The rise of the AI beauty pageant and its complicated quest for the ‘perfect’ woman

Ten women participating in a beauty pageant is nothing new. Some pose candidly, some play to the camera, their beauty forever frozen in this moment in time. Like many other pageants held in countries around the world, the contestants are young, thin and embody many of the standards defining traditional “beauty.”

But that is where the similarities to a traditional beauty pageant end. None of these women are real — everything about them, even the emotion that flickers across their faces, is generated by artificial intelligence (AI), for the world’s first ever AI beauty pageant. Each has a creator or team of creators, who use programmes like Open AI’s DALL·E 3, Midjourney or Stable Diffusion to generate images of the women from text prompts.

These 10 contestants have been selected from a pool of more than 1,500 entrants to make the final of “Miss AI,” scheduled to be held at the end of June and broadcast online by its organizers “The World AI Creator Awards.”

For those involved, the event is an opportunity to showcase and demystify the technology’s extraordinary abilities. But for others, it represents a further proliferation of unrealistic beauty standards often linked to racial and gender stereotypes and fueled by the ever-increasing number of digitally enhanced images online.

“I think we’re starting to increasingly lose touch with what an unedited face looks like,” Dr Kerry McInerney, a research associate at the Leverhulme Centre for the Future of Intelligence at the University of Cambridge, told CNN in a video interview.

Each of the contestants has a unique and distinctive personality, as well as face. One red-haired, green-eyed avatar named Seren Ay poses for Instagram photos as she travels around the world and through time, appearing next to Turkey’s first president Kemal Ataturk, on the Oscars red carpet or wandering through the neon-lit streets of Kyoto, Japan at night.

Related card
A history of beauty trends – and the standards that shaped them

And like real life pageant contestants, some AI avatars promote specific causes. One, named Aiyana Rainbow, posts in support of the LGBTQ community, her allyship literally displayed by her rainbow-colored hair, and name. Another, Anne Kerdi, posts about cleaning the oceans, her native region of Brittany in France and travelling. Zara Shatavari, posts tips on her blog for dealing with depression or strategies for losing “stubborn belly fat.”

All are beautiful. But, echoing the reality of most modern Miss USA beauty pageant winners since the competition’s inception in 1921, most are White, thin and have long hair and symmetrical features, detailed Hilary Levey Friedman — a sociologist and author of “Here She Is: The Complicated Reign of the Beauty Pageant in America” — in a phone interview.

Racial and gender biases ingrained within beauty standards also seep into programmes that use AI to generate images — since they have “learned” from the troves of data on the Internet that already contain these biases. As such, research has found that AI reflects these gender and racial stereotypes when generating images, reducing beauty into a homogenous ideal.

Most of the models on the “Miss AI” shortlist, McInerney said, are “very very light-skinned and the vast majority are still White women, still thin, still really not diverging very much from that norm.”

“These tools are made to replicate and scale up existing patterns in the world,” she added. “They’re not made necessarily to challenge them, even if they’re sold as tools that enhance creativity so when it comes to beauty norms… They’re capturing the existing beauty norms we have which are actively sexist, actively fatphobic, actively colorist, then they’re compling and reiterating them.”

Open AI has acknowledged that it finds “DALL-E 3 defaults to generating images of people that match stereotypical and conventional ideals of beauty.” But while AI images can perpetuate these standards, some argue the technology doesn’t represent a completely new phenomenon due to the huge number of digitally edited images online, enhanced by filters or airbrushing. “When we look at the beauty standards of influencers, they are not real as well…” Furkan Sahin, one of Seren Ay’s creators told CNN in a video interview. “They look perfect, it’s like an AI.”

Though judge Sally-Ann Fawcett acknowledged “there’s a long way to go,” she told CNN in a phone interview that “we wanted women who are more diverse in every way, in size, in age, in flaws… It’s taken 50 years for pageants to get where they are today, with AI it can be done on fast forward.”

Fawcett, who has written four books about beauty pageants and is the head judge at Miss GB, added that she had “doubts” when she was first approached by the competition’s organizers, but that she saw it as an opportunity to shift the public perception of AI-generated women.

Creators of these AI models add that the technology itself is not necessarily the problem. “AI makes it perfect but perfect is how people want it,” said Sahin, “and we are not really changing any beauty standards.”

Similarly, Sofía Novales, a project manager at The Clueless Press which created the popular AI model Aitana López who “sits” on the pageant’s judging panel told CNN by email that “we are not here to solve this long-standing problem.”

“But we aim to encourage AI personalities to be diverse and acknowledge the existing issues surrounding beauty standards.”

AI and robotics have long been used, often by men, to create the image of a “perfect woman,” said McInerney, referencing the Stepford Wives trope and the 2014 movie “Ex Machina.”

Related article
Forever young, beautiful and scandal-free: The rise of South Korea’s virtual influencers

As technology becomes increasingly entwined with creating this version of an ideal woman, the in-person beauty pageant world has responded with a shift towards emphasizing authenticity, says Levey Friedman. “There’s been a turn in the past decade that’s really focused on be yourself, be authentic, be perfectly imperfect, all these sorts of catchphrases,” she added.

Such notions have found their way into pop culture too — Merriam Webster’s 2023 word of the year was “authentic,” partly thanks to “stories and conversations about AI, celebrity culture, identity and social media,” the dictionary said at the time.

Competition organizers say entrants will be judged on more than just their beauty. They will earn points for their creators’ use of AI tools as well as their social media influence and have to answer questions like “if you could have one dream to make the world a better place what would it be?”

Fawcett said that she is looking for “someone with a powerful, positive message,” while Novales said that they are “not just evaluating beauty, but also the technology behind it… and, above all, the backstory behind each avatar.”

Many of these AI avatars were originally created as marketing tools, to act in the same way as a human social media influencer might. Seren Ay was created to promote an online jewellery store when its founders found it difficult to work with human influencers, they said. Aitana López, can earn up to €30,000 (around $32,000) a month from sponsored posts, Novales said.

Such AI influencers have already proved their worth in recent years — one named Lil Miquela has amassed millions of Instagram followers and worked with brands like Calvin Klein and Prada. Unlike their human counterparts, they appear flawless, ageless and free of scandal. They don’t need to be paid and they can be directly owned by a marketing agency or by the company whose products they are promoting.

“Influencers are just behind a screen,” Mohammad Talha Saray, one of Seren Ay’s creators, said. “They’re not real for us, they’re just a girl or guy on the Internet and when you think about that, there’s not much difference between AI and an influencer.”

He told CNN by email that he created Anne from different AI systems and programmed her so that “she is free to say what she wants as long as it does not involve misinformation.”

“It is sometimes frustrating for me to see her on video at important events expressing a view different from mine, or writing in a way I would have imagined differently but… we each have our own free will.”

Both Anne Kerdi and Seren Ay exist as more than simply images for their followers who often interact with them, asking advice of Seren as if she were their “big sister,” said Sahin, or wishing Anne goodnight, said Keranvran.

“Just as we become attached to literary or movie characters, some people are attached to Anne,” he said. “She responds affectionately and sometimes humorously when someone asks how she is doing.”

Creators of some AI avatars use this relationship with people for the adult entertainment industry. “Miss AI” is sponsored by Fanvue — a site that is similar to OnlyFans and hosts both AI and human content creators. Understanding the data that is being used to train AI avatars used for sex work is crucial, McInerney said, “because so much of the available data out there is not only really sexist, it’s also very heterosexual, it might not leave spaces for other kinds of sexual orientations, identities, experiences.”