But why do these programs ruin your hands (not to mention bare feet) so much? It’s a question that many people have. I ask.
To find out, I emailed Midjourney; Stability AI, which does Stable Diffusion; and OpenAI, which created DALL-E 2. Only Stability AI answered my questions.
“It is generally understood that within AI datasets, human images show hands less conspicuously than faces,” a Stability AI spokesperson told BuzzFeed News. “The hands also tend to be much smaller in the source images, as they are rarely seen in large form.”
To understand more, I reached out to Amelia Winger-Bearskin, an artist and associate professor of AI and the arts at the University of Florida, who has been exploring the aesthetics of AI art in your blog. “I’m obsessed with this question!” Winger-Bearskin exclaimed on our video call.
Generative artificial intelligence training on billions of images pulled from the internet, Winger-Bearskin explained, doesn’t really understand what a “hand” is, at least not in the way it connects anatomically to a human body.
“It’s just looking at how the hands are represented” in the images he has trained on, he said. “The hands, in the images, have quite a few nuances,” he adds. “Usually they cling to something. Or sometimes they cling to another person.”
In the photographs, paintings, and screenshots the AI learns from, the hands may be grasping curtains or grasping a microphone. They may be waving or looking at the camera in a way where only a few fingers are visible. Or they may be closed into fists where the fingers are not visible.
“In pictures, the hands are rarely like thisWinger-Bearskin said, holding up her hands with fingers spread. “If they were like this in all the images, the AI could reproduce them perfectly.” AI, he said, needs to understand what it is like to have a human body, how exactly the hands connect to it, and what its limitations are.
Hands have a fundamental place in the world of art: the handprints on the walls of the caves are the the first of all type of art that Homo sapiens created that we know of, and are considered some of the most difficult objects to draw or paint. In the paintings of Ancient Greece and medieval Europe, depictions of human hands were still flat and devoid of complexities.
It was only in the era of Renaissance art, between the 14th and 16th centuries in Europe, when artists like Leonardo da Vinci I started to study and drawing hands, including their structural elements such as bones and ligaments, which human hands began to represent in all their complexity. (This era also gave us one of the most recognizable frescoes with two hands: Michelangelo’s Adan creationdepicting God as a bearded man reaching out his right arm to touch Adam’s outstretched left arm).
“Da Vinci was actually quite obsessed with hands and did many, many studies of hands,” Winger-Bearskin said. Meanwhile, when the AI trains on an image, “it just looks at that and says, ‘Oh, in this case, there’s only half a thumb,’ because the rest of it is hidden under the cloth or attached to something, and So when you play it back, it’s kind of warped.”
However, one day, generative AI will get significantly better at rendering images of hands, feet, and teeth. “It has to be this way,” Winger-Bearskin said. “For AI to become a useful tool for humanity, it must understand what it is to be human and the anatomical reality of being human.”
