Scarlett Johansson, OpenAI, and the sexism of ChatGPT’s “Her” voice

If a man instructed you his favourite sci-fi film is Her, then launched an AI chatbot with a voice that sounds uncannily just like the voice from Her, then tweeted the only phrase “her” moments after the discharge… what would you conclude?

It’s cheap to conclude that the AI’s voice is closely impressed by Her. 

Sam Altman, the CEO of OpenAI, did all the issues talked about, and his firm not too long ago launched a brand new model of ChatGPT that talked to customers in a flirty feminine voice — a voice that distinctly resembles that of Scarlett Johansson, the actress who voiced the AI girlfriend within the 2013 Spike Jonze film Her. 

Now, Johansson has come ahead to object, writing in an announcement that the chatbot’s voice sounds “so eerily just like mine that my closest mates and information shops couldn’t inform the distinction.” 

Altman’s response? He claims the voice “shouldn’t be Scarlett Johansson’s and was by no means meant to resemble hers.” 

That’s, at first blush, an absurd declare. 

Whereas the voice could not actually be skilled on or copied from Johansson’s — OpenAI says it employed one other actress — there’s loads of proof to counsel that it might need been meant to resemble hers. Along with Altman’s professed love of Her and his “her” tweet, there are the brand new revelations from Johansson: Altman, she says, reached out to her agent on two separate events asking for her to voice the chatbot. 

When the primary request got here in final September, Johansson stated no. A second request got here in two days earlier than the brand new chatbot’s demo, asking her to rethink. “Earlier than we may join, the system was on the market,” Johansson acknowledged, including that she had employed a lawyer to demand an evidence from Altman. 

OpenAI revealed a weblog publish saying that it went via a months-long course of to seek out voice actors final yr — together with the voice for “Sky,” the one many individuals discover just like Johansson’s — earlier than introducing some voice capabilities for ChatGPT final September. In response to Altman, “We forged the voice actor behind Sky’s voice earlier than any outreach to Ms. Johansson.” September, thoughts you, is the month that Johansson says Altman first requested to license her voice. 

If OpenAI did certainly forged the actor behind Sky earlier than any outreach to Johansson, it nonetheless doesn’t essentially observe that Sky’s voice was by no means meant to resemble Johansson’s. Nor does it essentially observe that the AI mannequin behind Sky was solely ever fed the employed actor’s voice, with no use in anyway being made from Johansson’s voice. I raised these inquiries to the corporate. OpenAI didn’t reply to a request for remark in time for publication. 

OpenAI took down Sky’s voice “out of respect for Ms. Johansson,” as Altman put it, including, “We’re sorry to Ms. Johansson that we didn’t talk higher.” 

But when OpenAI didn’t do something mistaken, why would it not take down the voice? And the way a lot “respect” does this apology actually convey, when Altman insists in the identical breath that the voice has nothing to do with Johansson? 

From Apple’s Siri to Amazon’s Alexa to Microsoft’s Cortana, there’s a cause why tech corporations have been giving their digital assistants pleasant feminine voices for years. From a enterprise perspective, it’s good to offer your AI that voice. It probably improves your organization’s backside line. 

That’s as a result of analysis reveals that when individuals need assistance, they like to listen to it delivered in a feminine voice, which they understand as non-threatening. (They like a male voice on the subject of authoritative statements.) And corporations design the assistants to be unfailingly upbeat and well mannered partially as a result of that form of habits maximizes a person’s want to maintain participating with the system.  

However the design alternative is worrying on an moral stage. Researchers say it reinforces sexist stereotypes of ladies as servile beings who exist solely to do another person’s bidding — to assist them, consolation them, and plump up their ego. 

In response to Johansson, conveying a way of consolation was precisely Altman’s objective in making an attempt to license her voice 9 months in the past. 

“He instructed me that he felt that by my voicing the system, I may bridge the hole between tech corporations and creatives and assist shoppers to really feel snug with the seismic shift regarding people and AI,” Johansson wrote. “He stated he felt that my voice can be comforting to individuals.” 

It’s not simply that Johansson’s breathy, flirty voice is soothing in itself. Johansson voiced Samantha, the AI girlfriend within the romance Her, a narrative that’s all about how an AI may join with, consolation, and enliven a lonely human. Notably, Samantha was additionally way more superior than something fashionable AI corporations have put out — so superior, in actual fact, it evolves past its human person — so associating the brand new ChatGPT with the movie in all probability helps as effectively.

There’s a second layer right here, one which has to do with a girl’s consent. Regardless of Johansson’s clear “no” to Altman’s request final yr, he used a Johansson-like voice after which, when she complained, instructed the world that the actress is mistaken in regards to the voice being meant to resemble hers. 

I wasn’t certain what to name that, so I requested ChatGPT about any such state of affairs extra usually. Right here’s how the chatbot replied:

The Johansson controversy is the most recent in a string of occasions inflicting individuals to lose belief in OpenAI — and particularly in its CEO Altman. 

Final yr, artists and authors started suing OpenAI for allegedly stealing their copyrighted materials to coach its AI fashions. In the meantime, specialists raised the alarm about deepfakes, which have gotten extra worrisome by the day because the world approaches main elections.

Then, final November, OpenAI’s board tried to fireplace Altman as a result of, as they put it then, he was “not persistently candid in his communications.” Former colleagues then got here ahead to describe him as a manipulator who speaks out of either side of his mouth — somebody who claims that he desires to prioritize deploying AI safely, however contradicts that in his behaviors. Since then, workers have been more and more coming to the identical conclusion, to the purpose that some are leaving the corporate.  

“I step by step misplaced belief in OpenAI management,” ex-employee Daniel Kokotajlo instructed me, explaining why he give up his job final month.

“It’s a means of belief collapsing little by little, like dominoes falling one after the other,” one other particular person with inside information of the corporate instructed me final week, talking on situation of anonymity.

Some workers have prevented talking out publicly as a result of they signed offboarding agreements with nondisparagement provisions upon leaving. After Vox reported on these agreements, Altman stated the corporate has been within the course of of fixing them. However the public may effectively ask: Why would OpenAI have had such restrictive provisions if it wasn’t doing something that it was eager to maintain out of the general public eye? 

And at a time when a number of of OpenAI’s most safety-conscious workers are leaping ship as a result of they don’t belief the corporate’s leaders, why ought to the general public belief them?

In reality, in keeping with a brand new ballot from the Synthetic Intelligence Coverage Institute, practically 6 in 10 People say the discharge of the souped-up ChatGPT makes them extra apprehensive about AI’s development, whereas simply 24 % say it makes them excited. What’s extra, 52 % of People now maintain an unfavorable opinion of OpenAI.

At this level, the burden of proof is on OpenAI to persuade the general public that it’s worthy of belief.