Ethical Questions Arise After AI “Completes” Keith Haring Painting

A recent thread on X spotlighted some of the inherent ethical problems of artificial intelligence after someone used an AI image generator to “complete” a painting by Keith Haring that was intentionally left unfinished as a commentary on the AIDS crisis. The artist died from complications of the disease just a few months after he created the work.

Haring’s “Unfinished Painting” (1989), a white canvas with a purple background that covers only the top-left quadrant, is layered over with the illustrator’s iconic, heavy-lined figuration and designs that end abruptly. Aside from a few streaking purple drip lines that slide down the untouched canvas beneath, the punctuated painting is ear-splittingly silent beyond the broken edges of what Haring had rendered in his recognizable visual language, alluding to the thousands of lives cut short during the beginning of the epidemic that affected many in the LGBTQ+ community. 

Last summer, artist Brooke Peachley, who preferred to go by the last name she uses on her social media accounts to protect her privacy online, posted a photo of the work on X alongside a prompt asking others to respond with a visual art piece “that never fails to destroy [them] every time they see it,” which gained momentum and continues to receive responses daily. Over six months later, another user responded to the original post with a generative AI image that “completed” Haring’s purposely half-painted work, writing, “now using AI we can complete what he couldn’t finish!” 

The post went viral and the backlash was immediate: Commenters used words like “travesty,” “gross,” and “cruel” to describe what they saw as an obvious affront to both Haring and the other individuals whose lives were lost to AIDS. Some X users expressed their outrage at the impudence of using AI to complete Haring’s meaningful work, with one user calling generative art “a way for lazy, ignorant people to fulfill their selfish desires while also going against everything the art world stood for.” Still others applauded the post, presumably for its ability to drum up controversy.

The AI-generated image, posted by user @DonnelVillager, is likely an example of “bait” — intentionally polarizing or anger-inducing content meant to generate engagement by eliciting strong reactions from others. X and other algorithmic social media platforms tend to “reward” posts that get a lot of engagement by hoisting them to the top of people’s feeds regardless of their interest in the topic at hand. Since Elon Musk took control of what was then-called Twitter in October 2022, the site has seen an increase in hateful and incendiary speech. This past summer, Musk introduced a policy that allows users who pay for X Premium to share in their Tweets’ ad revenue, offering a monetary incentive for people to post intentionally controversial content that will receive engagement, whether positive or negative. 

Regardless of whether the post is bait or not, Peachley told Hyperallergic, “This is the kind of bait you go to hell for,” noting the ethical transgressions associated with distorting the message of Haring’s original artwork.

“I find the ‘completed’ version of the artwork to be abhorrent,” she said. “Not only does ‘completing’ the painting completely negate it of its original meaning, but spits on the tens of thousands of queer individuals who lost their lives to the AIDS epidemic in the ’80s and ’90s.”

“And to do so using generative AI, a computer program that cannot feel the weight of what it is doing nor create with any sort of human intention, only adds to the disrespect,” Peachley continued, going on to note that the AI was unable to accurately recreate Haring’s figures, depicting abstractions rather than the artist’s characteristic human forms. The Haring Foundation declined to comment.

Artist and writer Molly Crabapple, a vocal critic of AI art who penned an open letter this spring urging the restriction of AI-generated illustrations in publishing, did not mince words when Hyperallergic asked for her thoughts on the “completed” Haring work.

“This is the actual use for AI,” she said. “It’s a way for witless dullards to suck every bit of anima, pathos and humanity out of art. It lets screen-deadened consumers turn the artists they claim to love into undead sock puppets churning out content slop, and it lets them pretend they are the real creators while doing it.”

Last month, a database of artists used to train the Midjourney AI generator was leaked online. Haring was one of the over 16,000 visual creators on the list, which included figures such as Salvador Dalí, David Hockney, and Yayoi Kusama. AI image generators scrape artwork from the internet and train their tools to create work in the style of specific creators. Artists have pushed back on multiple fronts, from posting “No to AI art” on social media to adopting a tool that “poisons” image-generating software to filing multiple lawsuits accusing AI companies of infringing on intellectual property rights. 

“Generative AI is hurting artists everywhere by stealing not only from our pre-existing work to build its libraries without consent, but our jobs too, and it doesn’t even do it authentically or well,” Peachley said.