AI and Art: The Childish and the Childlike

“As stupid as a painter” —Marcel Duchamp.

The disappointment I have with visual generative AI is the deluge of trite, cliched and derivative imagery that’s produced with it. Midjourney’s explore homepage, for example, is an endless scrolling wall of fantasy art, cutesy images, cartoonish animations, and sexy virtual models. (Aesthetically disappointing, but perhaps anthropologically insightful, a snapshot of our collective cultural consciousness. Also disappointing.)

Often, the impulse with art technologies is to create what people can already do, but better and with more fidelity. A shortcut to displays of virtuosity. AI generated imagery reminds me of the hyper-realistic airbrush art of the 70s and 80s. Work created with the goal of eliciting a “wow! I can’t believe that isn’t a photograph” response. Paintings with detailed renderings of reflections in chrome or glass and mirror were among the best examples off this type of work.

“At twelve I knew how to draw like Raphael, but it took me a lifetime to learn to paint like a child.” —Pablo Picasso

In mid 2023, generative AI started getting hands right. It could finally render hands with the correct number of fingers, proper hand to body side positioning, and natural looking gestures. It was a big step closer to achieving “wow, I can’t believe that isn’t real” results. It was at this time that I began to feel as if we were letting a big opportunity pass by, a developmental phase we were overlooking. What if, instead of trying to get AI to do what we can already do, but better, we trained it to do things we might never have thought to try? We’ve been like misguided adults, treating visual AI like a child learning to draw and constantly correcting it when it gets things “wrong”—like a goat’s head on a butterfly. With six fingers. Maybe we should be nudging it to double down on the things it does unexpectedly, differently, or wrong, in an attempt to seek out fertile new territories of possibility or different takes, if not altogether entirely new aesthetics and genres. Or better yet, to rethink the very nature of creating visual expressions and narratives.

I found the quirks and errors in early visual AI to be one of the things that what was most interesting about it. It was something you didn’t encounter with any other medium and something you couldn’t entirely control either. Sometimes it produced areas in images that were glitchy. Sometimes areas in the images would look like the AI was struggling with that part and just left it as a smeary mess.The technology has advanced so quickly that people working with the systems didn’t have that long to explore those faults. 

The problem with a technology that advances so rapidly is that we’re missing out on opportunities to test of the limitations of it as a medium as it develops. Sometimes the most interesting characteristics are where the medium fails. For artists, exploiting the faults can be a rich ground for experimentation and discovery. For an audience, the unexpected deviations can be what makes any art exciting.

Participation in art ends for most children as they approach adolescence. The struggle and frustration experienced with not “getting things right” breaks them during this increasingly self-critical phase. But the process of working through this is foundational to artistic development which is about more than just the refinement of craft, it’s about recognizing and giving form to what is unique and irreproducible to one’s experience and vision. AI makes it possible to bypass more than just the development of craft, by its very nature it bypasses the self-discovery that is so central to what art is and what it’s for. This doesn’t mean that AI can’t be art, but the art does not reside in the AI. The question isn’t whether AI is art or not. The real question is: “is your work art?”

“I'm an artist, man. Give me a tuba, and I'll get you something out of it.” —John Lennon