OpenAI is shelving an artificially intelligent voice assistant that has drawn scrutiny and criticism in recent days over its suspicious resemblance to the voice of actress Scarlett Johansson.
“We’ve heard questions about how we chose the voices in ChatGPT, especially Sky,” the company said on Monday morning. “We are working to pause the use of Sky while we address them.”
In a lengthy statement shared with The Daily Beast, Johansson said she was “shocked, angered and in disbelief” that Sky’s demo voice sounded “so eerily similar to mine that my closest friends and news outlets could not tell the difference.”
ADVERTISEMENT
The actress, who voiced an AI bot in the dreamy Spike Jonze sci-fi romance Her, revealed that she had been approached by OpenAI founder Sam Altman eight months ago with a proposal to lend her voice to one of the ChatGPT assistants.
“He told me that he felt that by my voicing the system, I could bridge the gap between tech companies and creatives and help consumers to feel comfortable with the seismic shift concerning humans and AI,” Johansson said. “He said he felt that my voice would be comforting to people.”
She said that she declined the offer after weighing the pitch. But about nine days ago, Altman reached out to her team, Johansson said, “asking me to reconsider.” Before she could get back to him, OpenAI released its demo of GPT-4o, a package that included Sky. Johansson said that her “friends, family, and the general public” were quick to draw comparisons between her and the artificial system.
“Mr. Altman even insinuated that the similarity was intentional,” the actress said, pointing out that he’d tweeted a single word—“her”—in the middle of the update’s rollout.
In a detailed rundown of how the company had allegedly come to choose the five distinct “voices” of ChatGPT, OpenAI denied that Sky’s voice was an imitation of Johansson’s, saying that the project had used the voice of “a different professional actress using her own natural speaking voice.” The company declined to provide the performer’s identity, citing her privacy.
The statement echoed remarks made by OpenAI CTO Mira Murati, who told tech blog The Verge last week that Sky had not deliberately been designed with Johansson in mind. Joanne Jang, the company’s model behavior lead, told the tech blog on Monday that OpenAI was “in conversations with ScarJo’s team” over the “confusion.”
“We want to take the feedback seriously and hear out the concerns,” Jang added.
In her statement, Johansson said she’d been “forced to hire legal counsel, who wrote two letters to Mr. Altman and OpenAI, setting out what they had done and asking them to detail the exact process by which they created the ‘Sky’ voice.” She said that the company’s decision to yank Sky was a direct result of her legal pressure.
The two-time Oscar nominee then widened her lens to address the issue of disinformation and the lack of regulation and legal safeguards around artificial intelligence.
“In a time when we are all grappling with deepfakes and the protection of our own likeness, our own work, our own identities, I believe these are questions that deserve absolute clarity,” Johansson said. “I look forward to resolution in the form of transparency and the passage of appropriate legislation to help ensure that individual rights are protected.”
The real people whose voices were used to model Sky and the other ChatGPT assistants—Breeze, Cove, Ember, and Juniper—were plucked from a talent pool of more than 400 people, OpenAI said in its blog post. The company’s casting team looked for voices that sounded “timeless” and were easy to listen to, it said.
After auditions, the selected actors traveled to San Francisco to record last summer. The company said the actors were compensated with “above top-of-market rates,” adding they’d all been made aware of the “scope and intentions” of the operation.
Johansson’s reaction and the company’s rationale comes amid rising concern that AI could threaten the livelihood or even replace human workers, up to and including celebrated Hollywood professionals. OpenAI and similar companies have been buffeted in recent months by lawsuits and legal threats from creators, artists, authors, and media companies for allegedly using their material to train their models without permission.
The issue was of central importance to SAG-AFTRA during last year’s Hollywood strikes. The contract the actors’ union eventually cemented with the studios and streamers includes measures restricting the use of artificial intelligence in film and television, from voiceovers to full-body scans.
