An award winning media company committed to sharing the perspectives of people of colour from marginalised genders

Lensa AI/Maya Kotomori

Searching for a silver lining in the age of eerie AI art

Can the AI world open up possibilities for artists of colour?

21 Dec 2022

Edmond Belamy was the first artificial intelligence (AI)-generated painting to be sold at auction. Created by the Parisian art collective, Obvious, the 2018 image was the result of an algorithm trained on a set of 15,000 portraits from WikiArt, resulting in a series of blurry paintings evoking 18th Century aristocratic portraiture. To be honest, I found the images themselves to be unremarkable. They could almost blend into any Western palace, museum, or traditional gallery – alongside other portraits of white, stuffy noblemen. Still, there was evidently an appetite for it and Edmond Belamy sold for $432,000. 

Now it’s four years later and AI has taken over my Instagram feed. Lensa AI, a machine learning AI app, was the most downloaded free app in many countries by 7 December. My friends, old classmates, and coworkers were suddenly turned into cartoons by the app. After the initial gold rush of Lensa, came the scrutiny. The small print revealed that the app can distribute your images without compensation, can leave users vulnerable to appearing in AI-generated nudes, and artists are concerned their styles are being aped and hard work replicated as a quirky filter en masse without compensation while Lensa gets the $7.99 fee. 

“Work being ‘stolen’ from other artists is not a new concept to me”

The tools used to generate AI images have become increasingly accessible, fuelling an explosion in the interest, creation, and discussion of such images. But these concerns aren’t slowing the acceptance or proliferation of AI-generated images as art’s rising star. They’re being displayed in MoMA, have won art competitions, and have been used as magazine covers. High-profile “moments” for these images and wider-spread use have sparked debates around copyright infringements, data privacy, and the fear of job losses to mechanisation. 

To be completely frank – despite practising as an artist and art writer – I was disengaged with the topic. Work being “stolen” from other artists is not a new concept to me. I have walked through prominent British art galleries or museums that are literally full of stolen goods. Other times, the theft is more subtle. Art spaces are filled with the works of white artists like Picasso, who mimicked the work of African artists and refused to credit them. Being a “talented” artist won’t necessarily sell your art – being a POC or woman already sets you up against the odds. I already know the art world as a place that doesn’t operate on rules of fairness or integrity. 

Can the AI world open up possibilities for artists of colour? 

When Diet Paratha, an online community platform that highlights the work of South Asian creatives, shared the work of four different artists creating AI-generated images I was surprised by my delight. I found the images incredibly attractive. Prateek Arora‘s series ‘The Goth Family’ features portraits of ghoulish and demonic Indian families, set against backgrounds or homes featuring elements of South Asian traditional decor or clothing. This series resonated with me. It blended the strange and familiar by combining iconography from Indian folklore and horror with scenes that reminded me of the domestic. 

I was intrigued by Arora’s use of Midjourney, a text-to-image generator which produces images based on a prompt. It was the same program that was used to create a 90s black gothic scene and fake metal concerts, presumably to increase visibility in subcultures that are often thought of as exclusively white. The process of creating AI images involves training algorithms on a large dataset of images or other data and then using this knowledge to generate new, original images. 

For the first time, I saw the potential of AI imaging tools as a way for POC to visually re-imagine culture, with these tools allowing POC to “cast themselves” into images, stills, or sets that they’re not used to seeing themselves in. 

The work of Aanaya Dayaram similarly moved me. Aanaya is a multidisciplinary creative with a background in art direction, menswear design technology, photography, and styling. Her series, ‘The Indian Lizard Tribe’, is a self-described dark dreamworld “emphasising the surreal way our people are rarely seen.” I spoke to Aanaya about her experience working with Midjourney. “I’m using it to generate ideas for photography, and also in the hopes of directing films and TV series,” she explains. “I want them to come to life. And obviously, I can’t do that if I don’t have those resources.” She points out that it’s a struggle to execute her ideas with limited equipment, an insubstantial budget, and without the connections needed to bring them to life. Using Midjourney allows her to create concept boards which bring her ideas to life, ultimately allowing her to visualise and pitch projects with more detail, “I think it’s a great tool as long as you state it’s AI.”

Prateek, like Aanaya, has discussed his desire for his AI-generated images to inspire the production of Indian-inspired science fiction or horror films. His portfolio includes a 1980s Bollywood-style disco/alien film and cyborgs shopping in a crowded market in Mumbai. In a recent interview with homegrown, Prateek mentioned this aspiration: “My hope is to use the content I’m putting out there to catalyse an increase in creative output in these genres here in India, and in the process incubate characters, stories, and worlds that I can develop further into streaming shows, movies and games.” As for whether these images are “art” themselves? The answer is unclear. In some ways, the very act of creating and using these tools is an inquiry into what art-making is and what human functions it serves. 

Looking at Prateek and Aanaya’s work, I can’t help but wonder whether these AI imaging tools could be a powerful way for diasporic communities or POC to re-engage with elements of their culture and bring them to life. Of course, we should also consider the implications of being able to generate an image of a POC without their input or involvement. In fact, companies can already auto-generate photographs of people who do not exist, deciding whether they want an “Asian, African American, Black, Latino, European/white” model in order to cash in “diversity” brownie points.

Frederic Fol Leymarie is a researcher and professor in the Department of Computing at Goldsmiths University, London. He points out that we’ve created these tools to experiment with creativity. These tools are producing what we find quite interesting and beautiful, not what the AI finds interesting itself,” he says. “It’s humans that find anything produced by internet tools like Dall-E interesting. The question of making an AI competent at the level of a human as a sort of curator or critic is a real challenge.” 

Ultimately, humans have the final say over whether these tools and the way we use them are to be considered art. AI-generated images are a tool we have sharpened to our own liking. Our usage of the internet and our surrendering of data to companies like Google have provided a massive archive for AI algorithms to be trained on. 

“It comes as no surprise that AI image generators are regurgitating the ugly parts of our society back at us”

Tools like Midjourney pull its training data from the internet, and it’s worth noting that AI algorithms are only as unbiased as the data they are trained on, and if that data is biased, the resulting images may also be biased. Facial recognition software doesn’t recognise black faces as accurately as white faces, and there have been multiple instances of AI image-generating tools reinforcing racial stereotypes. Wired writes that entering text like “a photo of an angry man” returned images of men of colour. The generation of these images highlights systemic racism and inequality within the internet. We’ve created the internet, we’ve created a comprehensive archive, we’ve created the tools, we’ve created the companies selling the archive and the tools, and we’re feeding everything its reference materials. It comes as no surprise that AI image generators are regurgitating the ugly parts of our society back at us.

These black goths never existed.

When art in galleries is stolen, when job opportunities are already taken by people in the art world through nepotism, when art is already a commodity that businesses buy and use to sell, when the art world has shown itself to be inherently skewed towards white audiences – why is it a shock when the AI we have trained and created is a new vehicle for these problems?

The founder of Midjourney, David Holz said that its goal is to “make humans more imaginative.” AI art in its early stages yielded a classic portrait of a rich white man and his family. It was created and sold because that was what its creators, auctioneers, and buyers were interested in. Open-access AI tools in their current inception have a chance to be useful in expanding this imaginative power. POC creatives have been given a moment to reimagine the past, present, and future and turn it into something physical. We could take it as a rare chance to generate new associations – for both machine-learning tools and the human audience.