Menu Menu
[gtranslate]

Are AI-generated images perpetuating toxic beauty standards?

Viewing photos generated by apps like Midjourney, DALL-E, and Adobe is no longer a novelty. Could frequent exposure to highly idealised – and unrealistic – images of human beings potentially reshape society’s beauty standards?

If you’ve ever toyed around with AI photo-generating apps, you may have noticed a trend emerging – especially when asking the technology to depict realistic-looking humans.

The trend I’m referring to is that all AI-generated humans seem to be stunningly beautiful – borderline perfect, in fact – despite the presence of visually flawless individuals rarely being a reality in our day-to-day lives.

Take, for example, the below image.

I requested Midjourney to create a photo for an article about young people forcing the beauty industry to change. The prompt was something along the lines of ‘a diverse group of Gen Z teens wearing creative and colourful makeup.’

The result is striking. I mean, they look like ethereal fairies.

Their features are incredibly Eurocentric – containing attributes not quietly favoured in society throughout history – complete with button noses, light eyes, and high cheekbones.

The Black girls have white-textured hair, while the white girls have the kind of full, plump lips that most Caucasian women are achieving these days through regular trips to cosmetologists. Each of them is digitally blessed with pore-less, plump, hydrated skin.

While the making of AI tools is still shrouded in some level of mystery, we generally know that this technology is first armed with baseline information input by programmers. From here, AI tools are also taught to accumulate ‘knowledge’ from data they collect online, whether they’re using this information to generate text-based or image content.

But if the majority of reference photos have to be of flawed, real-life people for Midjourney to be accurate, why is it that photos of humans generated by AI look so… freakishly perfect? And what are the consequences of perfection being overrepresented in the images it produces?

Midjourney photo prompt: a young man sitting on a bench.

 

Limitations and bias

Bias is clearly playing a huge role in the way Artificial Intelligence ‘thinks’ humans look and it’s likely that much of this starts at the beginning.

When programmers are teaching AI what people look like, the largest and most widely available available catalogue of human photographs are of celebrities and other famous figures, who are photographed hundreds upon thousands of times a year.

Using photos of celebrities as an AI learning resource likely means including a vast library of advertising imagery – specifically those used in fashion, makeup, and beauty campaigns. These images are close-up shots of the face, which in many ways can act as great learning opportunities for AI image generators.

The downside of this though, is that tools like Midjourney start out with a skewed idea of what regular human faces (along with skin texture and hair) look like because they have been overwhelmed by heavily photoshopped photos of the world’s most beautiful faces to begin with.

Midjourney photo prompt: a photo of an Indigenous woman.

On top of this, further chances to learn how to depict an average face are lost when AI begins scouring the internet for further image data points.

This is because everyday people are increasingly inclined to only show the most attractive sides of themselves on social media, with users piling on filters and using teeth-whitening or skin-blurring editing tools to eradicate perceived flaws.

Finally, AI tools are constantly learning what kinds of images users are searching for. They also make note of what users download for use. If we are always choosing the most attractive photos, AI photo generators will have their actions affirmed and keep serving up more hyper-beautiful people in the future.

It is quite amusing to note that one of the only times I’ve noticed things get a little wonky – yet sadly accurate – is when you ask Midjourney to create ‘a hyperrealistic photo of politicians’. These depictions are likely based on cartoonish images used as political satire across the internet.

What this tells us is that AI tools are essentially acting as a mirror, reflecting our idealised digital selves – and aesthetic biases – back onto us.

Midjourney prompt: a group of politicians.

 

The repercussions of biased AI-images

While bloggers, journalists, and artists will surely continue to use AI-generated imagery in the future, there is a growing debate about whether AI should be legal for use in advertising.

The use of AI imagery could lead to false promises about the real-life design or benefits of any given product – especially in cosmetic and beauty industries – essentially tricking customers into making purchases based on false pretences.

While it seems obvious that tools like Midjourney are simply echoing society’s preference for Eurocentric features, a lack of discernment for the widespread use of these images could end up perpetuating an even higher standard of beauty that is virtually – and literally – unachievable.

Though this suggestion might sound far-fetched to some, ‘I know AI images aren’t real people, so how could that influence my self-esteem?’, we can’t forget how the introduction of selfie-camera face filters and increased screen time during the pandemic caused a boom in cosmetic surgeries and non-invasive procedures such as Botox and filler.

It appears that by viewing the work of machines we’re training to think like us, we’re seeing the ugly side of humanity’s superficiality and biases emerge at the click of a button. Only time will tell if AI images will heighten the beauty standards of the future, but one thing is for certain – our desire for perfection is certainly influencing them.

Accessibility