Is This by Rothko or a Robot?

I have found many abstract-type paintings (and non-abstract for that matter) to be much more interesting in person, where you can experience the actual scale, texture, colors, multiple viewing angles and perspectives, reflections/interaction with the environment and other viewers, etc..

Perhaps that's an argument for high-resolution VR museums, with better scans of paintings to capture 3D texture, layers, transparency, reflectivity, etc.. ;-)

Somehow it doesn't surprise me that the experts had the hardest time with the abstract art.

> With this genre it’s very hard and the idea of cohering logic is important. Most abstract paintings come from a debate about what’s necessary in the mind of the artist, rather than what’s arbitrary.

This is why I've never been a fan of this particular style - it feels to me that if you have to know and understand the history of the artist in order to appreciate a piece of art, that means it's lacking something fundamental. Basically every other style of art can stand alone - you see the painting and you can appreciate it solely for what it is, without any other context (though context may certainly add to it).

Though perhaps I am just an uncultured boor!

Amusing, but expertise in art doesn’t equate to expertise in detecting AI-generated images. The one with the flowers has most of the flaws we’ve come to expect from 2022-23 AI - superficially similar but it doesn’t have a clear idea of what a flower actually is. Some of the petals are attached randomly, some other structures in the pistils are confused. No doubt that AI will eventually handle these details, but for now there are some tells.

Many credentialed people can barely tell forgeries from authentic works of art so I doubt this sort of inquiry yields anything other than public humiliation of so-called experts. (Aesthetic expertise kinda is, mostly.)

The most interesting part is the (IMO usually poor!) reasoning by the judges, whether right or wrong. It's clear that most people have no idea what AI finds "difficult", and try to generalize either from what's difficult for human artists, or what it's difficult to simulate with traditional digital techniques.

I would be interested in a comparison between image generators and artists with high technical skill but who focused their efforts into extremely kitsch/commercial work. I think the AI would see more "success" there.

Somewhat OT: could we use a training set of movie scripts for good and not-good ones. It could then be used for generating scripts or for evaluating written scripts. Maybe even used for written scripts to make them better.

We should move past "look what AI can do!" into ways to improve what we have by making use of them.