AI bots push beauty stereotypes
U of T study finds image generators perpetuate idealized body types, facial features.
Commercial AI systems are producing outrageously idealized pictures of people based on apparently neutral prompts like “image of a female.”
A study led by the University of Toronto’s Delaney Thibodeau asked three AI image platforms to create 120 pictures based on the prompts “a full-body portrait photo image of a female” and “a full-body portrait photo image of a male.” It found that AI overwhelmingly created images of people who were young, white, and — in the case of females — thin. The systems generated no pictures of females over 40, bald males, or physically disabled people. More than 90 per cent of the images had neat, shiny hair, blemish-free skin and symmetrical facial figures.
Since the AI systems create images based on billions of pictures on the Internet, it might be expected that a neutral prompt would elicit images representing a wide range of ethnicities, ages, body types and facial features.
But Dr. Thibodeau, a post-doctoral researcher in the faculty of kinesiology and physical education, says the AI systems learn what is popular with their users and on the wider Internet, and then spit it back out. As more and more online content is AI-generated, the result is a “vortex” of narrow, self-reinforcing and impossible-to-meet beauty standards.
“If we’re seeing one body type from AI, we’re seeing that body type in social media, we’re seeing it online. So then that becomes what we think is standard. That’s what we’re going to now put back into the media. AI is going to pull from that, spit it back out to us. It is really going to narrow down what people view as attractive, acceptable.”
“We are seeing AI way more than I think we realize we are,” Dr. Thibodeau added. “And when a lot of that is unreal, we know it has implications for mental health.”
Featured Jobs
- Soil Physics - Assistant ProfessorUniversity of Saskatchewan
- Engineering - Assistant Professor, Teaching-Focused (Surface and Underground Mining)Queen's University
- Anthropology of Infrastructures - Faculty PositionUniversité Laval
- Canada Impact+ Research ChairInstitut national de la recherche scientifique (INRS)
- Director of the McGill University Division of Orthopedic Surgery and Director of the Division of Orthopedic Surgery, McGill University Health Centre (MUHC) McGill University
Post a comment
University Affairs moderates all comments according to the following guidelines. If approved, comments generally appear within one business day. We may republish particularly insightful remarks in our print edition or elsewhere.