In the 1960s when The Beatles started releasing music using electric guitars, keys and amps, many members of the older generation started questioning whether it was ‘real’ music as it was produced by electronic means. The same thing happened in the 80s with the synth boom.
Now, a new culture clash is occurring with the use of AI for art. For some artists, especially artist with health conditions, text to image generators like Stable Diffusion and Dall-E, last year, can change everything for the better allowing for productive work. Others are not in agreement.
Many AI artists who share their work on Twitter – who have created images like magenta heads with lotuses blooming from their crowns – have encountered waves of vitriol and abuse from those questioning their right to call themselves artists.
Much of the criticism has arisen due to perceived intellectual property theft by platforms like Stable Diffusion, Midjourney, and OpenAI, which train their image-generating models using existing images scraped from the internet. “I think people have ethical concerns, because it’s like: OK, you’re creating art—but even though you’re not directly stealing an image, you’re using data from created artwork to source it,” said Zach Hillegas, a 31-year-old illustration artist from Colorado to Weekend. “Nobody wants to see people posturing themselves as a professional artist when they’re really just using a machine.”
The debate itself is a bit reductive; there are many different ways in which traditional artists are learning to use AI and just tech in general to create novel works. Modern music producers, for example, aren’t technically playing the drum tracks but sourcing them from samples.
Pindar Van Arman, a 48-year-old roboticist and artist from Texas, has lately been using AI models to create portraits. “I’m using diffusion [models] for prompts to get models of all types and ethnicities,” he said. “I love it.” Van Arman has been using AI to create art for the last two decades, largely with the help of self-programmed robots he said possess the “creativity level of a college student.” “I’m not releasing pictures of the models and calling that art,” he said. “I’m using AI to help my art.”
Such has become one of the internet’s hottest debates: what is the value and purpose of AI art? Believers in the craft insist their work requires skill and effort—that even feeding platforms imaginative yet precise image prompts is itself an art form. Its detractors, however, claim AI art is more akin to art forgery, with some artists saying it threatens their livelihood, and questioning whether the art world can coexist with AI in the future.
All of this is made possible by the data AI models are trained on, like the mammoth LAOIN-5B dataset, which hosts more than 5 billion images. “Over time, these text-to-image models learn a mapping between the words used to describe an image and the image itself,” said Rijul Gupta, co-founder of DeepMedia, a company that uses AI to overcome language barriers in media. But there’s a catch—one that irks traditional artists.
The data AI models are trained on, like the mammoth LAOIN-5B dataset, hosts more than 5 billion images. “Over time, these text-to-image models learn a mapping between the words used to describe an image and the image itself,” said Rijul Gupta, co-founder of DeepMedia, a company that uses AI to overcome language barriers in media. But there’s a catch—one that irks traditional artists.
“Because some images and ideas are much more pervasive throughout our culture,” said Gupta, “they may be present in the dataset at a rate 1,000 times greater than other, less common images.” This overrepresentation, or overfitting, allows AI artists to replicate the unique styles of better known artists by putting those artists’ names into their prompts. Suhail, for example, has come under fire for using prompts containing the name of artist Greg Rutkowski–whose name has been input by a multitude of AI users to generate roughly 93,000 AI artworks that mimic his unique style. Very much like 90s producers who made their trademark sound akin to samples of 60s jazz musicians.
But, according to traditional artists, mimicry isn’t the only ethical issue when it comes to AI art. “There are artists who make a quick profit off secretly AI-generated content by presenting themselves as regular, experienced commission artists,” said Thaumana, a nonbinary digital anime-style artist from Germany. These AI artists then sell designs “as self-mades that they neither created themselves nor put any amount of effort into for huge service prices.”
Others from the AI media sphere, however, don’t entirely agree. “I think there isn’t as much artistic value—potentially no artistic value—in just generating new images in the style of another artist,” said Edward Saatchi, head of metaverse guild The Culture DAO and nephew of Charles Saatchi, founder of the Saatchi Gallery.
Edward Saatchi, who was on the founding team of Oculus Story Studio, runs a collective made up of AI engineers, artists, designers and storytellers who are trying to create “the next great art forms in AI,” a future he is betting will be 3D and dynamic. Only when creators use technology for telling new and original stories, like films with unique plots or immersive video games, he said, does it become art in its own right: “I don’t really get where the future for [AI art] is.”