Hello DALL-E?
AI-powered image generation is here to stay. Our collective nightmare is only just beginning.
Dolly Parton, as drawn by DALL-E.
You can't turn on the Internet these days without hearing about Generative AI. In just a few months it has become the Taylor Swift of technology topics. (It's called Generative because it generates stuff that previously was only created by humans. Like: writing, art, music, software code, etc.)
Depending on whom you ask, Generative AI will either usher in a new era of content creation, freeing up creative professionals to be even more creative by performing boring/repetitive-but-necessary tasks, or it will put them all out of work. The truth is probably somewhere in between.
Right now, the tools are mostly free and fun to play with. ChatGPT, which generates all kinds of text -- essays, poetry, fiction, song lyrics -- has become a viral sensation, with more than 100 million people logging onto it in just two months. (Stay tuned for more on ChatGPT in future posts.) And it already has people up in arms about plagiarism, cheating, copyright violations, and misinformation.
But in this essay, I'm going to talk about Generative AI art.
The best known artificial image generator is DALL-E. Its name is a play on WALL-E, the movie featuring the adorable robot -- as well as humans trapped on an interstellar cruise ship who have to ride around on scooters because they're too obese to support their own weight (1) -- combined with the name of the famous Spanish surrealist.
Wall-E meeting Salvador Dali in front of a Walmart. Source: DALL-E
DALL-E creates new images from existing ones, using what's known as a Diffusion Model. Essentially, it deconstructs an image to learn its basic properties and then attempts to recreate it. No one knows exactly how the model does that, they only know what it produces. It's a bit like handing a picture of the Mona Lisa to a five year old, who tears it into pieces, then tries to glue the pieces back together in the right order while also scribbling on them with a crayon.
The results can be, well, the word disturbing comes to mind. Here's some of what DALL-E produced when I asked it to draw a picture of Dolly Parton:
To me, that looks more like a drag queen doing a not-very-convincing Tammy Faye Baker. Your mileage may vary.
I then asked Craiyon, a free image generator based on the DALL-E engine, to produce an image of Dolly Parton singing "Hello Dolly" as drawn by Salvador Dali. It took about a minute.
(While I waited, the site showed me a "holiday inspiration" video featuring Christmas ornaments, polar bears wearing wreaths of holly, winter wonderlands, and zombie elves -- presumably, art it had recently created for other intrepid users. Apparently, Craiyon is badly in need of a calendar. )
Here the results were not as bad as I'd feared, though they look more like something Ralph Steadman (2) might produce:
I get it. The images are not supposed to be photo realistic. I imagine if they were, Dolly Parton's lawyers would have something to say about that. But what are they supposed to be? Right now they're mostly a curiosity.
At the very least, Generative AI raises intriguing questions about what it means to declare something "art" -- a debate that's been going on since the invention of photography made realism in painting kind of redundant, which then lead to impressionism, cubism, surrealism, and so on. (Is that glued-together Mona your five-year-old created art? Yes, if the right people say it is.)
We used to think that humans were the only animals that made and used tools. Jane Goodall proved us wrong there. Now our belief that only humans can create art is being tested. Maybe AI will invent brand new forms of expression, or different ways of looking at the world.
We already know that machines can be smarter than us. Perhaps we will discover they also have more soul.
(1) Imagine a Walmart in the year 2147.
(2) Admittedly, Ralph owes a lot to Dali.