. But before that could happen, Drott said, huge numbers of live performers lost their jobs in the early 20th century—recordings replaced ensembles in movie theaters and musicians in many nightclubs, for instance.
Sanchez, of Udio, told me that he believes generative AI will allow more people to create music, as amateurs and professionally. Even if that’s true, generative AI will also eat away at the work available to people who make music for strictly commercial and production purposes, whose customers may decide that aesthetic vision is secondary to cost—those who compose background music and clips for sample libraries, or recording engineers. At one point in our conversation, Udio’s Ding likened using music-generating AI to conducting an orchestra: The user envisions the whole piece, but the AI does every part autonomously. The metaphor is beautiful, offering the possibility of playing with complex musical concepts in the same way one might play with a simple chord progression or scale at a piano. It also implies that an entire orchestra is out of work.
What is different about AI is a matter of scale, not kind. Record labels are suing Udio and Suno not because they fear that the start-ups will fundamentally change music itself, but because they fear that the start-ups will change the speed at which music is made, without the permission of, or payments to, musicians whose oeuvres those tools depend on and the labels that own the legal rights to those catalogs. (Udio declined to comment on the litigation or say where its training data come from. Mikey Shulman, the CEO of Suno, told me in an emailed statement that his company’s product “is designed to generate completely new outputs, not to memorize and regurgitate pre-existing content.”) Humans already sample from and cover others’ work, and can get in trouble if they do so without sharing credit or royalties. What AI models are being accused of, although technologically different—reproducing likeness and style more than an exact song—is fundamentally a similar heist carried out at unprecedented speed and scale.
Herein lies the issue, really, with AI in any setting: The programs aren’t necessarily doing something no human can; they’re doing something no human can in such a short period of time. Sometimes that’s great, as when an AI model quickly solves a scientific challenge that would have taken a researcher years. Sometimes that’s terrifying, as when Suno or Udio appears capable of replacing entire production studios. Frequently, the dividing line is blurred—for an amateur musician to be able to generate a high-quality beat or for an independent graphic designer to take on more assignments seems great. But somewhere down the line, that means a producer or another designer didn’t get a contract. The key question AI raises is perhaps one of speed limits.
Also, unlike technological shifts in the past, the tremendous resources needed to create a cutting-edge AI model today mean the technology emerges from—and further entrenches—a handful of extremely well-resourced companies that are accountable to nobody but their investors. If AI replaces large numbers of working artists, that will be a triumph not of machines over human creativity but of oligopoly over civil society, and a failure of our laws and economy.
Or perhaps, amid a deluge of AI-generated jingles and podcast music and pop songs, we will all search even harder for the human. When I learned, a few months after the Belgium concert, that Birdy would be performing in New York City in the fall, I immediately bought tickets for myself and my sister. Birdy performed a version of one of her songs as a ballad, which built into a cascading sequence involving a looper pedal, that gave me goose bumps. The pedal layered, or “looped,” her voice over itself live—a piece of technology that, instead of replacing humanity, amplifies it.
Source link