Why Can't OpenAI Make ChatGPT Sound Like Scarlett Johansson?
OpenAI didn't name the new ChatGPT voice 'ScaRleTt JoHAnsSoN' but might as well have.
(Photo by Monica Schipper/Getty Images for David Yurman)
OpenAI recently previewed its new ChatGPT voice — “Sky” — and several folks immediately noticed it sounded eerily like Scarlett Johansson.
So that’s where you go when you’re trapped in the Soul Stone.
Raising The Bar in Bar Prep
Specifically, the new ChatGPT sounded like Scarlett Johansson from Her, where she played an artificial intelligence-driven assistant in a movie expressing the threat of emotional damage that a sufficiently developed AI could exact upon vulnerable human users.
Apparently, OpenAI’s Sam Altman loves the movie and desperately wants his product to be inspired by it, forcing us to yet again repeat:
In a statement released yesterday, Johansson claims Altman asked her to provide the voice back in September — and then again two days before releasing the demo — and she declined both times. Then OpenAI released the very, very similar sounding product anyway and Johansson called her lawyers:
Sponsored
Raising The Bar in Bar Prep
Trust The Process: How To Build And Manage Workflows In Law Firms
How Savvy Lawyers Build Their Law Firm Rate Sheet
Trust The Process: How To Build And Manage Workflows In Law Firms
As a result of their actions, I was forced to hire legal counsel, who wrote two letters to Mr. Altman and OpenAl, setting out what they had done and asking them to detail the exact process by which they created the “Sky” voice. Consequently, OpenAl reluctantly agreed to take down the “Sky” voice.
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. I look forward to resolution in the form of transparency and the passage of appropriate legislation to help ensure that individual rights are protected.
If OpenAI built this voice off sampling Johansson’s voice — which is what she’s trying to surmise through counsel — then someone likely has a copyright claim. Whether it would be Johansson herself or the studios she performed the sampled work for raises more complex questions, but the company would’ve taken copyrighted work and used it without permission. Unlike the cases being brought by other creatives, arguing that AI “training” on their work constitutes a copyright claim, reappropriating clips wholesale is obviously a problem.
Altman feeling that he had to continue to try to find a way to pay her up to a couple days before the demo… well, it’s not great evidence for OpenAI. On the other hand, maybe he’ll claim that they only got so far with non-infringing imitation and he still wanted to hire her to build out the system.
If Sky didn’t use any of Scarlett’s audio samples, there’s not much of a case. Kentucky Law professor Brian Frye thinks the mere sound of a voice — particularly if it’s just an impersonator — isn’t going to cut it here.
Sponsored
How To Build And Manage Your Law Firm Rate Sheet
Back To Basics: 6 Tech Tips Every Small Firm Needs To Hear
But, to add another wrinkle, what if OpenAI didn’t just hire “a lady that sounded like her” but instead employed some sort of algorithmic autotune wrapper that took an impersonation from another actress and smoothed out any minor differences to match the timbre of the Her character? It’s a completely achievable technological feat recently used in an effort to build a historically accurate recreation of the Brown v. Board oral argument. Does that change the situation? Is it then outright stealing copyrighted aspects or is it back to the same “training” scenario that writers face?
The most likely route she’d have to take is a publicity rights claim against the company… and that seems like an uphill battle.
The landmark cases in this area are Midler v. Ford and Waits v. Frito-Lay, a pair of Ninth Circuit decisions that found personality rights in protecting a voice from commercial impersonation. But neither case is quite on point here.
First off, both of those cases involved companies using impersonators in endorsements. Would using her as the model for the platform’s interface create that sort of confusion? Especially if OpenAI tacked on some well-tested, boilerplate “any resemblance is purely coincidental” language?
As endorsements, those cases also involved companies suggesting that the celebrities themselves approved of the product. Perhaps a subtle difference, but OpenAI’s product doesn’t sound like Johansson’s natural conversational tone, but like her character in a specific movie. How much of the Her AI voice is Johansson as opposed to the writing and direction that collaborated to create that characterization? And how much is OpenAI trying to copy Johansson as opposed to that character?
If they’d tried to copy that grating New Jersey accent from whatever her name was in Don Jon we’d (a) immediately swear off ChatGPT forever, but also (b) probably not be talking about personality rights. It’s the fact that this character was, while not the same, but close to the performer’s normal voice that seems to be strengthening this argument. But to what extent is cashing in on a characterization akin to cashing in on a performer? If they made a sequel to Her and told the new actress to attempt an impersonation, no one would think Johansson could sue the studio… yet that would seemingly be on the table if the law extends publicity rights to copying characters.
This isn’t merely academic. Rick and Morty parted with its co-creator who happened to voice both iconic characters. The show hired impersonators to keep producing episodes. Bob’s Burgers replaced the Jimmy Pesto character after the previous actor, um, stormed the Capitol.
Also, while both Bette Midler and Tom Waits have taken turns acting to some acclaim, but their celebrity remains rooted in their singing. Scarlett did sing the snake song in that Jungle Book remake, but fundamentally doesn’t make her career in music — a point indirectly but comedically made by her husband on this weekend’s SNL:
Facetiousness aside… to what extent is an actress, as opposed to a singer, defined by voice? James Earl Jones could, perhaps, make a claim that copying his voice intrudes upon trading on his celebrity. How many other actors can you honestly say are famous principally for their voice? And, again, the natural qualities of their voice as opposed to a character that they play?
Fine, but no one really wants their virtual assistant to sound like the late Gilbert Gottfried.
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter if you’re interested in law, politics, and a healthy dose of college sports news. Joe also serves as a Managing Director at RPN Executive Search.
