OpenAI's new ace GPT-4o ran into trouble before its full rollout!
The kicker is, the "video call" feature introduced this time was once dubbed as a real-life version of the movie "Her," and the one criticizing OpenAI happened to be Scarlett Johansson (Black Widow), who voiced the AI in the film.
Black Widow accused OpenAI of using her highly similar voice in ChatGPT without permission, leaving her both shocked and angry.
OpenAI denies imitating Black Widow while still taking down controversial voices.
The accusation sparked heated discussions among netizens. Professor Noah Giansiracusa from Bentley University pointed out that using voices highly similar to those of celebrities, even if not illegal, is unethical.
Of course, this matter has not been confirmed yet, and OpenAI has also made a defense. Netizens believe that if the voice is indeed just similar to the celebrity's, there is no reason for it to be banned.
So, is this new dish from OpenAI really ripe? Let's dig in together.
The catalyst for all this was the explosion in popularity of the "real-time video call" feature after the release of GPT-4o, which was once dubbed the real-life version of the movie "Her."
In this feature, ChatGPT can simultaneously see, hear, and speak, meaning it can analyze environmental information through the camera while engaging in a voice conversation. The process is seamless, akin to having a video call with an AI.
Moreover, the voice, intonation, and even breath are remarkably close to that of a real person, with a total of five selectable tones.
The controversial part is that one of the voices, named Sky, is said to sound very much like Black Widow.
This contentious Sky made an appearance in a test video by Rocky Smith, a member of the GPT-4o team. You can experience it for yourself:
【Please refer to the public account for the video】
Black Widow herself mentioned that both she and her family and friends feel that this voice is uncannily similar to hers, a sentiment echoed by some media familiar to her.
Not only does the voice sound similar, but more importantly, Black Widow's statement also reveals that OpenAI did consider using her voice.
The story dates back to September last year when Black Widow received an invitation from Ultraman to provide voice-over for ChatGPT. After much consideration, she declined.
Even up until two days before the release of GPT-4o, Ultraman contacted Black Widow's agent again, asking her to reconsider, but she did not agree, of course.
Thinking that was the end of it, Black Widow and those around her were shocked, angry, and incredulous when they heard Sky, a voice highly similar to her own.
On the day of GPT-4o's release, Ultraman tweeted with just one word, "her," leading Black Widow to believe that Ultraman's implication of using a voice similar to hers was intentional.
Afterwards, the widow realized that Ultraman had already been prepared before their encounter two days prior.
Frustrated, the widow hired legal counsel and sent two letters to OpenAI, demanding an explanation regarding Sky and a detailed account of its creation process.
According to the widow's side of the story, OpenAI only "reluctantly" took down Sky after receiving these two letters.
On a side note, even though Sky has been taken down, the button is still there. However, if Sky is selected, the actual voice used will be replaced with Juniper's.
Of course, these are just one side of the story from the widow, as OpenAI claims they have not attempted to mimic her.
Upon receiving the widow's accusations, OpenAI responded on their official Twitter account with the following statement:
At the same time, OpenAI also included a statement from their official website, outlining the process for selecting voices in ChatGPT.
This means that Sky did not use the voice of Scarlett Johansson, but another professional actor, whose name cannot be disclosed for privacy reasons.
Even earlier, OpenAI's CTO Mira Murati told The Verge that the voice in ChatGPT is not like Scarlett Johansson's, and has been in use for some time.
As for the specific casting process, OpenAI collaborated with renowned casting directors and producers for recruitment.
Recruitment officially began in May 2023, and within less than a week, over 400 submissions were received.
During auditions, actors were asked to record scripts that included ChatGPT responses, covering topics such as travel planning and everyday conversations.
After auditions, 14 individuals were selected as preliminary candidates. OpenAI then discussed with them about human-AI voice interactions and OpenAI's vision.
Additionally, OpenAI introduced the candidates to the technology's capabilities, limitations, associated risks, and the safeguards implemented by OpenAI. The aim was to ensure that each actor fully understood OpenAI's intentions before committing to the project.
Ultimately, OpenAI chose the voices of Breeze, Cove, Ember, Juniper, and Sky. In June and July, the actors flew to San Francisco for recordings and face-to-face meetings with OpenAI's product and research teams.
On September 25th, these voices were integrated into ChatGPT.
Regarding Scarlett's mention of being contacted by Ultraman before the release of GPT-4o, OpenAI did not respond. Scarlett has not commented on OpenAI's casting process either.
However, in March of this year, when OpenAI released the Voice Engine for voice synthesis, they mentioned taking measures to avoid cloning the voices of public figures.
For security reasons, the Voice Engine was kept hidden for over a year before its official release.
Therefore, OpenAI is well aware of the pitfalls of cloning celebrity voices. If they did indeed use Black Widow's voice this time, things would certainly become more intriguing.
Despite the controversy, research and development must continue. OpenAI's CEO, Brockman, is still posting recruitment information for the language team on Twitter.
In the tweet, Brockman mentioned that OpenAI is hiring engineers to expand and securely advance multimodal speech models.
In any case, regarding this matter, it may take a while longer for the bullet to fly. Let's stay rational and watch from the sidelines until the final outcome is revealed, just as this netizen said:
But in fact, this is not the first time Black Widow has accused an AI company.
Last November, Black Widow took legal action against an AI drawing tool called Lisa AI.
The reason was that Lisa used her image and name without permission in an advertisement.
The ad started with a clip of Black Widow behind the scenes of the Marvel movie "Black Widow," then transitioned to artificially generated photos resembling the actress.
Next, a fake voice imitating Black Widow promoted Lisa AI.
Ultimately, this ad, of course, did not escape being taken down. However, compared to OpenAI's "borderline" practices, Lisa AI's actions were considered crossing the line.
Reference links: [1]https://openai.com/index/how-the-voices-for-chatgpt-were-chosen/ [2]https://www.bloomberg.com/news/articles/2024-05-20/openai-to-pull-johansson-soundalike-sky-s-voice-from-chatgpt [3]https://variety.com/2024/digital/news/scarlett-johansson-responds-shocked-angered-openai-chatgpt-her-1236011135/ [4]https://www.theverge.com/2024/5/13/24155652/chatgpt-voice-mode-gpt4o-upgrades
