OpenAI apologises to actress Johansson over AI voice similarity

SAN FRANCISCO — OpenAI chief Sam Altman publicly apologised on Tuesday to Scarlett Johansson after the movie star said she was “shocked” by a new synthetic voice released by the ChatGPT maker that sounds “eerily similar” to her.

At issue is “Sky”, one of several voices OpenAI unveiled last week with the release of its higher-performing and even more humanlike GPT-4o artificial intelligence technology.

In a demo, Sky was at times flirtatious and funny, capable of seamlessly jumping from one topic to the next, unlike most existing chatbots.

The technology — and sound of the voice — quickly drew similarities to the Johansson-voiced AI character in the 2013 film “Her”.

Altman has previously pointed to the Spike Jonze-directed movie — a cautionary tale about the future in which a man falls in love with an AI chatbot — as inspiration for where he would like AI interactions to go.

He furthered speculation last week with a single-word post on X, formerly Twitter, saying “her”.

Johansson on Monday expressed outrage at the new voice, saying in a statement she was “shocked, angered, and in disbelief that Mr Altman would pursue a voice that sounded so eerily similar to mine that my closest friends and news outlets couldn’t tell the difference”.

She said Altman had offered in September to hire her to work with OpenAI to create a synthetic voice, saying it might provide people comfort engaging with AI.

OpenAI said on Monday on X that it was working to “pause” Sky, with a company blogpost explaining that “Sky’s voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice”.

In a statement shared on Tuesday with AFP, Altman said the company “cast the voice actor behind Sky’s voice before any outreach to Ms Johansson”.

“Out of respect for Ms Johansson, we have paused using Sky’s voice in our products. We are sorry to Ms Johansson that we didn’t communicate better.”

 The company, in its blogpost, explained that it worked with professional voice actors on synthetic voices it named Breeze, Cove, Ember, Juniper and Sky.

It began working to cast the voice actors in early 2023, “carefully considering the unique personality of each voice and their appeal to global audiences”.

Some of the characteristics sought were “a voice that feel timeless” and “an approachable voice that inspires trust”, the company said.

The five final actors were flown to San Francisco to record in June and July, it said, with their voices launched into ChatGPT on September 25, 2023.

“To protect their privacy, we cannot share the names of our voice talents,” OpenAI said.

“We believe that AI voices should not deliberately mimic a celebrity’s distinctive voice.”

So far in the AI frenzy, most tech giants have been reluctant to overly humanise chatbots.

Microsoft Vice President Yusuf Mehdi told AFP his company, which has a partnership with OpenAI, sought to make sure that AI was not “a he or a she”, but rather a “unique entity”.

“It should not be human. It shouldn’t breathe. You should be able to... understand [it] is AI,” he said.

Just days ago OpenAI said it disbanded a team devoted to mitigating the long-term dangers of artificial intelligence.

OpenAI began dissolving the so-called “superalignment” group weeks ago, integrating members into other projects and research.

Company co-founder Ilya Sutskever and superalignment team co-leader Jan Leike announced their departures from the San Francisco-based firm last week.