From top left: Celebrities including Demi Lovato, T-Pain and Tom Brady have lent their likenesses to artificial intelligence.
Robot: Adobe Stock (4). Lovato: Santiago Felipe/Getty Images. T-pain: Monica Schipper/WireImage. BRADY: Philip Farahon/Getty Images for Paramount Pictures.
The sound is unmistakable. As a video of a woman wading into the ocean at sunset plays, a melodic tune is superimposed: “I wake up, wake up, wake up in the sun, this is my life.”
The voice is clearly T-Pain’s, including his signature Auto-Tune quirk. What’s the gain? The lyrics, as well as T-Pain’s vocals themselves, were created by a generative artificial intelligence technology called “Dream Track” on YouTube. AI programmed the song and recreated T-Pain’s voice so that users could soundtrack it to their personal videos. YouTube has also signed up the likes of John Legend, Demi Lovato and Charlie Puth for its Dream Track technology.
Facebook and Instagram owner Meta has launched a series of artificial intelligence “characters” in beta, all of which resemble celebrities such as Tom Brady, Snoop Dogg and MrBeast. Tech startups like AI-focused Soul Machines, in which SoftBank is also an investor, are rolling out AI-powered celebrity avatars that let users interact with and interact with “digital twins” of boxer Francis Ngannou, K-pop stars and more. chat. Mark Tuan or golfer Jack Nicklaus.
As Hollywood guilds grapple with the potential of generative artificial intelligence to transform film and TV production, tech companies are harnessing the power of celebrities to educate the masses about the underlying technology. “There’s a huge potential business there, and I think that’s what YouTube and the music companies see, for better or worse,” said former executive producer Gavin Purcell. The Tonight Show Starring Jimmy Fallon Who is hosting now? human artificial intelligence podcast. “Facebook and YouTube are trying to get people into what they see as the next world of UGC, user-generated content, which is this artificial intelligence thing.”
This is no small bet. Warner Music Group CEO Robert Kyncl shared the technology behind creating the press at an event at YouTube’s New York offices in September. “Since then, technology has transformed the industry and made it better and more prosperous to a large extent,” he said. “But whenever that happens, there’s a period of uncertainty because there’s profound change, and the change is unsettling. “We’re in that period of change. “
Currently, AI-powered celebrity likeness (whether it’s a look-alike chatbot or a similar-sounding audio) is in the experimental stages, but with the expansion of Meta’s chatbot and YouTube’s Dream Track, this is expected to happen in the coming years. That will change over the course of the month (Meta, for example, you already have recordings of many participating stars, so their chatbot may be able to talk to users outside of text in the future). “Look at how many celebrities today are creating products, brands and e-commerce stores as an extension of their social media presence,” said Greg Cross, CEO of Soul Machines. “So it becomes digital engagement Another aspect of.”
But the technology also raises thorny questions about identity, intellectual property, compensation and safeguards. A movie studio owns the IP for a movie and the script is written in advance. With AI-generated content, the technology can write everything instantly. Contracts can stipulate compensation, but things are not so simple when a celebrity’s likeness or voice is used to train a model.
In an industry where the famous and powerful protect their image so zealously, how confident can they be that their AI-generated counterparts won’t cheat?
The compensation part is in some ways simpler—at least for now. Meta paid millions of dollars each to some of the celebrities who signed on to its chatbot project during the deal, sources confirmed. YouTube is also considering how to pay participating musicians (it already shares revenue with artists whose songs are used in videos, so one could see a similar model applied to AI-generated tunes based on the style of real artists).
Things get more complicated when you think about the future. What happens when these deals expire and one party or the other chooses not to renew? Personal data used to train AI models can be deleted, but only to a certain extent, sources said.
“It’s like saying, ‘You should go see a movie Star Wars Then forget you’ve ever seen this movie. That’s not possible,” said a source familiar with the discussions. “You can delete the data, but there’s still residual data that tells us how the system was trained. “
“These models essentially require some kind of perpetual license,” the source added. “I don’t know that the big tech companies are thinking about or preparing to deal with this.”
Then there’s the issue of guardrails. It’s one thing to have AI John Legend sing about your dog, or AI Tom Brady (sorry, his character’s name is “Bru”) talk about his Super Bowl pick. But what can be done to stop these AIs from being misused to issue threats or say things that are inappropriate for their real-life alter egos?
Guardrails exist, but are imperfect. “We specifically train them so they don’t use profanity or deal with offensive themes,” Cross said.
Will Chan, creative director at Mark Tuan, said: “Soul Machines is very keen to ensure that if fans take extreme routes and ask questions (touching on taboo topics), the filter is in place. We don’t want to overdo it with the potential for outrageous reactions. Control. I think it’s just trying to steer the narrative toward, you know, a fun, interactive, entertaining service.”
But given the models involved, there are limits to how much fencing can be done. Take Kendall Jenner, for example, who provides the face for “Billie,” one of Meta’s chatbots that the company describes as a “no-nonsense, ride-or-die companion.”
Users can encourage Billie to stop drinking, which is interesting considering Jenner has 818 Tequila. Another user asked Billie what her favorite tequila was, and she recommended Don Julio and Patrón, saying she wasn’t familiar with Jenner’s brand.
“It’s likely to happen with something like this because you can’t train for everything, and basically the basis of a chatbot based on one of these LLMs (large language models) is to be able to answer all questions based on its training, “Purcell said.
Earlier this year, influencer Caryn Marjorie (2.6 million followers on Snapchat and 745,000 followers on YouTube) partnered with a company to create a virtual AI chatbot based on her likeness. However, users were able to have the chatbot engage in sexually explicit conversations that went beyond what was planned.
“There’s only so much you can be sure of, and if you mess with it enough, it’s going to start to break,” the source said.
While tech giants like YouTube and Meta are focused on releasing the technology safely, some other companies may be more fast and loose in their approach. What’s to stop someone from impersonating Drake’s voice and threatening their neighbors?
“I don’t think you can,” Purcell said. “I think it’s really going to be a complicated world and there has to be some kind of ‘this is an official model from me’, some kind of watermark or statement.”
“You have to get that person to come forward and say, ‘This is not me.'” And then, whether you like it or not, you’re probably going to have to have some serious lawsuits and laws based on it,” he added.
One senior media executive grimly predicted that AI-powered models “will be everywhere anyway” and that it would be better if companies set up a “rules-based place” to experiment. “Media companies need to learn from their mistakes and create safe spaces for these things,” lest they hand over control entirely to big tech companies, the executive added.
The proliferation of celebrity AI avatars, chatbots and music experiments shows that many celebrities have already embraced this future.
“Every celebrity today is interested in their digital footprint and how they use it to connect with their fans,” Cross said. “So this becomes an extension of social media.”
But there is also a cautious side.
Charli XCX, who participated in the YouTube experiment, said in a statement: “When YouTube first contacted me, I was cautious and I still am now. Artificial intelligence will change the world and the music industry in ways that we do not yet fully understand.” Related to the launch of Dream Track.
“It’s our job — the platform and the music industry — to make sure artists like Charli (Pooce) who read can benefit; it’s also our job together to make sure artists who don’t want to read are protected,” Kyncl said at the YouTube event.
Exactly how they will be protected in an AI-first world and how much they will benefit from licensing and compensation is still nascent and uncertain.
“I think any big tech company that’s developing these technologies is going to have a lot of work to do and a lot of ongoing maintenance and tracking of it,” said a source familiar with the work of big tech companies. “And I don’t know if it’s ever going to be perfect.”
A version of this story first appeared in the Nov. 29 issue of The Hollywood Reporter magazine.Click here to subscribe.