Annabelle Gordon/Sipa USA via AP Images
Actress Scarlett Johansson attends the 2024 White House Correspondents’ Dinner at the Washington Hilton in Washington, April 27, 2024.
A couple of weeks ago, shortly before it became apparent that Congress won’t be regulating the rapid development of AI technology anytime soon, the leading avatar for chatbots and other generative AI tools, Sam Altman, showed us all the true face of Big Tech and Silicon Valley, once again.
Altman, the CEO of OpenAI, tweeted out a single word—“her”—on May 13, as the company’s engineers demonstrated the capabilities of its latest AI-powered voice assistant model, ChatGPT-4o, on a livestream. The message was a clear reference to the uncanny resemblance between the assistant’s voice and the voice of a real, live, and very identifiable woman, actress Scarlett Johansson, who played an AI assistant that forms a romantic and sexual relationship with a human in the movie Her.
The intentional decision to employ a flirtatious female voice on a tool that will probably become near-ubiquitous one day was questionable enough. But days later, Johansson implied in a statement that OpenAI may have simply stolen her voice, or created a fake facsimile, after she rejected an offer to work for the company.
OpenAI has since denied that the voice was meant to sound like Johansson, and produced records showing it hired a different actress—although it’s also true that the company telegraphed repeatedly that its new feature was inspired by Her, and intimated that Johansson agreed to let the company use her voice.
The move came days before Senate Majority Leader Chuck Schumer (D-NY) closed several months of closed-door discussions with tech CEOs by announcing a “roadmap” that calls for billions in government spending on further AI research and development, and not much meaningful regulation.
A group of academics, tech accountability, and civil rights groups responded with a “shadow” report, explaining that researchers, labor organizations, and other public-interest groups “were marginalized in the conversation,” while “some of the loudest and most-self serving voices from the industry,” including Altman, Elon Musk, and Marc Andreessen, had lawmakers’ ears.
“Any legislative process that begins with industry in the driver’s seat is fated not only to fail the public interest, but even to exacerbate AI’s ongoing harms,” the group said in the May 20 report.
The case involves overt sexism and bias, which has been a troubling feature of AI applications in the past.
The dispute between OpenAI and Johansson is a neat distillation of the cluelessness, the soft bigotry, and the apathy with which the titans of industry and Big Tech conduct themselves as they develop what are widely expected to become the most world-altering, and dangerous, technologies in human history. The case also involves overt sexism and bias, which has been a troubling feature of AI applications in the past. And as much as anything, it’s a clear indication of the industry’s willingness and ability to unfairly exploit and abuse the labor of even the most powerful, well-off workers in our society.
Kate Devlin, a lecturer in AI and society at King’s College London and author of Turned On: Science, Sex and Robots, said the company’s offer to Johansson was a very deliberate choice to align its vision and products with the imagined future in Her—i.e., people substituting interaction with attentive computers for human relationships, and even falling in love with them. The movie, incidentally, was a depiction of dystopia, not a love story.
“This is despite years of criticism about the reductive and unhelpful predominance of female voices for subservient virtual assistants,” Devlin added. “Sex sells, and OpenAI knows they can get a lot more engagement with a persona for their model that coos and giggles at your every command.”
Suresh Venkatasubramanian, a Brown University professor who participated in the discussions with CEOs and Congress, said the dispute shows the need for immediate regulation, so there are rules in place to protect the public while companies are debuting AI products and experimenting with the emerging technology.
“They did this with a fair amount of impunity,” Venkatasubramanian said. “The idea that ‘you do this first and if someone complains we’ll see what happens’ is very much the ethos of a lot of what happens in this area for a while now. This is just an attention-grabbing example.”
OpenAI did not respond to a request for comment.
JOHANSSON SAID IN HER STATEMENT that she received an offer “to voice the current ChatGPT system” last September, which was around the same time that Altman said publicly that Her is his favorite movie about AI, during a demo and Q&A session at a customer relationship management conference.
The movie “inspired us,” and “certainly more than a little bit,” Altman said.
Reporters even noticed the resemblance during the demo between the so-called “Sky” voice and Johansson’s, all the way back in September 2023.
Johansson declined OpenAI’s initial offer; and she never responded to a renewed offer from Altman this month, two days before the release of ChatGPT-4o, according to her statement.
Yet, despite the lack of consent, the new voice assistant feature dropped last week, alongside Altman’s “her” tweet, and with a voice “so eerily similar” to Johansson’s that her “closest friends” could not tell the difference, the actress said in her statement.
Whatever else might be said about its portrayal of the future, Her is a movie—an expression of human imagination meant to tell a story and to maybe say something about life and living. On the other hand, Altman is a real guy, whose company apparently ignored a real woman’s denial of consent, and capitalized on her likeness and on-screen personas (her labor, in other words) in the real world. Most people probably understand on an intuitive level exactly what OpenAI and Altman were doing.
Research suggests that casting actors who are known mainly for non-voice acting into voiceover roles allows filmmakers to take advantage of the star’s on-screen persona(s). In this case, we have an actress who is well known as a sex symbol, and whose on-screen personas, including in Her, have been described as “sweet, sexy,” and “seductive” by film critics.
In short, by “casting Johansson as Samantha, the film allows us to associate Samantha with how Johansson looks, thereby giving Samantha a ‘phantom’ sexual presence,’” according to research by Devlin, at King’s College, and others.
Similarly, by trying to “cast” Johansson, and by using a similar voice, OpenAI apparently sought to imbue its product with those same qualities, in order to entice more customers. It’s quite a move for a company whose technology is expected to help form the AI infrastructure on Apple and Microsoft’s ubiquitous devices.
It’s also worth noting that OpenAI was criticized by members of Congress in December for having an all-male and all-white board, and for nominating Larry Summers, who has a history of sexist remarks. The company added three women to its board in March.
Venkatasubramanian said he was struck by how AI companies evade responsibility for their design choices.
“On one hand, they convey the impression that what they’re doing is like sci-fi, alien and completely different from anything ever done before,” he said. “They do that through the voice that’s been described as ‘flirty,’ and by cryptically referring to the movie. These are deliberate choices. But when confronted they tend to back off and say we didn’t mean that.”
THE LABOR IMPLICATIONS OF OPENAI’S DECISIONS are no less worrisome.
As a background matter, the “magic” behind AI often includes legions of gig workers. OpenAI, a company worth billions, has outsourced its grunt work to Kenyan workers making less than $2 an hour.
The development points to how AI threatens a broad range of professions, including white-collar or so-called knowledge workers, and even the most powerful in our society.
Johansson chose not to work for Altman and OpenAI; still, the company was able to use its huge public profile and its technology to reap much of the attention and other benefits that it would have gained if Johansson had decided to sign a contract with them.
Johansson has said that the company only took down the “Sky” voice after she was forced to hire legal counsel, and it’s likely she will pursue the apparent violations of her publicity rights. She may have a good case with some precedent: Going back to the 1980s, artists like Tom Waits and Bette Midler, who are famous for their distinctive, recognizable voices, have successfully sued companies that hired people to impersonate their voices for commercial use. The Midler case went all the way to the Ninth Circuit Court of Appeals, which ruled that a famous voice is distinctive enough that an imitator cannot be used in a commercial enterprise without consent and approval.
The dispute also parallels more recent circumstances, when the popular artist Drake released a song that relied on AI to generate the voice of the late Tupac, who remains one of the best-selling artists within the genre.
Polls show that most Americans don’t trust companies to use and develop AI responsibly.
Lawyers representing the late rapper’s estate sent a cease and desist letter saying their client “would never have given its approval,” and was dismayed by the “flagrant violation of Tupac’s publicity” rights and “the legacy of one of the greatest hip-hop artists of all time.” The song was scrubbed from social media after the threat of legal action.
Ironically, Drake himself was the victim of a possible publicity rights violation in April, when a content creator uploaded a new song believed to have featured AI-generated vocals from Drake and The Weeknd.
At the time, the artists’ label, Universal Music Group, made the stakes clear in a statement condemning AI-generated content that infringes on publicity rights: The issues raise a question as to which side of history these companies want to be on, UMG said: “the side of artists, fans and human creative expression, or on the side of deep fakes, fraud and denying artists their due compensation,” a UMG spokesman said in a statement reported by NPR.
UMG has also sued artificial-intelligence platforms over using its copyrighted works to “train” models; and other major businesses, like The New York Times, the Chicago Tribune, and a host of other publications, have filed lawsuits against companies like OpenAI and Microsoft.
OpenAI had to have known about these controversies before deciding to rip off Johansson’s likeness. And yet it wasn’t a deterrent to the multibillion-dollar company.
CREATIVE LABOR ON INTERNET PLATFORMS, or the fast-growing category of work known as content creation, is uniquely vulnerable to artificial intelligence. In a sense, content creators, who are effectively contract workers for social media platforms, often lack access to some of the basic terms and conditions of their employment—the algorithmic systems that handle ranking and content moderation.
The dispute with Johansson is simply a high-profile example of the power Big Tech and AI companies wield.
Of course, the dispute also demonstrates a kind of cluelessness in the industry, much like the recent ad from Apple that showed a hydraulic machine crushing a trumpet, piano, books, and other instruments of creativity and imagination, and replacing them with the latest iPad. (Apple pulled the ad and apologized after receiving major backlash.)
Polls show that most Americans don’t trust companies to use and develop AI responsibly. A recent Pew Research Center survey showed that 70 percent of Americans say they have “little to no trust in companies to make responsible decisions about how they use AI in their products,” and 80 percent said the information companies collect in the process “will be used in ways that people are not comfortable with.”
Still, industry leaders are trying to sell us on a vision of the future and of AI focused on deep relationships with flirty, attention-trapping computers that sound like Scarlett Johansson, which will also read our children bedtime stories and make things like musical instruments and pencils obsolete.
It’s almost laughable, until you remember the actual stakes at hand when it comes to AI, and the fact that these companies are still unregulated in the U.S.
