I'm watching a guy who built Kowloon City in Minecraft, and even using 3D modeling tools and cutting a lot of corners, it was a tedious task. This was the densest city that ever existed on earth, built haphazardly, with things like make-shift bridges between the buildings. To simulate connectivity *just* by putting doors between connected buildings (not even with the bridges) it took hundreds of tedious hands-on edits.

I thought, now this is one thing AI would be cool to use for!

I realized the difference: If it's something extremely difficult for one or a couple of humans to do, then it feels ok to use AI. That makes it exactly like any other artist's tool, from paint brushes to word processing to cameras — the tool does the thing the artist *physically can't do*, leaving the artist to do what the machine cannot.

That's what makes art special, and why AI can't make art.

@corbden I'll add some provocations:

- Humans sometimes do extremely tedious things, and often that very aspect makes the work more impressive to others. Art is defined not just by the final product, but by the process and its origins, or even just the story around it.

- It's not uncommon to work around the tedium by writing procedural generation tools. Does the output of those feel different than the output of generative AI?

@fleep Don't I know it. I have a tendency to do tedious art, but in the end it isn't very satisfying.

As a teen in the 80s, I once made a dragon sitting on top of a pile of gold using sculpey, about a foot in diameter, and I *cut and attached each coin* (less than 1/8th inch) separately. (At least I placed them on the surface of a blob of clay rather than making the whole pile out of them.) But by the time I got to the dragon, I'd lost vision and steam and was just in a rush to get the project over with. The dragon itself looks ridiculous and sad. That process did not contribute to the art or to my enjoyment, and the disaster has spent all this time in a box.

I think if he could use various 3D modeling automations, then telling a machine to add 3-4 connecting doors any place the buildings touch, would be of a similar type of automation, as if I had used a mold to make clumps of many coins all at once, directly on the surface of the pile.

Yes, I think it would be doing the work of procedural generation, but on top of an existing template (that he made to fit the accuracy of the real city).