Last Friday, DJ David Guetta tweeted a video in which he explained that he used Eminem’s vocals in a recent set and “people went nuts.” One tiny thing to note: Artificial intelligence had more to do with it than the actual rapper. Guetta said he used A.I. tools to recreate Eminem’s voice and generate the lyrics. In other words, the real Eminem had nothing to do with his “appearance” in the set.
Guetta added that he would “obviously” not distribute the track commercially. He did not address the legality of using an A.I.-generated version of the rapper’s voice singing A.I.-generated lyrics (“This is the future rave sound. I’m getting awesome and underground”), in this case in a concert setting that he was presumably paid for. Vocalists and other artists feeling anxiety over such capabilities can be forgiven.
What appears certain is that such tools are only just getting started, that both Big Tech and startups are racing to offer them, and that the legal ramifications have yet to be sorted out. Depending on your perspective, the possibilities are either exciting or worrisome—or perhaps a mix of both.
While Guetta did not specify which tools he used, he made clear that A.I. tools were behind Eminem “appearing” in the track.
“Eminem bro, there’s something that I made as a joke and it works so good—I could not believe it,” Guetta said. “I discovered those websites that are about A.I. Basically you can write lyrics in the style of any artist you like, so I typed: ‘Write a verse in the style of Eminem about future rave,’ and I went to another A.I. website that can recreate the voice. I put the text in that and I played the record and people went nuts.”
Guetta’s stunt comes amid the rise of ChatGPT, the A.I. chatbot from OpenAI that has helped to popularize “generative A.I.,” which refers to tools that can, among other things, deliver answers, images, or even music within seconds based on simple text prompts. Microsoft has invested billions in OpenAI and is weaving its technology into a wide range of its products. This week for example it released a new version of its search engine Bing enhanced with ChatGPT.
Google is responding to worries that it’s falling behind Microsoft by planning or releasing a raft of its own A.I. tools. This week it launched a test version of a ChatGPT rival called Bard, and last week it announced an $300 million investment in the startup Anthropic, which makes another challenger called Claude. Last month, Google showed off a tool called MusicLM, not yet released to the public, that can transform hums and text into actual music in a variety of styles.
In a TED talk five months ago, multidisciplinary artist Holly Herndon discussed the ability to create music using any musician’s voice. She also introduced Holly+, an A.I. tool that lets others sing in her voice, including in other languages.
“If I can allow people to play with my IP, my digital identity, my intellectual property, what might they come up with?” she said. “Could someone else go on tour as me, with my permission? Could I be in a thousand different bands in multiple languages?”
Or in the case of Eminem, could he appear in random DJ sets rapping lyrics he never actually penned or performed? And what if those “appearances” were awful, in his mind?
Herndon foresees opportunities for artists who consent to such arrangements. But she also sees downsides to anyone being able to make tracks featuring a vocalist who never agreed to it.
The legal site Nolo notes, “If you use ‘samples’ of other people’s music in your own music, you should first obtain written permission, so as to avoid allegations of copyright infringement. This is particularly true if you intend to release your music commercially and profit from the sales…Failure to obtain the proper permission could lead to serious consequences, including lawsuits for money damages.”
It’s easy to imagine talented but non-famous vocalists earning extra income by allowing others to use their voice in A.I.-enhanced tracks or performances. On the other hand, A-list musicians—not to mention their labels and lawyers—might be less receptive to the idea.
A.I. systems can produce “clones” of someone’s voice with even small samples to train on. Microsoft recently showcased a tool called VALL-E that, after being fed just three seconds of somebody’s voice, can synthesize audio of them appearing to say anything, as Ars Technica reported last month.
Up-and-coming musicians might want to be careful about the contracts they sign today, suggests Herndon. Presumably it would be a bad idea to give others free use of their likeness or voices forever.
“I could see people signing away contracts right now that could have really detrimental impacts on their future ability to make work as themselves,” Herndon told Wired in November. “I do want people to understand how powerful these systems are and how having sovereignty over training data is really important. The thorny questions that are being asked right now—it’s really important that we get them right.”
For Herndon, the key is permission. She and colleagues at an organization called Spawning offer the website haveibeentrained.com, where artists can check to see if their art or likeness is being used in popular A.I. training sets. If it is, the thinking goes, artists should be able to opt out. One company agreeing to honor opt-out requests from the website is Stability AI, which uses generative A.I. to create images from text prompts.
Yet artists, vocalists, and others who take a permissive approach to their intellectual property, or IP, could profit from doing so, if the right systems are set up, Spawning suggests on its website: “Some may choose to take the permissive IP approach to AI models we pioneered with Holly+, where Holly offered her voice model for others to use in return for a share of profits in officially approved derivative works.”
