Earlier this year, Bad Bunny emphatically rejected rumors that he was about to release a new song with Justin Bieber. “That’s fake,” he told TIME in an interview for a cover story on his meteoric rise. “You never know what I’m going to do.”
But last month, a song featuring what sounded like his and Bieber’s voices started circulating on TikTok, garnering millions of likes. Bad Bunny hadn’t lied in the interview, though: the song was created with AI. An artist named FlowGPT had used AI technology to recreate the voices of Bad Bunny, Bieber and Daddy Yankee in a reggaeton anthem. Bad Bunny himself hated it, calling it a “shit of a song” in Spanish and discouraging his fans from listening, and the clip was removed from TikTok. But many fans of all three megastars loved it all the same.
The song and the polarized reactions to it are emblematic of the fraught ways in which AI has stormed the music industry. Over the past couple of years, advancements in machine learning have made it possible for anyone sitting in their homes to reproduce the sound of their musical idols. One artist, Ghostwriter, went viral for mimicking Drake and The Weeknd; another creator jokingly set Frank Sinatra’s smoky voice to profane Lil Jon lyrics. Other AI tools have allowed users to conjure songs just by typing in prompts, which are effectively the audio versions of text-to-image tools like DALL-E.
Some boosters argue that these advancements will further the democratization of music, allowing anyone with an idea to create music from their bedroom. But some artists have reacted with fury that something so personal as their voice or musical style could be co-opted and commodified for someone else’s gain. The push-and-pull between protecting artists, forging innovations, and determining the complementary roles for man and machine in music creation will be explored for years to come.
“If there’s a huge explosion in music created at infinite scale and infinite speed, will that return us to thinking about what we are actually bringing to the table as humans?,” asks Lex Dromgoole, a musician and AI technologist. “Where does imagination exist in this? How do we bring character to our own creations?”
AI is already being used by music producers for more mundane parts of their jobs. AI can help correct vocal pitch and allow engineers to mix and master recordings much more quickly and cheaply. The Beatles recently used AI to isolate John Lennon’s voice from a 1978 demo, stripping out the other instruments and ambient noises in order to build a new, pristinely-produced song. AI is also ingrained in many peoples’ listening experiences: streaming platforms like Spotify and Apple Music rely on AI algorithms to suggest people songs based on their listening habits.
Then there’s the actual creation of music using AI, which has triggered both excitement and alarm. Musicians have embraced music tools like BandLab, which suggests unique musical loops based on prompts as an escape valve for writer’s block. The AI app Endel generates customized, constantly-mutating soundtracks for focusing, relaxing or sleeping based on people’s preferences and biometric data. Other AI tools create entire recordings based on text prompts. A new YouTube tool powered by Google DeepMind’s large language model Lyria allows users to type in something like “A ballad about how opposites attract, upbeat acoustic,” and a song snippet belted by a Charlie Puth-soundalike is instantly generated.
These technologies raise all sorts of concerns. If an AI can create a “Charlie Puth song” instantaneously, what does that mean for Charlie Puth himself, or all the other aspiring musicians out there who fear they are being replaced? Should AI companies be allowed to train their large language models on songs without their creators’ permission? AIs are already being used to summon the voices of the dead: a new Edith Piaf biopic, for example, will include a reassembled AI-created version of her voice. How will our understanding of memory and legacy change if any voice throughout history can be re-animated?
Even those most excited about the technology have become worried. Last month, Edward Newton-Rex, the vice president of audio at the AI company Stability AI, resigned from the company, saying he feared that he might have been contributing towards putting musicians out of jobs. “Companies worth billions of dollars are, without permission, training generative AI models on creators’ works, which are then being used to create new content that in many cases can compete with the original works,” he wrote in a public letter.
These questions will likely be decided in courts in the coming years. In October, Universal Music Group and other major labels sued the startup Anthropic after its AI model Claude 2 started spitting out copyrighted lyrics verbatim. A Sony Music executive told Congress that the company has issued almost 10,000 takedown requests for unauthorized vocal deepfakes. And many artists want to opt out entirely: Dolly Parton recently called AI vocal clones “the mark of the beast.” AI companies, conversely, argue that their usage of copyrighted songs falls under “fair use,” and is more akin to homages, parodies or cover songs.
The singer-songwriter Holly Herndon is among the artists trying to get ahead of these seismic changes. In 2021, she created a vocal deepfake of her own voice called Holly+, allowing anyone to transform their own voice into hers. The purpose of the project, she says, is not to force other artists to also surrender their voices, but to encourage them to also take on a proactive role in these larger conversations, and claim autonomy in a top-down music industry in which tech giants play an increasingly large role. “I think it’s a huge opportunity to rethink what the role of the artist is,” she tells TIME. “There’s a way to still have some agency over the digital version of yourself, but be more playful and less punitive.”
The musician Dromgoole, who co-founded the AI company Bronze, hopes that AI music will evolve out of its current stage of mimicking singers’ voices and instantly generating music. Over the past few years, Bronze has worked with musicians like Disclosure and Jai Paul to create ever-evolving AI versions of their music, which never sound the same when played back twice. The goal is not to use AI to create the perfect, monetizable static song—but to use it to challenge our conceptions of what music could be. “It seems like the tech industry thinks that everyone wants a shortcut, or a solution to creativity,” he says. “That’s not how imagination works. Anyone who’s studied flow state or spent time with people who are creating music knows that we love that process.”
