Human beings may have sang before they spoke. Scientists from Charles Darwin onward have speculated that, for our early ancestors, music predated—and possibly formed the basis of—language. The “singing Neanderthals” theory is a reminder that humming and drumming are fundamental aspects of being human. Even babies have some musical instinct, as anyone who’s watched a toddler try to bang their tray to a beat knows.
This ought to be kept in mind when evaluating the rhetoric surrounding the topic of music made by artificial intelligence. This year, the technology created songs that amassed millions of listens and inspired major-label deals. The pro and anti sides have generally coalesced around two different arguments: one saying AI will leech humanity out of music (which is bad), and the other saying it will further democratize the art form (which is good). The truth is that AI is already doing something stranger. It’s opening a Pandora’s box that will test what we, as a society, really want from music.
The case against AI music feels, to many, intuitive. The model for the most popular platform, Suno, is trained on a huge body of historical recordings, from which it synthesizes plausible renditions of any genre or style the user asks for. This makes it, debatably, a plagiarism machine (though, as the company argued in its response to copyright-infringement lawsuits from major labels last year, “The outputs generated by Suno are new sounds”). The technology also seems to devalue the hard work, skill, and knowledge that flesh-and-blood musicians take pride in—and threaten the livelihoods of those musicians. Another problem: AI music tends to be, and I don’t know how else to put this, creepy. When I hear a voice from nowhere reciting auto-generated lyrics about love, sadness, and partying all night, I often can’t help but feel that life itself is being mocked.
Aversion to AI music is so widespread that corporate interests are now selling themselves as part of the resistance. iHeartRadio, the conglomerate that owns most of the commercial radio stations in the country as well as a popular podcast network, recently rolled out a new tagline: “Guaranteed Human.” Tom Poleman, its president, decreed that the company won’t employ AI personalities or play songs that have purely synthetic lead vocals. Principles may underlie this decision, but so does marketing. Announcing the policy, Poleman cited research showing that although 70 percent of consumers “say they use AI as a tool,” 90 percent “want their media to be from real humans.”
The AI companies have been refining a counterargument: Their technology actually empowers humanity. In November, a Suno employee named Rosie Nguyen posted on X that when she was a little girl, in 2006, she aspired to be a singer, but her parents were too poor to pay for instruments, lessons, or studio time. “A dream I had became just a memory, until now,” she wrote. Suno, which can turn a lyric or hummed melody into a fully written song in an instant, was “enabling music creation for everyone,” including kids like her.
Paired with a screenshot of an article about the company raising $250 million in funding and being valued at $2.5 billion, Nguyen’s story triggered outrage. Critics pointed out that she was young exactly at the time when free production software and distribution platforms enabled amateurs to make and distribute music in new ways. A generation of bedroom artists turned stars has shown that people with talent and determination will find a way to pursue their passions, whether or not their parents pay for music lessons. The eventual No. 1 hitmaker Steve Lacy recorded some early songs on his iPhone; Justin Bieber built an audience on YouTube.
But Nguyen wasn’t totally wrong. AI does make the creation of professional-sounding recordings more accessible—including to people with no demonstrated musical skills. Take Xania Monet, an AI “singer” whose creator was reportedly offered a $3 million record contract after its songs found streaming success. Monet is the alias of Telisha “Nikki” Jones, a 31-year-old Mississippi entrepreneur who used Suno to convert autobiographical poetry into R&B. The creator of Bleeding Verse, an AI “band” that has drawn ire for outstreaming established emo-metal acts, told Consequence that he’s a former concrete-company supervisor who came across Suno through a Facebook ad.
[Read: The people outsourcing their thinking to AI]
These examples raise all sorts of questions about what it really means to create music. If a human types a keyword that generates a song, how much credit should the human get? What if the human plays a guitar riff, asks the software to turn that riff into a song, and then keeps using Suno to tweak and retweak the output? Nguyen replied to her critics by saying that the “misconception here is that ‘there’s no effort put in by a human,’ when so many musicians I know using Suno are pouring hours and hours into music production and creation.”
In practice, however, AI is helping even established musicians work less, or at least to work faster. The Verge reported this month that the technology has become ubiquitous in the country-music world, where Nashville pros are using it to flesh out demos and write melodies. The producer Jacob Durrett said in that story that Suno affords him “a productivity boost more than a creative boost”; the publisher Eric Olson said that it allows him to spend more time with his kids. Similar practices are happening in other genres. The Recording Academy’s CEO, Harvey Mason Jr., recently said many of the producers and songwriters he knows are using AI in some capacity.
While the technology quietly reshapes the industry, the first-order effect of AI’s ease of use is simply the existence of more music—a lot more. Suno users generate 7 million new tracks a day, which every two weeks nets out to about as many songs as exist on Spotify. Most of those tracks are likely never heard by anyone. Still, the streaming service Deezer has disclosed that nearly a third of the music uploaded daily to its platform is AI-generated. Spotify has said that it will crack down on obvious slop and spam—but definitively detecting when AI has been used to make a song is hard, and only going to get harder.
I admit to feeling some sadistic curiosity about what a full fire hose of AI music, intersecting with streaming algorithms tuned to deliver to users exactly what they want to hear, might reveal about listening desires. Historically, popular art tends to progress by the logic of MAYA: “Most Advanced Yet Acceptable.” Hit songs are usually neither wholly original nor wholly derivative but rather some delectable combination of the two. At a glance, AI may seem incapable of newness because it’s trained to replicate music’s past; many of the breakout AI songs to date sound incredibly familiar. But other examples show that the technology can, perhaps accidentally, carve new pathways.
Those are not always good pathways—but they are ear-teasingly novel, the results of choices a person probably wouldn’t or couldn’t make on their own (again: creepy). In Bleeding Verse’s viral track “Only When It’s You,” the vocals have a scuzzy vibrato, almost like someone’s blowing bubbles into the mic; the post-chorus jumps to a whistle tone that’s less Mariah Carey and more steaming teapot. An even more harrowing example: Spalexma’s “We Are Charlie Kirk,” which has been sarcastically memed into infinity. The nu-metal tribute to the late right-wing activist is deeply catchy and sickeningly soppy, like Creed but a lot worse. If a real person had attempted to record it, I’m convinced that he would have fainted from embarrassment. The song is an example of the entropic abominations that might catch on should AI music continue unimpeded, at scale: Spalexma, an entity whose authors aren’t known, published about 280 songs in less than a year.
However, in a bit of a twist, the full slopocalypse may be held off for a bit. Recently, a number of lawsuits by major labels against AI companies have ended in settlements that dictate significant reforms. One service, Udio, is now obligated to become, according to Billboard, a “walled garden” whose contents cannot be widely distributed. A deal reached in November between Suno and Warner Music Group mandates that the platform retire its current model and replace it with one trained solely on licensed data. Users will be able to remix work from artists who have opted to participate in the system, and they will now need to pay a fee to download their creations.
These developments will not end AI music, but they may slightly rein in the creative and copyright free-for-all, at least temporarily. The music industry—for obvious reasons—wants to control AI tools, making it harder, or at least costlier, for amateurs to jump its gates. But the generative possibilities and consumer demand demonstrated in the brief history of AI music to date aren’t going to be forgotten. A cultural countermovement that emphasizes flesh-and-blood talent and craft seems inevitable—but so is a future for recorded music that grows more crowded and chaotic with each day. In that way, too, AI music is accelerating something very human: competition.
The post AI Is Testing What Society Wants From Music appeared first on The Atlantic.
