- Added Scarlett Johansson's statement
- Added Washington Post sources
Update from May 23, 2024:
The Washington Post has obtained documents from OpenAI and statements from sources indicating that the voice behind Sky was recorded months before Altman's contact with Johansson. Neither Johansson nor the movie "Her" were ever mentioned in the briefings for the voice, according to the sources.
The conditions in the job posting were that the actress must not be unionized, and the voice must sound between 25 and 45 years old and be "warm, engaging [and] charismatic."
Moreover, the two voices do not sound particularly similar. The actress herself says she feels personally attacked by the criticism, as it only concerns her natural voice. She has never been compared to Johansson in her private life.
Why Altman wanted to contact Johansson again so close to the presentation remains unclear. However, the harsh criticism of OpenAI and Altman in some publications and social media is inappropriate given the facts.
Update from May 21, 2024:
Scarlett Johansson is "shocked, angered and in disbelief" by OpenAI's use of a voice that sounds very similar to her own, NPR reports.
OpenAI CEO Sam Altman allegedly made several unsuccessful attempts to license Johansson's voice for ChatGPT, most recently two days before the launch of GPT-4o. OpenAI did not mention this in its announcement of the Sky pause (read below).
Johansson turned down the deal, and OpenAI reportedly licensed a similar-sounding voice. Altman explains that the voice, which is now paused, was cast by OpenAI before the first contact with Johansson. The fact that the voice is now on hold is a gesture of respect to the actress, he said.
The US actress still sees this as a personal affront and a threat to her privacy rights.
"I 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 could not tell the difference," she said in a statement to NPR.
At a time when generative AI and deepfakes bring the protection of one's image, work, and identity to the forefront, such issues need "absolute clarity," Johansson said.
Scarlett Johansson was one of the most high-profile victims of the first wave of deepfakes, which began in 2017. Her face was deepfaked into numerous porn videos. In late 2023, she sued AI startup Lisa AI for allegedly using her name and image for online advertising without her consent.
Original article from May 20, 2024:
Scarlett Johansson's voice doppelganger axed from OpenAI's ChatGPT, but it wasn't her
OpenAI has stopped using one of five voices for its ChatGPT AI chatbot because it supposedly sounded too similar to actress Scarlett Johansson's voice, the company said.
The voice, called "Sky," was one of the "carefully selected" voices ChatGPT uses to give itself a human touch in the smartphone app.
OpenAI said the goal in selecting the voices was to find a timeless and "warm, engaging, trustworthy, charismatic voice" that would give ChatGPT warmth.
According to OpenAI, the aim in selecting the voices was to find a "timeless, confidence-inspiring and natural voice with a rich sound" that would give ChatGPT warmth and charisma.
The company emphasized that it doesn't intend to imitate the voices of celebrities. "Sky’s voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice," OpenAI explained.
In the science fiction film Her, Scarlett Johansson voiced the AI character "Samantha," with whom the main character falls in love. According to OpenAI, Her is a role model for the advanced AI assistance the company aims to create. While introducing the new AI voice synthesis last Monday, OpenAI CEO Sam Altman wrote "her" on X.
For privacy reasons, OpenAI will not disclose the name of the actress who provided Sky's voice. It is unclear why OpenAI removed the voice, even though it is demonstrably not Scarlett Johansson's, is legally licensed, and at best bears a resemblance to Johansson.
To select the voices, OpenAI says it worked with renowned casting agencies and directors for five months. From more than 400 submissions, 14 actors were selected, and OpenAI met with them in person to explain its vision for human-AI interaction.
The actors whose voices were ultimately selected flew to San Francisco for recording sessions in June and July 2023. Their voices were then integrated into ChatGPT in September 2023, the company said.
The new voice feature, which was unveiled last week, will be rolled out to paying ChatGPT Plus users in the coming weeks. The chatbot will then be able to smoothly handle interruptions, respond faster, effectively manage group conversations, filter out background noise, and adapt to the tone of voice, according to OpenAI.
The current version of the voice is still the old version from the fall of 2023, although it is already running with GPT-4o in the background.
Correction: Clarified that Altman said OpenAI hired the Sky actress before Johansson turned down the offer.
