Who’s Afraid of A.I. Music?
Earlier this summer, a small night club in Passaic, New Jersey, hosted a brief performance by one of the year’s most talked-about musicians. A post on Instagram had promised attendees “an unforgettable night,” although not necessarily a long one: the only assurance was that the Los Angeles rapper known as Fenix Flexin would be “performing his viral hit ‘Rubberz’ live!” Until recently, Fenix was known for high-spirited and slightly mush-mouthed hip-hop tracks, but “Rubberz,” which was released in June, marked a dramatic change: it was sad but sprightly synth pop, like an unheard outtake from the nineteen-eighties; the vocals were crisp, with a hint of vibrato and a faintly British accent. Listeners seemed to delight in the incongruity—a semipopular rapper suddenly sounded nothing like himself.
Plenty of people thought they knew why. They guessed that the song had been created with artificial intelligence, which makes it trivially easy to generate a realistic song using nothing more than a brief description. Perhaps the voice on the track didn’t belong to Fenix, or, in fact, to anyone at all. The rapper appeared to deny this. When the YouTube journalist Andrew Callaghan asked Fenix whether he wanted to clear up misconceptions about “Rubberz,” he declined. “Oh, like the A.I. shit?” he said. “Nah. Shit is recorded like any other song.” He also sang some of the words for Callaghan, although neither the sound of his voice nor the lyrics he delivered were an exact match. “The watch is getting quicker,” he crooned; on the recording, it’s “The watch is moving quicker.”
On this night in New Jersey, no one seemed inclined to conduct a forensic investigation into the authenticity of the song, which Fenix performed twice, singing along with a prerecorded version of it and reproducing the slightly weird pronunciation: “Money talk is much louder now,” he sang, stressing the second syllable in the line. “Hatin’ that I’m able,” he sang, pronouncing the sharp “t” in the first word, rather than softening or eliding it, the way most Americans would. He even did a few lines a cappella, which might have been a message to the doubters, or else a concession to tradition: suffice it to say that Fenix Flexin is not the first rapper to find himself in a night club, at 1 A.M., performing a notably imprecise rendition of his own hit single.
All summer long, as “Rubberz” was becoming harder to ignore, so, too, were questions about A.I. in music. When Kanye West released the album “Bully,” in March, fans tried to figure out whether certain tracks had been created with A.I., a technology West had previously praised. (West was later sued by a producer who claimed he had helped the artist create A.I. vocals on “Bully” and its predecessor, “Vultures II.”) Startups including Suno allow users to create songs partly or wholly powered by A.I., and some A.I. entities have begun to attract followings: Eddie Dalton, an old-school soul singer who corresponds to no physical human being that we know of, has half a million monthly listeners on Spotify—about as many as Clarence Carter, who died in May, at the age of ninety. In June, an Australian producer and d.j. named Josh Fawaz made it to No. 4 on the Billboard Hot Dance / Electronic Songs chart with an A.I. version of “Like a Prayer.” And HUGEL, a popular French d.j. and producer, has one of the summer’s biggest dance hits with “Movin’ to the Sun,” which seems to be a remix of an A.I. track that was originally associated with a social-media account said to belong to a Reiki healer from Hungary known as Imael Angel. The new version is jointly credited to HUGEL, Imael Angel, and the legendary house-music singer Ultra Naté, who apparently lent her human voice to this transhuman collaboration.
It would be an understatement to say that the growing prominence of A.I. music is generating backlash—sometimes it seems this music generates nothing but backlash. A producer named Medasin made a video arguing, convincingly, that Fenix and his collaborator Purps on the Beat had used an A.I. program called Treblo to make “Rubberz,” and he showed how easy the process might have been. (Using simple prompts such as “synthwave 80s” and “lyrics that are resemblant of a rapper lifestyle,” Medasin got the app to spit out a distinctly “Rubberz”-like song.) Afterward, Fenix changed his story. “Never said I didn’t use AI,” he wrote on Instagram, adding, “ain’t nothing wrong w using new technology as a tool keep up w the times or get left behind.” When a puckish British electronic duo known as Two Shell talked, in a recent interview, about using A.I., Tom Breihan, a senior editor for the music site Stereogum, delivered an unambiguous verdict that summed up the general mood: “No! Wack! Fuck that shit!” A month ago, a rapper named Tyga released a synth-pop album of his own, and he spoke freely about using A.I. to make it. His reward was a score of 0.0 from Pitchfork (the site’s first zero rating in nearly twenty years), and a similarly harsh review from the pop star Doja Cat: “Tyga is a penis for making an A.I. album.”
Sometimes this backlash concerns the ethical implications of A.I., which produces music by drawing from preëxisting songs, often without compensating the humans who made them. (“I feel like I’m at war because of A.I.,” the singer SZA said, in a recent interview.) But even more people seem concerned about the way A.I. lowers the barriers to musical entry, allowing would-be musicians to skip the step where they learn a skill or express a feeling. Many listeners—and not a few musicians—regard the process of making a song with A.I. as not just uncreative but anti-creative, and maybe anti-human, too. Earlier this year, the musician and producer Jack Antonoff posted a brief manifesto online, which said, “To everyone who is gassed up about the new ways you can fake making art, by all means drive right off that cliff.” To Antonoff and others, A.I. music seems neither good nor bad but something else altogether: fraudulent.
The fight against musical fraudulence has a long and illustrious history. The current battle dates at least to 1906, when John Philip Sousa testified before the House Committee on Patents. Sousa, who wrote military marches such as “The Stars and Stripes Forever,” was perhaps the most popular composer America ever produced, and he had come to deliver a warning. “These talking machines are going to ruin the artistic development of music in this country,” he said, of a device that was a few decades old and growing in popularity—the phonograph. In an essay published the same year, Sousa referred to phonographic cylinders and disks as “canned music,” a degraded form that could never, he said, capture or transmit the qualities that make music special. “Music teaches all that is beautiful in this world,” Sousa wrote. “Let us not hamper it with a machine that tells the story day by day, without variation, without soul, barren of the joy, the passion, the ardor that is the inheritance of man alone.” Sousa’s concern was not entirely aesthetic: part of his complaint about what he called “infernal machines” was that phonograph manufacturers paid no royalties to composers; his testimony helped inspire the 1909 Copyright Act, which changed that. In 1930, when theatres were replacing live orchestras with prerecorded soundtracks, the American Federation of Musicians launched a new campaign against “canned music,” buying a series of newspaper advertisements to decry the “corruption of an art by soulless mechanization.” One image depicted a man playing guitar flanked by a robot playing a banjo, and the text made an argument that many A.I. skeptics repeat today: “Where there is no feeling, no emotional capacity, there can be no music.”
Of course, this canned music was composed and performed by humans; “mechanization” referred only to the technology that allowed the performances to be recorded and played back. Aldous Huxley, in his novel “Brave New World,” from 1932, imagined, instead, a “synthetic music machine” able to autonomously create an endless soundtrack; around the same time, a company called Muzak was commissioning instrumental recordings that it delivered, via telephone wire, to workplaces and homes, creating a non-stop stream of background music. (In this magazine, in 1947, a Muzak executive described his product as “music you can eat to, shave to, work to, but not listen to.”) Muzak was a corporate success, although it swiftly became a cultural punch line—people used “Muzak” as a pejorative term for anything dull and low-quality. Starting in the eighties, the company began to supplement its library with original recordings of popular songs: a slightly less canned version of canned music. The company abandoned the name Muzak in 2013, by which time most of us had long stopped noticing the omnipresence of background music in our daily lives. Sousa was wrong to imagine that America might be able to resist the lure of the phonographic disk and its successors, but he was right to imagine that the recording industry would transform our ideas about what music is, while more or less killing off the tradition of the after-dinner sing-along and all but eliminating pianos from the nation’s living rooms. In popular music especially, the authentic artifact is now the recording, while a live performance is typically an ancillary experience.
As recording studios became more sophisticated, and more central to popular music, artists occasionally sought to reassure fans that they were really hearing what they thought they were hearing. The self-titled début album from the band Queen, from 1973, included in its instrumental credits a stern disclaimer: “ . . . and nobody played synthesizer.” The rock band Boston issued a similar statement (“No Synthesizers Used”) in a 1978 album, and so did Rage Against the Machine in 1992 (“NO SAMPLES, KEYBOARDS OR SYNTHESIZERS USED IN THE MAKING OF THIS RECORDING”), and the White Stripes in 2003 (“No computers were used during the writing, recording, mixing or mastering of this record”). By contrast, Stevie Wonder seemed proud to tell listeners that his 1972 album, “Music of My Mind,” contained both Moog and ARP synthesizers, which were then a novelty. He emphasized that his album was “virtually the work of one man”; technology was making it easier for musicians to work alone, helping them to create records that could be touted as more personal. Many pioneers of electronic pop acknowledged this tension by making music that was self-consciously robotic, like Kraftwerk paying tribute to “the man-machine” or Ultravox singing, deadpan, “I want to be a machine.” And yet these synth-pop musicians were nevertheless eager to insist that their music was real music. In a documentary about the synthesizer era, Andy McCluskey, from the English electronic-pop group Orchestral Manoeuvres in the Dark, described his creations in terms Sousa might have appreciated: “It was all written by real human beings, and it was all played by hand.”
It may be no coincidence that “Rubberz” drew inspiration from this period in particular. (It sounds a little bit like “If You Leave,” an O.M.D. hit from 1986.) The song is a pastiche of a form of musical innovation that now sounds quaint, precisely because of its perceptible artificiality; the sharp backbeat and arpeggiated synthesizers remind us of an earlier and perhaps simpler time. By the end of the eighties, discussions of musical authenticity usually centered on the use of prerecorded sounds. In hip-hop and other genres, producers were using snippets of other people’s records; courts often ruled that this was a form of theft, although hip-hop fans tended to see it as righteous theft. And, onstage, pop musicians frequently performed alongside recorded tracks, which audiences tolerated—to a point. In 1990, the German duo Milli Vanilli was exposed for having lip-synched both onstage and, in effect, on record, as the two ostensible members of the duo had nothing to do with its music. This, everyone seemed to agree, crossed a line; the group eventually disbanded, and in the aftermath of the scandal the state of New Hampshire passed a law requiring concert promoters to inform audiences when they know, in advance, that “all the lead vocals consist of played recordings rather than the actual singing of those vocalists.” (New York recently enacted a similar law, aimed at A.I., requiring companies to notify the public if advertisements feature “synthetic performers,” although this law explicitly excludes movies and other “expressive works”—presumably including songs.) Nowadays, most big concerts include at least some prerecorded elements; one duty of a touring drummer is to play in time with a click track, which keeps the music onstage in synch with the music on tape. Since so much modern popular music consists of electronic elements, it’s not always clear what it might mean to play these songs “live.” Often, what a concert offers is the undeniable thrill of being in the same room as the person who made the music, regardless of the precise meaning of “made.”
Museumgoers generally agree that a great artist need not wield a paintbrush. They understand that some celebrated artists have assistants, or studios full of them, while others incorporate found or readymade objects into their practices; indeed, the word “practice” encourages viewers to think holistically about the creative process, instead of fixating on the final product. Avant-garde composers have sometimes built work out of noise or silence, or they’ve written scores that consisted only of instructions, some of which may remind contemporary audiences of A.I. prompts. (La Monte Young, “Composition 1960 #13”: “The performer should prepare any composition and then perform it as well as he can.”) In the world of popular music, different genres tend to have slightly different rules. Outside songwriters are esteemed in pop and country but less so in hip-hop; outside producers are celebrated in hip-hop but less so in house and techno, where the producer is the artist. As the process of music-making has grown more electronic, and therefore more mysterious, the human voice has become more central to ideas of musical authenticity. (Or, rather, the semi-human voice—singers in just about every genre have learned to love Auto-Tune.) Usually, we can treat the person or the group on the album cover as the album’s author, but not always. In the old days, record companies sometimes packaged albums with little or no input from the artists themselves, and K-pop fans still argue over how much creative control their favorite idols really have. Britney Spears is an extreme case: her last three albums were released while she was subject to a legal conservatorship; she later said, of some of the music, “I wasn’t into it.”
Part of the reason for the A.I. backlash is that people worry that this complicated system of attribution will be overwhelmed by a tide of interchangeable, computer-generated songs that weren’t really made by or for anybody. I think that’s unlikely. Most A.I. music tends to be disdained by listeners as inauthentic slop—sometimes because it lacks quality, and more often because it lacks context. On its own, A.I. music is bootleg music—not “bootleg” like an unauthorized live album but “bootleg” like a knockoff pair of sneakers, widely and cheaply available. It hasn’t been claimed by an artist people love or authenticated by a brand that people care about, and so it’s not the real thing. As the Muzak corporation discovered, decades ago, people don’t like to think of themselves as consumers of undifferentiated slop. They want “real” music for the same reason they want “real” Nikes: because they want to feel connected to a tradition, a narrative, a perception of quality, a person.
Sometimes, of course, a bootleg can become authentic. This past summer, when the Knicks won the N.B.A. Finals, a flood of bootleg Knicks T-shirts engulfed the city; years from now, collectors might distinguish a genuine 2026 bootleg (purchased, perhaps, on Seventh Avenue, on the night of the championship) from a worthless knockoff of a knockoff. The designer Dapper Dan, who helped invent hip-hop fashion, was originally known for his high-fashion knockoffs, but consumers came to view him as a brand unto himself. Anyone who hopes to use A.I. technology in a singular and memorable way faces a similar question: How do you make valuable slop?
One answer is: you don’t. Righteous refusal has often been a compelling musical strategy, and sometimes an effective political strategy, too. In the nineteen-forties, the American Federation of Musicians fought for higher royalty rates by going on strike, forbidding its members from making new commercial recordings for more than two years. Today’s pop musicians don’t have that kind of union power, but the A.F.M. is currently suing two of the big labels, Universal Music Group and Warner Music Group, over their deals with A.I. companies. In the meantime, we may see more acts pledging, Queen-like, that their albums contain no A.I., although this disclaimer might be hard to define. Few people seemed to mind when Charli XCX and her producers talked about having used an A.I.-enabled plug-in to stitch together sampled sounds in a recent song. (Especially with instruments, it can be hard to draw a line between A.I.-enhanced and A.I.-generated.) Queen eventually changed its policy on synthesizers—the group’s eighth album, “The Game,” from 1980, began with the eerie, spiralling sound of an Oberheim OB-X. Most listeners have no idea how their favorite songs were made, but, if A.I.-powered musical tools continue to proliferate, eschewing A.I. might soon be less like eschewing synthesizers and more like eschewing electronics altogether: possible, but only for musicians who really want to make a point.
At the other extreme, audiences may learn to love A.I.-generated songs. Pat McAfee, a top sports commentator, recently released a country album, and one of his collaborators, a bona-fide country singer-songwriter named ERNEST, said that the process began with McAfee “chanting” original lyrics into Suno. The album is awful, but pop history is full of examples of memorable music arising from unpromising circumstances; perhaps a celebrity with a more distinctive lyrical sensibility, or a better ability to identify an exceptionally catchy Suno melody, will use these same tools to build a pop career. In music, as elsewhere, curatorial skill may grow more valuable, along with the ability to write a creative A.I. prompt. (If this sounds to you like a musical dystopia, then you can imagine how Sousa would have felt if he’d learned that d.j.s would eventually conquer the world.) Yet outmoded musical expectations have a way of sticking around: listeners still love the sense that their favorite performers are working hard for them. And so anyone who manages to reliably generate popular A.I. songs will be incentivized to act like a musician—to go out and put on some kind of performance, giving audiences a live version of music that was canned to begin with.
That, more or less, was the fate of Fenix Flexin this summer; after about fifteen minutes on the microphone in New Jersey, he headed to Queens for his second club gig of the night. He has been mercilessly roasted online, but the months of non-stop discourse have given “Rubberz” a kind of historical authenticity. The song is now indelibly tied to this summer, a time of excitement and anxiety over A.I.; it is also indelibly tied to the artist behind it, even if he didn’t actually sing it. But, despite its seeming ubiquity, “Rubberz” has been only a modest hit. It peaked at No. 58 on the Billboard Hot 100, and remains notably less popular on YouTube and Spotify than the hip-hop hits Fenix Flexin used to be known for. The anti-A.I. movement seems driven by a worry that people will be fooled by slop, but the listening public tends to have strong and rather unpredictable tastes, and also an insatiable hunger for personal connection. It has grown dramatically easier to make a song in the century since Sousa sounded the alarm about proliferating music technology. But it hasn’t grown any easier to make a song that sticks. ♦