One cheer for “‘AI’ art”?

I am generally skeptical of breathless “AI” boosterism. LLMs have given us some astonishing surprises in the last few years, and we can expect more surprises over the next few years at least, but it is preposterous to see artificial superintelligence is on the horizon. LLMs have differently-shaped strengths and limitations from familiar proceedural computing systems, but the contrasts with what we see from human cognition make it obvious that there are a lot of things humans do which they simply cannot.

I am generally disgusted by the disdain both for the arts and for artists in how “AI” boosters imagine replacing artists with robots. Surprising and weirdly impressive as developments in “‘AI’ art” art have been in the last few years, I cannot imagine thinking that “AI” is on a trajectory toward replacing novelists, painters, filmmakers, et cetera. I join those who refer to “‘AI’ art” as “slop” because of its deeply schlocky quality.

But I don’t want to be a one-dimensional “AI” hater, and have been challenging myself to think about ways these tools can be useful. So here are some ideas I am not sure are good, offered as a provocation to my thinking and others’.

Slightly better clip art?

Maybe “AI” illustrations are not a big problem because their distinctive janky quality signals what they are, meta-communicating I am using this as an ephemeral little instrument to communicate a simple intent, rather than as a subtle work.

When generative “AI” image tools were new, I experimented a bit. I made this little meme image quoting a Lou Reed song by asking “AI” to give me a “sneering Statue Of Liberty”. I am satisfied with it delivering a little something I could neither draw nor afford to pay an artist to do.

I think that plays because it is not a subtle work. A picture is worth a thousand words and I wouldn’t ask “‘AI’ art” to do that … but sometimes you just need an illustration worth less than a dozen words. There are plenty of bland corporate filler illustrations which express no profound human intent. I know using “AI” deprives illustrators of jobs making this stuff, but isn’t that the kind of elimination of toil we should embrace as enabling Fully Automated Luxury Gay Space Communism?

Maybe the right regulatory solution to “AI” images is to lean into their janky soullessness. Might we force them into a few house styles so we always recognize that they were not created by an artist?

For the TTRPG table?

There is currently a fight among tabletop roleplaying game creators about the use of “‘AI’ art”. I join with those who advocate against the use of “AI” slop in rulebooks and other materials. But I have been thinking about a possible LLM-driven artbot for creating images on the fly for use during play in games like D&D which has the necessary scale to pay for artists to create big corpus of art used as training data.

They would have a bunch of reference illustrations of what halberds, swords, beholders, dwarves, fireballs, et cetera are supposed to look like plus tagged examples of a handful of “house styles” like Cozy Fantasy, Heavy Metal Album Cover, Heroic Pseudo-Realism, Medieval Woodcut-Ish so that it can render any of the Things in any of the Styles.

Published settings & adventures could include an artbot module of data about creatures, locations, items, NPCs, etc to support relevant illustrations.

Then a gamemaster could instruct the artbot tool to spit out images of what is happening in play: “Liddel The Rogue and Vadania The Druid face the kobolds in the Cave Of Dread” or whatever.

The artbot tool to do this would require real programming, not just prompt “engineering”. It would be important to create images which are consistent, which align with the plans for an adventure, et cetera.

My hope is that this kind of artbot would end up benefitting artists, producing an ecosystem with new jobs to create reference images to train it for an endless succession of adventures and sourcebooks. Software-mediated marketplaces like Etsy have enabled artists a bit, making it easier to sell stuff like commissioned character portraits; I imagine this hypothetical TTRPG artbot enabling new kinds of commissions between players and artists.

Maybe none of that would work! But this is the kind of thought experiment which keeps me from being an absolutist in opposing “AI” illustrations.

As an ingredient in mixed-media art?

Over on the Bluesky thread where I first articulated this idea, someone countered …

“Fully automated luxury gay space communism” should have a higher aspiration than to feed money to the VIC collective to torch miles of rainforest and fresh water to generate an image of partying seniors with 9 fingers per hand for a failing rural elder care program.

Put a tick more precisely I think a use case for LLMs of artists doing drafts and preliminary scaffolding for their sincere work is orders of magnitude more socially beneficial than for making corporate PowerPoint sloptent even more repellent & sloppier.

Indeed. Even my dumb little meme images seem useful to me as an instrument for a small purposeful comment.

The more substantive uses of “AI” I find interesting are artists using these tools as one among many instruments in composing something with intention and meaning, whether it is James Curico weaving “AI” elements into complex mixed-media artworks or Damien Walter using “AI” illustrations just to liven up a videoessay so it’s not just him being a talking head. Curico says:

I’ve had very good results in this direction, personally. Though it’s kind of a lose-lose talking about the process since on the one hand it’s “doesn’t matter if you spent 90 hours on it you still deserve the electric chair” and “what's the point if can’t do it with one prompt?”