What AI means for Taylor Swift’s music

Black Eyed Peas frontman will.i.am sees a future where stars like Beyonce and Taylor Swift “will no longer die,” as bespoke AI models trained on their work keep churning out new hits in their style and voice even after they pass away.

The hip-hop icon’s prediction, made at a panel event hosted by Artificial Intelligence Los Angeles on Tuesday, echoes some of the narratives about AI taking hold in corners of the music industry, which is once again struggling with how to adapt in the face of new technology, two decades after digital streaming upended its business model.

In April, a song called Heart on My Sleeve that used a form of AI to mimic the voices of Drake and The Weeknd was listened to millions of times before Universal Music Group filed copyright complaints to get it taken down from platforms like Spotify and YouTube. Weeks earlier, an executive at the record label warned that AI could become a “calamity for artists.”

Still, will.i.am and other speakers at the event were mostly optimistic about what AI would mean for the future of music making, betting that it would more likely spawn new genres than make artists obsolete. But discussions about how it might impact those who hold the rights to it permeated the evening.

The president of AI LA, Todd Terrazas, said that artists had been “cloned without their permission,” a practice that another panelist referred to as “musical disinformation.”

Creating the eternal ghost of an existing artist may be one of the least interesting things people will do with AI music tools  — the equivalent of a cheap party trick. Instead, musicians can generate things like unique samples or a musical solo featuring an instrument they don’t know how to play. While they might channel or remix Taylor Swift’s voice, the best examples will be clever artistic references, like how hip-hop artists used drum machines and samples starting in the 1970s.

That’s essentially how the technology is already being used. Antony Demekhin, co-founder of the AI music startup Tuney, said that Heart on My Sleeve almost certainly wasn’t created with a generative program like ChatGPT or DALL-E, which can spit out a Shakespeare essay or a finished painting wholesale.

Instead, someone more likely wrote, sang, and recorded the track, and then put it through filters that imitated the rappers’ voices. That’s still impressive, but there’s significant differences between the kinds of tech involved.

“The expectation is that you can get an AI to sing like Drake, which is not the case,” Demekhin told Semafor. “Text-to-voice is not 100%, but it’s getting there. Text-to-true generative singing? That’s even harder.”

As the technology keeps improving, the unanswered question is whether record labels and the courts will let people openly publish music using programs trained on other artists’ work. One solution is for tech companies to simply license it for that purpose, which is what some AI music startups are doing.

“Telling people that we’re copyright safe matters,” said Karen Allen, the CEO and co-founder of Infinite Album, which focuses on generating music for gaming livestreamers. “I want people to know where we stand on that issue, it is a bit up in the air with a lot of companies.” She said that she saw an opportunity in the market after learning that streamers often receive copyright takedown notices when they play a game’s official sound track.

Demekhin said that Tuney is largely relying on audio recorded by artists he worked with in his previous life running a music production company. Musicians send Tuney unfinished tracks or song fragments that have few other uses. “A lot of the people we talk to that are really, really good do not see AI as a threat,” he explained. “They see it as a way to monetize something that they weren’t able to monetize before.”

Using generative AI to mimic another musician’s voice raises thorny ethical questions, particularly about race and identity. “It’s another way for people who are not Black to put on the costume of a Black person,” Lauren Chanel, a writer focused on Black culture in entertainment, told The New York Times. “This is just another example in a long line of people underestimating what it takes to create the type of art that, historically, Black people make.”

English-speaking AI fans rejoiced last week after a blog post went viral saying Japan had ruled that training artificial intelligence models doesn’t violate the country’s copyright laws. But Japanese media reported Friday that the government is considering a draft plan and had not yet finalized its stance on the issue.

The plan notes that “content created [by generative AI] that is similar to original material could result in a large number of copyright infringements,” according to Japan News.

• If you want to hear the genre for yourself, the website AI Hits is aggregating songs made by machines. Creator Michael Sayman told WIRED his favorite track right now is Por Que, an AI duet between Rihanna and Bad Bunny.
• WNYC’s On the Media podcast chronicled how copyright law suppresses the artistic voice of hip-hop producers in a fascinating segment featuring New York University professor Dan Charnas.