Chelsea Bonner uploaded a swag of images of herself to an artificial intelligence (AI) app, asked it to dress her in lingerie and, hey presto, out popped "cartoonish anime porn version Chelsea".
AI Chelsea had none of the wrinkles or curves that Bonner, 50, embraces as part of the confident, savvy woman she is, a woman who has been at the forefront of promoting diversity through her modelling agency for 21 years.
This hyper-sexualised AI Chelsea was about size eight, with flawless skin, incredible abdominal muscles and faker than a knock-off handbag.
"It's sort of like the anime version of what men think women should look like," Bonner tells Australian Story. Even when Bonner prompted the app multiple times to put more flesh on the bones, more pores in the skin, AI Chelsea still looked foreign to the real one.
This was no idle experimentation for Bonner. She had launched herself into the world of generative AI to better understand – and try to combat – what she believes is a serious and disturbing threat to the fashion industry and many others.
Instead of paying models, photographers, hairstylists, make-up artists, and other trades that help in the making of fashion images, companies could use AI to create photographs of a "realistic fake" model sitting in an exotic location, dressed in their apparel. It's already happening.
The more Bonner explored, the more her concerns grew beyond her industry. She worried about what these computer-generated images of the "perfect" woman would do to the self-confidence of the real-life ones, especially girls. She was aghast that AI models were being presented as real, luring people to porn or dating sites, or to buy goods.
What kind of future were we blindly marching into?
"It's a cataclysmic step backwards," Bonner says. "When I first started, every [model] … was a five foot nine, white, normally blonde or maybe brunette woman, and they were always a size eight to a maximum of size 10.
"Over the last 20 years, we've made huge inroads into size diversity, into ethnic inclusion, into the use of models with different abilities and disabilities. It's been a real fight [and] I just feel like we're going back in time."
So, Bonner is getting back into the ring, spearheading a petition to lobby the federal parliament to act. It carried the statistic that just in the past year, more than 15 billion images have been AI generated and are circulating on social media and online shopping platforms.
"Without rules, without governance, this technology can very quickly get out of control, and we just can't allow that to happen," Bonner says.
We're at a tipping point, she says, and what makes us human is in the balance.
Has your image been used to train AI?
Bonner grew up surrounded by beauty. Her parents were the "It" couple of the 70s; her father, Tony Bonner, played the ruggedly handsome helicopter pilot Jerry in the Australian television classic, Skippy, and her mother, Nola Clark, was one of the top fashion models of the day.
As a girl, Bonner loved the world of fashion, the art of creating a "look" for the camera, and decided to try modelling. But she wasn't a size eight.
"She couldn't get anyone to take her seriously as a big girl," recalls Clark.
"That's what they used to call them in those days, big girls, which is pretty disgusting."
So, Bonner created her own agency, specialising in plus-size models such as Robyn Lawley, who got her big break posing for the covers of Vogue Italia and Vogue Australia under Bonner's management.
Lawley, a size 14, has watched as Bonner paved the way for models such as herself to be used in fashion shoots, reflecting the 80 per cent of women over size 10. "Chelsea is always trying to fix things that are broken," she says.
The duo has united again, says Lawley, "fighting [for] diversity in a whole different realm" — the land of generative AI, where fake models can vary from cartoonish, big-breasted types to sophisticated creations that are very close renditions of the real thing.
Jeannie Paterson, a professor of law and director of the Centre for AI and Digital Ethics at the University of Melbourne, says while AI has been around for some time, the gamechanger in recent years has been generative AI, which can create reasonably realistic images that are easy to manipulate.
"Generative AI … is trained on lots and lots of images that are usually gathered from the internet … but then is able to create its own images," Paterson says.
"The reason that creates difficulties for law … is that it's not a straight copying of the images that it's found, which means it's not clear who owns the original data, whether the new images infringe copyright."
But to Bonner and Lawley, it's simple. It's theft. "It's stealing somebody's identity in order to create false identities to sell for financial gain," Bonner says.
Bonner fears for kids' perceptions of beauty
Bonner taught herself in a day to create her own AI models, switching faces from one body to the next. It's a whole new world of mix and match, where you can create a tall, leggy blonde and put someone else's face on it or blend facial features that appeal – the nose of Margot Robbie, the cheekbones of Rihanna.
The next step is to put the AI models to work.
Websites such as Lalaland.ai, a digital fashion studio, make fake AI models for clients to put their clothing on. Send the site the 3D renders of garments, and, says Bonner, "they will put them on a fake human model and send back the finished images that [companies] can then use in brand campaigns and catalogues, in-store posters and billboards". No more tricky and expensive photo shoots.
Jeans company Levi's was one of the first to use Lalaland's AI models, heralding its newfound ability to provide computer-generated models of different shapes and ethnicities to wear the brand's clothes on its online shopping site. These models, says Bonner, were very realistic.
"No-one would have picked up on it if they didn't decide to proudly use it as a branding exercise," Bonner says. Levi's was pilloried in the media for dispensing with real models, especially diverse models who had only just got a foothold in the industry. The company later issued a "clarification" saying it would not scale back the use of real models.
But the door is open. Bonner knows of smaller Australian companies that have switched completely to AI models. As a businesswoman, Bonner understands the hunt for cost savings but says the ramifications of this innovation are frightening not just for diversity, but the economy.
"Taking jobs off countless people in a woman-led industry [while the makers of AI models] are sitting there on the other side of the world not paying taxes in Australia, not contributing to our Medicare levies, not paying anybody's superannuation," Bonner says.
"The billions of dollars in the economy that our industry generates will very soon be gone."
Bonner and Lawley also discovered that the bodies of real women – women who had posted shots of themselves at the beach or out for dinner — were being given AI faces and used for porn or dating sites, or as influencers selling things. And they were being passed off as real.
Many of these AI models have thousands of followers, some have millions, and tend, says Lawley, to have big breasts and a ridiculously unblemished, childlike face. "These bodies do not exist," she says. "They're not real. This is fantasy."
Bonner says the "heaviness of the male gaze" is obvious in the stereotypical "perfect" AI woman created, fuelling her fears for impressionable girls.
"The prevalence of eating disorders after AI is introduced is just going to be catastrophic," she says.
It petrifies Lawley. She has a nine-year-old daughter and worries what these idealised AI models will do to children's body image and self-esteem. "What images are they going to look at and think, 'Is this a human? Should I compare my body to this body? Or is this a robot?'."
'We can't trust the images we see'
As Bonner went down the AI rabbit hole, she discovered the book Man-Made by journalist Tracey Spicer, and was so alarmed "I didn't know whether to cry, or who to hit, or who I needed to take out".
She got in touch with Spicer, who argues that "the bigotry, bias and discrimination" of the past is built into the data sets, algorithms and machine learning of AI.
"It's putting all of us into boxes as consumers and feeding us things that are stereotypical from the past."
On a more existential level, Spicer says the creation of fake images passed off as real is a threat to society.
"We can't trust images that we see, and words that we read," Spicer says. "That leads to a greater mistrust and distrust in democracy, in leaders and institutions, and ultimately leads to more social unrest, more wars, and a really dislocated, disrupted, dangerous society."
United in their fear, Bonner, Lawley and Spicer launched the petition, which has now been accepted by the federal parliament and awaits ministerial consideration.
Its headline demand is for AI images to be clearly labelled as such. Bonner says the technology is there to read metadata to differentiate AI photographs from real ones. What's needed is the political might to enforce it.
Regulation of AI is on the federal government's agenda, says Jeannie Paterson, but it is a complex, multi-faceted issue to legislate. AI offers great benefits in areas such as medicine but also comes with risks.
She backs the creation of laws that enforce the labelling of AI generated material but cautions "there's no silver bullet".
"We need to develop and design carefully a … hierarchy of initiatives that can help us preserve fundamental values. Like the real, the creative, the human."
Now is the time, says Bonner.
"We really need to all come together as a community … and say, 'This is dangerous … we as a society do not want this and stand together and ask for change and ask for the legislation that we need'.
"This is not something I can do on my own. It's not something that the three of us can do," Bonner says. "Everybody has to care about this in order for us to get it done. Everybody."
Watch Australian Story's Face to Fake, 8:00pm, on ABCTV and ABC iview.
[Zendesk]
