Editor’s Note: This article is part of our coverage of The Atlantic Festival. Learn more and watch festival sessions here.
Earlier this year, the technology company OpenAI released a program called DALL-E 2, which uses artificial intelligence to transform text into visual art. People enter prompts (“plasticine nerd working on a 1980s computer”) and the software returns images that showcase humanlike vision and execution, veer into the bizarre, and might even tease creativity. The results were good enough for Cosmopolitan, which published the first-ever AI-generated magazine cover in June—an image of an astronaut swaggering over the surface of Mars—and they were good enough for the Colorado State Fair, which awarded an AI artwork first place in a fine-art competition.
OpenAI gave more and more people access to its program, and those who remained locked out turned to alternatives like Craiyon and Midjourney. Soon, AI artwork seemed to be everywhere, and people started to worry about its impacts. Trained on hundreds of millions of image-text pairs, these programs’ technical details are opaque to the general public—more black boxes in a tech ecosystem that’s full of them. Some worry they might threaten the livelihoods of artists, provide new and relatively easy ways to generate propaganda and deepfakes, and perpetuate biases.
Yet Jason Scott, an archivist at the Internet Archive, prolific explorer of AI art programs, and traditional artist himself, says he is “no more scared of this than I am of the fill tool”—a reference to the feature in computer paint programs that allows a user to flood a space with color or patterns. In a conversation at The Atlantic Festival with Adrienne LaFrance, The Atlantic’s executive editor, Scott discussed his quest to understand how these programs “see.” He called them “toys” and “parlor game[s],” and did a live demonstration of DALL-E 2, testing prompts such as “the moment the dinosaurs went extinct illustrated in Art Nouveau style” or “Chewbacca on the cover of The Atlantic magazine in the style of a Renaissance painting” (the latter of which resulted in images that looked more canine than Wookiee). Scott isn’t naive about the greater issues at play—“Everything has a potential to be used as a weapon”—but at least for a moment, he showed us that the tech need not be apocalyptic.
Their conversation has been edited and condensed for clarity.
Watch: Atlantic executive editor Adrienne LaFrance in conversation with Jason Scott
Adrienne LaFrance: When we talk about AI art, what do we even mean? How does it work?
Jason Scott: So what we’re calling “AI art”—by the way, they’re now calling it “synthetic media”—it’s the idea of using analysis of deep ranges of images, not just looking at them as patterns or samples, but actually connecting their captions and their contexts up against pictures of all sorts, and then synthesizing new versions from all that.
LaFrance: So basically a giant database of images that can be drawn from to call to mind the thing that you prompt it to make.
LaFrance: And why is it exploding now? It seems like various forms of machine learning and AI have really accelerated in recent years.
Scott: They let it out of the lab and let regular people play with the toys. Across the companies that are doing this, some are taking the model of We’ll let everyone play with it now—it’s part of the world.
LaFrance: When you think about the implications for this sort of technology, give us an overview of how this is going to change the way we interact with art, or whatever other industries come to mind. For instance, at The Atlantic we have human artists making art. I’m sure they might have strong feelings about the idea of machines making art. What other industries would be potentially affected?
Scott: Machines are becoming more and more capable of doing analysis against images, text, music, movies. There are experimental search engines out there that you can play with and say things like “I need to see three people around a laptop.” And previously it would have to be three people in the laptop, but it actually is starting to make matches where there’s three people in the room. And the weirder and more creative you get with this toy, the more fun it gets. I see a future where you’ll be able to say, “Could I read a book from the 1930s where it’s got a happy ending and it takes place in Boston?” Or, “Can I have something where they fell in love but they’re not in love at the end?”
LaFrance: I have more questions, but I think now it’d be a good time to start showing people what we mean. Do you have some examples?
Scott: I have some examples of things that I did. So this is “detailed blueprints on how to build a beagle.”
“Detailed blueprints on how to construct a beagle” pic.twitter.com/blZSVJyMXd
— Jason Scott (@textfiles) July 15, 2022
LaFrance: So these are prompts that you gave the model, and this is what came out of it?
Scott: Yes. For the people who don’t know how this whole game works, it’s pretty weird. You usually type in some sort of a line to say, “I’m looking for something like this,” and then it creates that, and then people get more and more detailed, because they’re trying to push it. Think of it less as programming than saying to somebody, “Could you go out there and dance like you’re happy and your kid was just born?” And you’ll watch what happens. So it’s kind of amorphous. This is a lion using a laptop in the style of an old tapestry. This is Santa Claus riding a motorcycle in the style of 1970s Kodachrome. This is Godzilla at the signing of the Declaration of Independence. This is a crayon drawing of a labor action. These are bears doing podcasts. This is GoPro footage of the D-Day landing.
I’m always playing with it, and the reason you’re hearing all those strange prompts from me is because I want to understand: What are these systems seeing? What are they doing? It’s so easy as a parlor game to say, “Draw a cellphone as if it was done as a Greco-Roman statue.” But what about doing a bittersweet sky, or trying to draw a concerned highway? What does it see?
LaFrance: What does this suggest to you about the nature of art? This gets to be sort of an existential question, but is it still human-made art in the way that we think of it, and should we be bothered by that? I mean, we use all sorts of tools to make art.
Scott: Everyone is super entitled to their own opinion. All I can say is, I did drawings in a zine in my teens; I was a street caricaturist; my mother was a painter; my father does painting; my brother’s a landscape artist. And coming from that point of view, I am no more scared of this than I am of the fill tool or the clone brush [in Photoshop]. Everything has a potential to be used as a weapon—imagery, words, music, text. But we also see an opportunity here for people who never knew that they had access to art. I can almost hear the gears crack and start moving again when I go to somebody and I’m like, “Could you give me something to draw?” And they do it and they see how it goes. I can’t get angry at that particular toy. But I won’t pretend that this toy will stay in its own way neutral, or even is neutral now.
LaFrance: I was talking to a colleague about these sorts of tools the other week, and we were really compelled by the idea of being able to visualize dreams. What other sorts of things—fiction comes to mind—can we imagine but don’t normally get to visualize?
Scott: I love telling these AIs to draw “exquisite lattice work”—using phrases like exquisite or rare—or give me “leather with gold inlay on a toaster,” and watching it move into that world and design things in seconds that aren’t perfect, but are fun.
LaFrance: We’re going to experiment, which is always dangerous. You’re supposed to do stuff in real time. I have some prompts for you.
Scott: This is DALL-E. There are many others. Think of it just like early web servers or early web browsers. There’s a bunch of companies with various people funding them or doing things their own way.[Scott now leads LaFrance through a demonstration of DALL-E 2: It’s included in the video embedded above.]
Scott: We see the ability to do everything from intricate pen-and-ink drawings to cartoons. People are using it now to make all sorts of textures for video games; they are making art along a theme that they need to cover an entire wall of a coffee shop; they’re using it to illustrate their works. People are trying all sorts of things with this technology and are excited by it.
The Insidexpress is now on Telegram and Google News. Join us on Telegram and Google News, and stay updated.
- Adrienne LaFrance
- AI-art gold rush
- Art Nouveau style
- Artificial Intelligence
- Atlantic Festival
- Atlantic’s executive editor
- Colorado State Fair
- cover of The Atlantic magazine
- detailed blueprints
- easy ways
- Editor’s Note
- executive editor Adrienne LaFrance
- experimental search engines
- exquisite lattice work
- fill tool
- fine-art competition
- general public
- good time
- happy ending
- human-made art
- image of an astronaut
- Internet Archive
- Jason Scott
- Jason Scott Adrienne LaFrance
- live demonstration of DALL-E
- magazine cover
- parlor game
- part of our coverage of The Atlantic Festival
- people access
- pictures of all sorts
- prolific explorer of AI art programs
- recent years
- regular people
- Santa Claus
- sort of a line
- sort of technology
- strong feelings
- style of a Renaissance painting
- surface of Mars
- technology company OpenAI
- various forms of machine learning
- various people
- visual art
- whole game works
You May Also Like
- 6 min
- 4 min
- 2 min
- 3 min
- 5 min
- 8 min
- 5 min
- 4 min