The Monet question

27 May, 2026

Recently on Twitter, someone asked us to describe why a (purported) AI-generated Monet-style image was inferior to Monet. The trick is that the image was of a real Monet.

First: I fell for this. I looked at it for a few seconds, reacted to it, and wrote down some stuff about the outlines of the waterlilies and the balance between the parts of the painting. So all of this is coming from someone who "got owned" here.

That said, this sort of trick says only a little about the potential of generative AI, and almost nothing about Monet. Many commentators drew optimistic conclusions about AI-generated art or pessimistic conclusions about Monet. Those conclusions are unfounded, for at least these reasons:

1. People are not generally good at judging the quality of art quickly. Great art is great largely in how it rewards attention over time. (And analogously for good writing of any sort, which is a big reason that LLM prose is so dangerous.)
2. The quick reactions of Twitter users you've primed with false information have very little significance about anything, and almost no significance about the quality a work of art.
3. Many analyses of the situation assumed that the value or quality of a Monet lies entirely in how someone would react to, or assess, it in 2026. Art doesn't work like that. What makes Monet Monet is largely in its effects on (and reflection of) culture. Imagine asserting that, because Twitter users don't spontaneously distinguish between an AI-generated image of a steel ball and an image of one from a Yayoi Kusama installation, we should lower our opinion of Kusama. This would be an outrageous misunderstanding of Kusama's art.¹ It's a tricky question how much less outrageous the analogous misunderstanding is for Monet, but I'd say "not much less."
4. Paintings famously lose so much of their effect in reproduction, and especially in digital representation. (Benjamin Moser, in The Upside-Down World, goes out of his way to note that Girl with a Pearl Earring actually reproduces well: it's a big surprise when that happens.)

I'm quite eager to draw conclusions about aesthetics from the existence of generative AI. It's a great way to clarify intuitions and face hard questions about what really matters. But, for all those reasons and more, I found almost all of the reactions to the Monet trick to encode grotesque misunderstandings of what art is, does, and is for. It should not prompt worries about Monet, but worries about us.

#generative AI #things that make me sad