DALL-E 2 made me reconsider what 'art' means

I got access to the DALL-E 2 beta last week.

The first thing I typed was “a photograph of an astronaut riding a horse on Mars, golden hour.” Just to see. Just to test whether the hype was real.

The image that came back could hang in a gallery.

I’m not exaggerating. The lighting was warm, directional, consistent with a low sun angle. The astronaut’s visor reflected the Martian terrain. The horse’s legs kicked up dust that caught the light. The composition followed the rule of thirds. The color palette was coherent.

Nobody composed it. Nobody chose the light. Nobody framed the shot. I typed words and the machine returned an image that a professional photographer would need a crew, a location, and several thousand dollars to produce.

I sat there for a while, looking at it.

The skill question

We used to define art partly by skill. A painter who could render light on skin. A photographer who knew how to wait for the right moment. A sculptor who understood how marble wants to break. The craft was inseparable from the expression.

DALL-E 2 dissolves that connection. The technical execution is essentially free now. You can generate photorealistic images, oil painting styles, watercolor textures, pencil sketches. Any medium. Any composition. The machine handles the craft. The human provides the intent.

But is intent enough?

When I typed “astronaut riding a horse on Mars, golden hour,” was I making art? I had a vision. I described it. The result matched my vision, loosely. But I didn’t choose the angle. I didn’t decide where the dust caught the light. Those were the machine’s choices, if you can call them choices. Statistical patterns in a neural network, learned from millions of images made by millions of human artists.

It’s art made from the residue of art.

The analogy that doesn’t work

People say “photography didn’t kill painting.” True. But photography didn’t replicate painting. It was a new medium with its own constraints and possibilities. The camera could do something the brush couldn’t (capture reality exactly), and the brush could do something the camera couldn’t (distort reality deliberately).

AI image generation isn’t a new medium alongside painting and photography. It’s a layer on top of both. It can produce something that looks like a photograph. Or a painting. Or a pencil sketch. Or a Pixar render. Or a combination that has no name yet. It doesn’t sit next to existing mediums. It absorbs them.

That’s different. And I’m still figuring out what it means.

The thing that worries me

It’s not job loss. Illustrators and stock photographers will be affected, and that’s real and should be taken seriously. But that’s an economic problem with economic solutions.

What worries me is subtler. If images become free to generate, we’ll be flooded with them. The visual world will become even more saturated than it already is. And in a world of infinite images, what makes any particular image valuable?

Scarcity used to do some of the work. A beautiful photograph was partly valuable because taking it required being in the right place at the right time with the right skill. Remove the skill and the time and the place, and what’s left?

Meaning, maybe. The story behind the image. The context. The human reason it was made.

I typed “astronaut riding a horse on Mars” because I like the idea of absurd beauty in empty landscapes. The image DALL-E 2 gave me reflects that idea. But does the image carry the meaning, or do I?

Where I land

I don’t land anywhere. Not yet. I’ve been generating images for a week and I oscillate between awe and unease every few hours.

The MoMA has exhibited photography, video, and digital art. Someday it’ll probably exhibit AI-generated art. When it does, the interesting question won’t be “is this art?” It’ll be “who is the artist?”

I typed the words. The machine made the image. The answer might be: both. Or neither.

I’ll keep thinking about it.

Related thinking: