Is AI-generated art "art"?

🎉 I am writing these notes at Brick, a magical mystery no-bullshit publishing platform. Turns out writing goes much faster when I don't have to hit “Publish” or do git commit.

You can use it too — check it out at Brick.do.


I.

Imagine a beautiful tree. Not a picture of the tree, but the tree itself. Is it art? If you hesitate to say “yes”, try to figure out why.

I do hesitate to say “yes”; I think art should be intentional. But then the funny thing is: any intent matters. A tree is not art from the point of view of how it looks. But it is definitely art from the point of view of how it operates — an intricate biological machine. "This is how you extract nutrients from environment; this is how you do photosynthesis; this is how you self-heal".

(Note: from this perspective, engineering can also be art. I think this makes sense.)

A painting of a tree is art. Why? "I looked at many trees and picked this one to depict" — the intent is right there. The choice is not arbitrary and is guided by aesthetics. Perhaps the painter looked at many trees and synthesized a single tree out of them. Perhaps the painter did not depict the tree exactly as seen, but re-presented it — omitted some details and left just enough for us, other humans, to recognize it as something tree-like. All of this is art.

(Note: from this perspective, a random photo of a random tree is no more art than a random painting of a random tree. I think not counting mass-produced thoughtless paintings as “art” also makes sense.)

II.

Under what circumstances can a real tree be art?

One option is — if it was picked according to some specific criteria. The more subjective or complex the criteria, the more “art” it is. I found this tree and I made you see something in it that I also saw. I managed to intentionally transfer something from my mind into yours. Art.

Another option is — if the tree is made interpretable in some context. Somebody brought a tree to an art exhibition. Perhaps it was a bear on drugs and there was no intent. The tree will still be art, because now there is something to be interpreted. See Duchamp's Fountain.

III.

From this point of view, certain AI-generated pictures, texts, etc are surely, 100% art.

A GAN has painstakingly deduced what makes a face a face, a dril tweet — a dril tweet, etc. It spits the essence of [anything you want] into your face. Then you can argue whether it has found the essence well, or whether it has completely misjudged what a human would consider the essence of [something]. Whatever. Bad art is still art.

The harder was the design task, or the harder it was to spot the connection, the more "art" the result is. A short story that makes you uneasy is more art than someone screaming "war is bad!" at you. A dril tweet that doesn't look like other dril tweets but still somehow feels like one, is more art than a drill tweet that superficially resembles other dril tweets.

If you apply the criterion of interpretability — yes, GAN art is interpretable, too. Both in terms of “invites interpretation” and in terms of “you can roughly figure out what the GAN was aiming at”.

Overall, I don't see how AI art is not art. And the funny thing is that AI art might actually be deeper than human art — a GAN might have spotted a very deep connection and you wouldn't notice it or appreciate it.

In order for AI art to be accessible to us, we'll have to steer them into the direction of spotting deep connections that humans would also notice — not just any connections.

This might be the next frontier for AI art.