What that you must know
- Scarlett Johansson says she was approached by OpenAI final yr about utilizing her voice for a ChatGPT voice assistant.
- Although Johansson didn’t conform to the proposition, OpenAI shipped its GPT-4o mannequin with a voice known as “Sky” that sounds fairly just like Johansson‘s.
- After authorized stress, OpenAI eliminated Sky from GPT-4o, and mentioned that the voice was not based mostly on Johansson‘s.
- Nonetheless, in attempting to make use of a pleasant and welcoming voice to make AI really feel extra comforting, OpenAI ended up doing the other.
OpenAI made waves final week when it introduced GPT-4o, a multimodal AI mannequin that could be probably the most superior and futuristic one we’ve seen so far. It seems like a human, can work together with customers by way of imaginative and prescient and audio, and is educated. OpenAI ended up beating Google to the punch, and GPT-4o appears extra superior than Undertaking Astra, which Google previewed at Google I/O 2024.
However one of many voices OpenAI selected for GPT-4o has been drawing consideration on-line for all of the improper causes. First, some customers on social media identified they thought the voice “Sky” was overly flirty and sultry to the purpose it was unsettling. Then, folks began noticing the similarities between the voice of Sky and that of Scarlett Johansson, the award-winning actress. Now, it seems that might have been intentional.
To be clear, OpenAI denies that the voice of Sky was based mostly on Johansson and even launched a weblog publish explaining how the voices had been chosen. Nonetheless, Johansson put out a scathing assertion telling the story of how OpenAI approached her about formally voicing GPT-4o, which she declined. After going through authorized stress from Johansson’s legal professionals, the corporate eliminated the Sky voice choice from GPT-4o.
As distressing as this example is, it’s nearly ironic. OpenAI’s CEO Sam Altman informed Johansson that her voice, being the official voice of ChatGPT, could be extra comforting to customers. And but, by releasing a voice so just like Johansson‘s with out her permission, Altman and OpenAI ended up completely encapsulating every part that makes folks uncomfortable about AI.
Did OpenAI steal Scarlett Johansson‘s voice?
Although OpenAI says that it sought out skilled voice actors for GPT-4o and didn’t search somebody who seemed like Johansson particularly, the proof may inform a distinct story. It begins in September 2023, in keeping with Johansson, when OpenAI’s Altman reached out about hiring her as a voice actor for ChatGPT.
“He informed me that he felt that by my voicing the system, I may bridge the hole between tech corporations and creatives and assist customers to really feel comfy with the seismic shift regarding people and Al,” she mentioned in an announcement to NPR’s Bobby Allyn. “He mentioned he felt that my voice could be comforting to folks.”
Johansson finally determined to not go ahead with voicing GPT-4o. Nonetheless, it’s simple to listen to her resemblance within the Sky voice that ended up being demoed and shipped with the AI mannequin. To say that Johansson was displeased with the end result could be an understatement.
Assertion from Scarlett Johansson on the OpenAI state of affairs. Wow: pic.twitter.com/8ibMeLfqP8May 20, 2024
“We consider that AI voices mustn’t intentionally mimic a celeb’s distinctive voice— Sky’s voice isn’t an imitation of Scarlett Johansson however belongs to a distinct skilled actress utilizing her personal pure talking voice,“ OpenAI mentioned in a weblog publish.
The entire purpose that OpenAI needed a voice like Johansson’s, as Altman is alleged to have informed her, is to make AI extra comforting. Individuals could also be extra scared about AI than they’re enthusiastic about it. Particularly these in artistic industries are discovering that AI is getting used to automate writing, visible artwork, music, and different mediums. This isn’t one thing distinctive to OpenAI — Apple not too long ago got here beneath fireplace and apologized for an commercial that actually noticed devices being crushed into items and changed with an iPad.
By utilizing her likeness in a GPT-4o voice with out her permission — whether or not deliberately or unintentionally — OpenAI ended up validating the discomfort related to AI that it was desperately attempting to handle. Creatives, from actors and actresses to writers and photographers, are fearful about being changed by AI. The concept that OpenAI may have mimicked Johansson‘s voice for GPT-4o is precisely the sort of factor that worries and alarms folks in artistic industries.
“After I heard the launched demo, I used to be shocked, angered and in disbelief that Mr. Altman would pursue a voice that sounded so eerily just like mine that my closest associates and information shops couldn’t inform the distinction,” Johansson wrote, explaining that she requested OpenAI to indicate the way it developed the Sky voice. “In a time after we are all grappling with deepfakes and the safety of our personal likeness, our personal work, our personal identities, I consider these are questions that deserve absolute readability.”
We shouldn’t need AI to sound this human
Except for the unsettling concept that an organization may rip off an actress‘ voice after disagreeing with a deal, there are different the reason why we don’t need AI voices to sound like Sky. All of OpenAI’s GPT-4o voices, and particularly Sky, sound very human-like. This can be a downside, as a result of there’s a excessive degree of belief and familiarity folks have with human voices. While you discuss to a voice assistant like Siri or Alexa, it’s clear that you just’re speaking to — for lack of a greater phrase — a robotic. After having a dialog with GPT-4o, that degree of readability received’t at all times be at the back of your thoughts.
Proper now, AI fashions have an issue, and it’s that they confidently state their solutions as reality even when they’re blatantly improper. Individuals nonetheless find yourself believing AI responses as true regardless of the array of warnings that come together with them. As voices for AI fashions develop into extra human-sounding, this downside will solely worsen. It’ll be simple for the typical person of an AI software to consider what’s being mentioned because of the welcoming human voice it makes use of.
In attempting to make folks extra comfy with the way forward for AI, OpenAI ended up making it really feel extra dystopian. We shouldn’t need AI to sound as human as GPT-4o, and there are many the reason why. It may foster an unwarranted degree of belief between customers and AI fashions, in addition to put creatives like Johansson In a precarious place.