How I Created a Monster
I asked an artificial-intelligence tool to draw Jews. The results ranged from amazing to bizarre—and suggest that this technology is here to stay.
One of the fun new frontiers in artificial intelligence is computer-generated illustration. The concept is simple: Users provide a descriptive prompt for the AI, which then consults its large database of historical imagery and produces a brand-new piece of artwork. For example, here is President Barack Obama holding a lightsaber:
As you can see, the results here are…mixed. That’s because I was using DALL-E Mini, a sort of junior version of the real thing. (The name is a play on Pixar’s robot WALL-E and the Spanish artist Salvador Dalí.) The most advanced editions of this technology are not yet publicly available to consumers. To gain access to the real DALL-E or fellow travelers like Midjourney, you need to sign up and wait your turn. But more stripped-down versions of these cutting-edge creation tools are starting to roll out online for everyone, which means that anyone can employ a robo-illustrator to sketch whatever bizarre idea pops into their head.
In my case, I created a monster.
Personally, I can definitely see the Hanukkah Monster™ fronting an entire franchise of children’s books and matching lunch boxes. But more seriously, the success of even this lower-level generator at creating a novel Jewish icon gives you a glimpse of what this technology could do down the line. A child could employ DALL-E to illustrate a story they wrote for school; cash-strapped musicians could use it to produce art for their albums; families could use it to create signs and posters for their homes.
Or you could just do nonsense like this:
You can try DALL-E Mini yourself here. (If your prompt doesn’t go through when you click, just keep trying until it does.) You can also join the waitlist for beta access to its older brother, DALL-E 2, or do the same for the remarkable Midjourney.
Even at its most advanced, I’m skeptical that this technology would simply replace actual artists. Consumers would still place a premium on human-produced artwork, just as they prefer original pieces to reprints. Moreover, an AI trained on existing artistic styles will be hard-pressed to conceive of new ones, although it might stumble into doing so by accident. But regardless of such limitations, the reality is this: Computer-generated illustration is a lot further along than most people realize, and its uses will eventually extend well beyond amusing internet memes.
Not that there’s anything wrong with those:
Thank you for reading this paid subscriber edition of Deep Shtetl, a newsletter about the unlikely intersections of politics, culture, and religion. Send me your zaniest DALL-E creations at deepshtetl@theatlantic.com, and be sure to subscribe if you haven’t already.