AI Could Usher in a New Era of Music. Will It Suck?

Dec 27, 2023

Amos Barshad

Michael Sayman has worked at Facebook, Google, Roblox, and Twitter. At 26, the software engineer has already published a memoir, App Kid. But until he began work on his latest project, he’d never built a website. "I made it in five hours over the weekend, out of frustration that there wasn't anything like this," he says. "Now there's nearly a million streams on the site."

Sayman's site is AI Hits. Ever since it launched in April, it's been aggregating a controversial new musical … thing: songs created with artificial intelligence tools that mimic, with chilling accuracy, mainstream stars like Drake and Kanye West. The conversation around AI music has largely been frenzied and shot through with hand-wringing over What It All Means and What It All Portends. But Sayman is an AI optimist. So he built the Hot 100 for AI tunes.

Will Bedingfield

Will Bedingfield

Will Bedingfield

AI Hits sifts through the mushrooming detritus and ranks the tracks per their collective streams on the various platforms on which they’re posted, linking out to the tracks directly (unless, of course, they’ve already been taken down by the time you click). With its notable fealty to the real thing, the Drake-mimicking "Heart on My Sleeve" became the first "hit" of the AI music era, and various versions of it dominate AI Hits. (Sayman points out there's even one featuring AI-generated Joe Biden vocals, for some reason.) AI versions of Ariana Grande, Travis Scott, Juice WRLD, SZA, and Lana Del Rey are all also represented on the chart.

In conversation, Sayman uses the term "voice" to refer to the artist being mimicked, and he uses the term "artist" to refer to the username of whoever created the song. This use of nomenclature may seem small, but it is significant. It's a step toward the creation of a shared lexicon around all this stuff. The landscape of AI music is endlessly, discursively messy, but as Sayman points out, we’re all present at the outset of a conversation that will unspool over years. "How do you search? Who are the creators? How do you attribute labels to them? What do those revenue splits look like?" he says. "And how does that even work, when you can make a hundred remixes of the same song?"

That latter question, over the legality of the practice of AI music, is central. Spotify quickly took down "Heart on My Sleeve," and UMG, Drake's parent label, has pushed the company to purge thousands of other AI-made songs. On a recent podcast interview, Ice Cube urged Drake to directly sue the creator of "Heart on My Sleeve," and he has tweeted that he finds the idea of generating a song in the style of a dead artist without the approval of the artist's estate to be "evil and demonic." But when looking past AI's potential for legal or ethical blunders, other artists, from pioneering musicians like Holly Herndon to legacy acts like the the Pet Shop Boys, are bullish on AI as a creative tool. It could even unfurl a whole new genre of music.

Sayman believes AI can create a more democratic, open-ended music industry. "The record labels used to hold all the power—they were in charge of distribution, resources, production quality. We’ve seen social media replace distribution and discovery of music. Now we’re seeing AI expand production quality, so there's opportunities for more people to get involved in the music creation process. More Drake singles! Instead of having two or three producers, he can have millions of producers working on those songs!" He laughs. "I’m half joking."

Parth M.N.

Lauren Goode

Joel Khalili

Julian Chokkattu

Sayman may be speaking off the cuff, but the futurist popstar Grimes has actually already embraced the idea of empowering an infinite number of nonprofessional musicians to make music with her voice. Via a site called Elf.tech, she's handing her vocals to anyone for commercial use in exchange for 50 percent of the royalties. In a recent interview with the The New York Times, she reviewed some of the music that's already been created via Elf.tech, and her sincerity in embracing the tracks provided a refreshing and much needed counterweight to the hysteria over AI.

"What I like about the early AI stuff is that you can hear the technology very profoundly," she said. "I think people will appreciate that more in five years, when they realize people only made stuff like this for a couple months."

"Heart on My Sleeve" was met in part by an almost visceral repulsion, and it's possible that reaction is rooted in the horror of how easy it seemingly was to create a fake Drake. But as H. Drew Blackburn pointed out in a piece for Bloomberg, "Drake has been making music that sounds like AI Drake for years." People are collectively obsessed with the theoretical ramifications of AI music. But maybe there's another question everyone is forgetting to ask about all this content: Is it any good?

"Most people don't make good art by copying and pasting. What makes pop work is that it's always changing and always responding. This is just a feedback loop between systems."

Marc Weidenbaum is a writer, sound artist, and professor at the Academy of Art University. "The hand-wringing [around AI music], it's a strange thing to me," he says. "We’ve been concerned with creating artificial life at least since the Golem."

Ultimately, for Weidenbaum, something like "Heart on My Sleeve" is a negative, not because of any apocalyptic overtones but because it's, well, boring. He points to something like cybernetic music, in which an artist programs a machine to go off exploring and create sounds outside of the artist's control. Aleatoric music operates similarly. "Unintended consequences are a feature, not a bug," he says. "Brian Eno was interested in the idea of the composition being a garden, constantly changing."

But using AI to mimic Drake is a mind-numbingly straightforward thing to do. "Most people don't make good art by copying and pasting," Weidenbaum says. "What makes pop work is that it's always changing and always responding. This is just a feedback loop between systems."

Parth M.N.

Lauren Goode

Joel Khalili

Julian Chokkattu

For Joey DeFrancesco, a musician and organizer with the Union of Musicians and Allied Workers, what's alarming about AI is how the major labels may ultimately utilize it. "Any potentially interesting artistic uses of AI pale in comparison to the corporate domination of AI that will inevitably occur under our current music industry power structures," DeFrancesco says. "The tech capitalist fantasy" in the music industry "has always been to cut out the artists entirely and remove the need for any royalty payments."

DeFrancesco points out that artists have successfully combated damages from emerging tech in the past: "Musicians in the 1940s went on strike to demand that the profits created by new vinyl record technologies be shared with musicians, and they won." He also points to the ongoing Writers Guild of America strike, saying, "The big studios want free rein over AI so they can harvest writers' work then cut them out entirely. But writers are collectively saying no." And DeFrancesco does have reason for concern. After kicking up a fuss over "Heart on My Sleeve," UMG then moved quickly to embrace machine-learning technology by partnering with an AI company called Endel.

All of which is a reminder that theorizing about worst-case scenarios ignores the bad-case scenarios that musicians and fans are currently facing. The pressing questions over AI music are human ones. Does its existence mean that musicians will be getting ripped off in new ways? And is it worth a listen? We don't yet know the answer to either, but for the latter we do have a burgeoning resource in AI Hits.

Currently, Sayman is working on improving the search functionality and on fielding requests from AI Hits’ many impassioned users. As for Sayman's own AI listening habits, his current favorite is "Por Que," an AI duet between Rihanna and Bad Bunny. Sayman is a native Spanish speaker, and his family is from Peru. "It's kind of a Spanglish song," he says, "but it's funny—the Spanish doesn't sound too good." Apparently the AI that made the song was trained on a data set that didn't have enough Spanish pumped in. He laughs. Whatever. To him, it's still a banger.