Critics of AI-generated art often claim that it “steals” from human-made art because AI is trained on real (digital) artworks or photos of real (analog) art. However, I find the arguments supporting this claim unconvincing. Many critics misunderstand how AI generates its output.
When tasked with drawing a dog, for instance, an AI doesn’t simply copy the head from Andrew’s illustration, the torso from Sibylle’s, and the legs from Frank’s, piecing them together like some kind of Franken-dog. Instead, the AI analyzes countless examples of dogs, learning patterns and characteristics that define “dog-ness.”
To simplify, it learns concepts such as: “A dog has a head attached to a torso with legs. A head is roundish, but not as round as an apple or a watermelon.” Using this understanding, the AI creates a new image. The quality of its output depends on how well it interprets the examples it studied. If the AI hasn’t fully grasped certain nuances—like the correct number of legs a dog has—it might produce something odd, like a five-legged dog.
This reminds me of when I was a child learning to draw people. I never drew more than one head, two arms, and two legs, but when it came to fingers, it was a different story. Sometimes I drew three fingers, other times seven, depending on how much space I had left around the palm. Having seen cartoon characters with varying numbers of fingers, it didn’t seem like a big deal to me that the number was often incorrect.
Although the processes AI and humans use to learn and create are fundamentally different—AI relies on math and statistical probabilities, while humans use their senses and brains, or in one word, their bodies—the underlying approach to learning is surprisingly similar.
For centuries, aspiring artists have studied the work of their predecessors, imitating their techniques and styles. Young artists have often sat before masterpieces, meticulously copying every detail to understand proportions, composition, and brushwork. Once they had absorbed these lessons, they developed their own unique styles—much like AI does.
If learning from others is considered stealing, then every artist who has ever lived would be a thief.