“If an appreciation of Bourdain’s genius can fool audiences, what if it were a film about a controversial subject with another intent?” film professor Ted Braun asks
Morgan Neville’s documentary about the late chef Anthony Bourdain, “Roadrunner,” has raised a host of questions about the use of cutting-edge technology — particularly in supposedly nonfiction films. If Neville was able to use artificial intelligence to have “Anthony Bourdain” read aloud an email that he never actually spoke, some experts worry about other ways filmmakers could fool audiences and the lines of fact and fiction.
“Hey, maybe if I’m running for president, I can use Anthony Bourdain as my speaker. He can introduce me before every speech. Maybe that’s not something Anthony Bourdain would do, but who has the right to allow that?” Sameer Singh, an associate professor of computer science at the University of California, Berkeley, told TheWrap.
You've reached your article limit.
Unlock premium content with a subscription.
Click Here Already a subscriber? LoginBetween viral Tom Cruise deepfakes, Paul McCartney looking like a teenager in his latest music video, and a man using AI to simulate talking with his dead fiancée and the drama that surrounded the Bourdain documentary “Roadrunner,” in which his voice was digitally replicated for 45 seconds, there’s no doubt AI is becoming far more prevalent in Hollywood.
Singh explains that a single engineer over the course of a week, with enough available data, can create a lifelike, synthetic voice just like the short audio clips of Bourdain in “Roadrunner.” In fact, it wasn’t until “Roadrunner” director Morgan Neville pointed out that Bourdain’s voice was fake that people started asking questions about whether all this new technology is a good idea — particularly when it comes to nonfiction projects like “Roadrunner.”
“It’s fairly easy,” Singh said, adding that in most cases AI-engineered voice clips can be “indistinguishable” from the real thing. “There are some aficionados that can detect the difference, but I would say the technology is pretty good right now.”
Most audiences might not bat an eye at a digitally de-aged Samuel L. Jackson in “Captain Marvel” or a youthful, digitized version of Princess Leia after Carrie Fisher’s death in “Star Wars: The Rise of Skywalker.” And any debates about those instances are “uncanny valley” dilemmas or questions about whether it’s a respectful representation of someone’s legacy. Documentaries, however, raise different questions. Though filmmakers like Werner Herzog have blurred the line between fact and fiction and others, like Errol Morris, have pioneered actor re-creations of key moments for dramatic effect, it was always clear to the audience of the filmmaker’s intentions.
Thelma Vickroy, chair of the Department of Cinema and Television Arts at Columbia College Chicago, bristled at the use of AI in “Roadrunner” and felt director Morgan Neville could have found a different narrative method to get the same emotional impact.
“It can be AI, or it can be something else, or a doctored picture. You gotta make sure that the audience knows in a way that’s so pivotal of where it’s coming from,” she said, noting that “Roadrunner” never makes clear when Bourdain’s actual voice is heard and when it’s an AI approximation of his voice. “It detracts from Anthony Bourdain and his life, and it also detracts from the story the filmmaker has made. So now all that is talked about is this use of technology… It’s a slippery slope we’re going into.”
But Ted Braun, a professor of cinematic arts at USC, was less disturbed by Neville’s audio fakery. “It’s a relatively benign instance of creating the illusion that someone had spoken something when they hadn’t,” he said. “But it’s not hard to extrapolate and see how this tool could be used in dangerous and nefarious ways. Part of what alarms people is the Pandora’s Box element of this. If an appreciation of Bourdain’s genius can fool audiences, what if it were a film about a controversial subject with another intent?”
Zeena Qureshi, CEO of the AI vocal company Sonantic, said she’s gotten several inquiries from documentary crews looking to create AI voices based on deceased subjects — but finds the use of AI deepfakes in nonfiction films particularly troubling.
“Using a voice model to make it appear as if a politician said something they didn’t really say, for example, could have grave repercussions,” Qureshi wrote in a wrote a blog post about the ethics of AI voices in film. “AI models could also be used to capitalize on vocal talent without securing permission or providing adequate compensation. For example, unscrupulous companies could use AI models of celebrity voices for unauthorized product endorsements or unpaid participation in creative projects.”
Qureshi noted even just three hours of voice data fed into their engines can generate highly expressive, emotional and lifelike voices. The technology is very useful for gaming and film studios to create lifelike, expressive and emotional artificial voices to automate and simplify the preproduction processes. AI voices can help game designers get a better sense of how a video game character’s voice will sound before they bring in a real voice actor. And that’s especially useful for game studios when working with scripts that can run thousands of pages long.
But the use of AI can also raise red flags about ways the technology could become dangerous. Nhat Phong Tran, VFX supervisor at Scanline VFX, whose credits include work on “Justice League” and the upcoming Ryan Reynolds film “Free Guy,” said that deepfakes like the Bourdain example are only possible if you have enough data of a person to feed into a model — which is increasingly easy for celebrities with lots of data mining possibilities. “In theory, you could have an actor live forever,” he said.
Singh even raised an ethically dubious example known as “lip-syncing,” in which both a person’s voice and their face or lips can be deepfaked. He suggested an example in which episodes of “Friends” airing in foreign countries instead of using clunky voiceover dubs could instead by enhanced with AI to make it look as though Jennifer Aniston actually spoke those words in another language. But apart from the cost, such a scenario raises issues of consent that have prevented it from going mainstream.
There’s also the issue of AI amplifying the biases of the data it’s fed. Anyone who has seen Twitter bots go rogue will know how things can quickly get sexist, racist or toxic — and the more AI goes into the development of Hollywood scripts, visuals or even audio can inadvertently introduce those same problems.
“Maybe their sound sounds a little different from what they actually sound like. And when AI makes it really easy to do that, everybody who’s watching movies starts thinking that’s what normal is,” Singh said. “Small problems or biases in AI models can really proliferate or affect quite a bit. It’s something to be careful about, and it’s not something we always think about with AI, and it sometimes can have unintended consequences.”
Sonantic addresses ethical concerns by making sure it understands the full extent of the project before creating any synthesized voice models, and actors that are utilized in creating artificial voices can become part of the preproduction process and even get a profit share if their work is used. Sonantic also only works with businesses and creatives, rather than individuals.
And in the case of individuals either living or dead, the company obtains permission and make sure the owner of the voice is aware of its use. What’s more, the company only makes use of publicly available data to create the voice models.
“Roadrunner” initially raised concerns that Bourdain’s widow never approved the use of AI in the film, but Neville after the initial controversy clarified that he originally envisioned AI as a way of allowing Bourdain to narrate the film in the vein of “Sunset Boulevard,” one of Bourdain’s favorite films, and that he did ultimately get approval from Bourdain’s estate.
AI has for years been used in ways both big and small — both obvious and subtle — across all aspects of the film industry. It can simplify time consuming VFX tasks, it’s been used to help companies determine how a movie will perform with certain marketing or can even help edit a trailer.
But moving into the future, AI is slowly encroaching on more creative sides of Hollywood. Screenwriters can employ AI for script development, producers can use AI to help identify which actor would be best for a role and composers can use it to assist in assembling a score. It may never replace actors or directors entirely, but it can change the industry.
“If that’s the question, how do you define who the writer was, is this actually creative,” Singh said. “Whether that makes them more creative or is actually making them less creative is going to be the discussion. But that I think is the future and in some ways can get problematic.”
Brian Welk
Film Reporter • brian.welk@thewrap.com • Twitter: @brianwelk
Beatrice Verhoeven
Film Editor, Twitter: @bverhoev