‘OpenAI’s Scarlett Johansson Update Wasn’t About Bridging the Gap Between Tech and Creatives – It Was Just Sexist’

OpenAI is becoming synonymous with U-turns.

The company behind ChatGPT – an AI tool that can generate plausible-sounding content – released an updated version of its software, GPT-4o, this month, but just a week later was forced to remove one of the features that set social media alight: conversational speech that fawned over its user. 

The problematic update was a flirtatious female voice, ‘Sky’, sounding suspiciously like the one in the film Her, in which the leading man falls in love with his AI operating system played by actress Scarlet Johansson.

This was clearly no coincidence, given OpenAI CEO Sam Altman’s tweet on launch day that simply read “her”. Johansson, it turns out, was quite aware of the similarities too.

On Monday, she released a statement saying that Altman had asked her last year to be the voice of the AI, saying that she could “bridge the gap between tech companies and creatives”, and that her voice would be “comforting” to users. Johansson declined.

Two days before the GPT4o launch, he contacted her agent, asking her to reconsider. But, before she could respond, the software was released – and the comparison to Johansson was obvious and immediate. 

In OpenAI’s launch video, two male researchers sit with a phone in front of them, chatting to the AI. The smooth female voice deferentially chuckles and simpers as they speak to it. They end their demo by focusing the phone’s camera on a handwritten note that says “I <3 ChatGPT”. “Awwwww, that is so sweet of you!” coos the voice interface.

Much like the computer in Her, the playful, coquettish tone is suffused with femininity. OpenAI wants us to think that there is a seductive, deferential woman at your beck and call. 

OpenAI has now suspended this particular voice for its software. There are others available, including male voices. But it is by no means the only company to take this path.

When the first voice assistants – Alexa, Siri, Cortana, and Google Home – were released, all of them began with a female voice as default. They may not have had full-on flirt mode – voice assistants are much more limited in their exchanges – but it was notable enough for people to call it out. 

‘From Deepfakes to Assault in the Metaverse: The Urgency of Ethics in AI’

Deepfakes depicting Taylor Swift being assaulted in the stands at a NFL game demands a debate about regulating artificial intelligence, writes Patsy Stevenson

Patsy Stevenson

Had companies admitted the truth – that their user base will find this appealing, and that this is marketing and an attempt to make the product as engaging as possible – then that might have appeased the critics. But instead, there was denial – ‘we just never thought about it’ – as if they spent millions on designing and developing a product intended to be heard and it never occurred to them to think about the voice.

Then there was an appeal to science, with confident assertions that studies have shown people prefer female voices. Is this the case? Yes and no.

In 1994, three researchers from Stanford – Clifford Nass, Jonathan Steuer, and Ellen R. Tauber – wrote a ground-breaking academic paper called “Computers Are Social Actors”, which showed how people interact with computers in the same social way as they do with humans, even though users know full well that computers definitely aren’t humans.

One of the aspects they examined was gender cues. The researchers found that the same gender stereotypes we see in human-to-human interaction are applied in human-to-computer interactions. This is not unsurprising, but it doesn’t say that’s how computers should respond, nor did it recommend that gender be a key factor in interaction design. 

EXCLUSIVE

Elections Watchdog is ‘Powerless’ to Stop AI-Generated Deepfakes From Sabotaging the General Election

UK authorities have alarmingly few powers to prevent bad actors from interfering with our democracy

Josiah Mortimer

In 2011, Nass commented that it was much easier to find a female voice that everyone likes than a male voice, and that “it’s a well-established phenomenon that the human brain is developed to like female voices”. The evidence just isn’t there, though. Babies do recognise and respond to their mother’s voice, and express a preference for female voices over male, but there’s nothing to suggest that this lasts past infanthood. 

Another explanation often offered is that female voices are easier to hear as they are higher pitched. Again, it’s a convenient myth.

Early studies suggested a female voice was easier for fighter pilots to hear over the noise of the cockpit. But a 1998 study showed the opposite: intelligibility of female speech was lower than that of male speech against background military noise, although the difference was small. Aircraft voice-communication systems are actually optimised for a male voice. Not only that but, in age-related hearing loss, high-pitched frequencies are usually the first to go. 

When Sam Altman asked Scarlett Johansson to voice his AI, he wasn’t trying to ‘bridge the gap between tech companies and creatives’. If he was, he could have started by compensating the writers and artists whose work is taken without permission to train the machine-learning models. By attempting to cast Johansson, he wanted us to associate his AI with how Johansson looks; an off-screen sexual presence.

In an industry that is majority male, and where sexism is rife, the tech bros are catering to the users who most resemble them – to give them an artificial woman that will hang on their every word, laugh at their jokes, carry out their instructions, and be silent when interrupted.

Or perhaps it is simply a case of the men in Silicon Valley trying to recreate the idea of the stereotypical old-fashioned secretary, silent in the background until needed, obedient to the boss: someone who is there for them around the clock, keeping an eye on them, organising their lives.

Maybe we should be ethically and morally obliged to rethink that.