Securities

How will AI art generators affect human creativity?

Description

Here is my dilemma, I want custom art for the podcast. Ideally, I would have a new image for each episode. Up until recently, my choices were to use stock photography or work with an artist. I now have a third option, AI art generators. As of late, my Twitter feed has been filled with these AI-generated images and I have become obsessed with them. On one hand, I feel like a superpower of visual communication has been unlocked and on the other hand, I am an artist and creative who wants other artists and creatives to get paid, I am conflicted.  DALLE-mini  What Kind of Sorcery Is This? Why code is so often compared to magic. Hieronymous Bosch fusion reactor #dalle2 - Josh Wolfe DALL-E makes some fantastic World’s Fair posters - Sam Arbesman "Interplanetary Space Empire by Thomas Cole" via DALL-E - Sam Arbesman

Transcript

This is a human-generated transcript, however, it has not been verified for accuracy.

Chris Gates:
Hello, welcome to Securities, a podcast about science, technology, and the human condition. I'm your producer, Chris Gates, and I have a dilemma. I want custom art for the podcast. Ideally, I'd have a custom piece for every episode. Up until recently, my choices to get this done were to either use stock photography or to work with an artist. But now I have a third option, AI generated art.

All right, so as of late, my Twitter feed has been filled with images that are generated by an AI. The most robust of these tools was generated by OpenAI. They call it DALL-E 2, D-A-L-L - E-2. You've probably heard of it. The name is a clever combination of the surrealist artist, Salvador Dalí, and the lovable Disney robot, WALL-E. Get it? From what I understand, and look, I'm no computer scientist or have ever worked with AI, so if I get it wrong, please send me a note, cgates123 on Twitter.

From my understanding, OpenAI and some of these other AI art generators scraped the internet for a bunch of images and fed this vast data set of images and the labels associated with the images to their AI. For more detail, here's how OpenAI put it on their website. "DALL-E has learned the relationship between images and the text used to describe them. It uses a process called diffusion, which starts with a pattern of random dots and gradually alters the pattern towards an image when it recognized specific aspects of an image." All right, I don't really understand how that all works, but I guess that's a part of the magic, right? So basically you give it a written prompt and it builds an image, pixel by pixel, until it creates an image that most closely represents the prompt. Like I said, my Twitter feed has been filled with these AI generated images as people get access to them.

As I've been watching them, I've become obsessed, obsessed on all different levels. On one hand, I feel like a superpower of visual communication has been unlocked, and on the other hand, as an artist and a creative who wants other artists and other creatives to get paid, I'm conflicted. So there are some ethical dilemmas here, and some other questions like, how does this affect human creativity? I guess OpenAI's hope is that, "DALL-E will empower people to express themselves creatively, and DALL-E will also help us understand how advanced AI systems see and understand the world, which is critical to our mission of creating AI that benefits humanity."

That's a really rosy picture. I would love for that to be the case, but I still have a lot of questions. So to help me navigate these questions, I got Lux's scientist in residence, Sam Arbesman, to talk to me about his experiments with DALL-E. He got access, along with a few other folks on the Lux team, and he's been experimenting in doing some interesting things and putting a bunch of thought into it. So I'm going to drop you all into where our conversation started to get good, okay? Ready? 3, 2, 1.

Sam Arbesman:
Josh and I were just talking earlier today, and so we've been swapping different things we've been doing. I asked him if he's been getting his kids involved, and it sounds like his kids are not really. My kids though have actually been really, really into it. To give you a sense of how much they're into it, my daughter this morning woke up, came downstairs and said, "I have an idea for DALL-E that I want to try." I'll just sit between both my kids, we'll sit on the couch, and they'll just come up with ideas and we'll take turns going back and forth thinking of things. They come up with crazy stuff. My son is particularly good at breaking it. He'll come up with prompts that it doesn't know what to do with.

My daughter did one... I have a whole bunch of stuff open. She asked for... it was digital art of a woolly mammoth skeleton on fire in the tundra, and it was good. It came up with some good stuff. It's also just funny seeing my kids, especially my daughter now that she is understanding the jargon of DALL-E, she'll say, "I want this, but in cyberpunk. Or like this," or whatever it is. And she doesn't know what cyberpunk is, but she kind of knows that it's a certain aesthetic. The art itself that it's generating is not always that interesting. Oftentimes the excitement is the prompt coupled with the art itself.

Chris Gates:
Yeah. In my research, a lot of the artists that have talked about working with DALL-E, is that the art is actually in the prompt engineering. The prompt engineering is crafting the incantation to the spell.

Sam Arbesman:
Yes. I've seen it. I've seen some of this kind of stuff. Actually, I wrote an essay about the ways in which code and magic have deep similarities. So I'm watching Star Trek: The Next Generation with my kids right now. When you look at them programming a computer, like programming the Holodeck, it's like, what do you do? You say, "I want," I don't know, "1940s San Francisco," or whatever. And you're like, "That's not how you program a computer." But the truth is, it is now. It's finally become a reality. It's insane.

Chris Gates:
That's so real.

Sam Arbesman:
The one episode where they do the Sherlock Holmes.

Chris Gates:
Yes. Where they create Moriarty, and Moriarty takes over the Enterprise.

Sam Arbesman:
Right. But it was because they said that they wanted an adversary capable of defeating Data-

Chris Gates:
Totally.

Sam Arbesman:
... not defeating Sherlock Holmes. It was like they messed up the prompt, and it was the wrong prompt, and suddenly it went all amok.

Chris Gates:
Oh my God, that's such a cool connection, Sam. I love that.

Sam Arbesman:
That's going way deep into nerdom, but for the three listeners, they're going to love it. If you want to go that way.

Chris Gates:
100%. 100%. Okay, so the other thing that I have been interested in is using artists' names in the prompt and how that-

Sam Arbesman:
Like as a personal genre almost?

Chris Gates:
As a personal genre. In the research that I did, that is where some of the artists had pushback.

Sam Arbesman:
Oh, interesting.

Chris Gates:
Maybe artists should have a right to not have their art ingested into these AI art generators. Coming back to what Josh had written in his prompt and saying, "I collaborated with this artist to create this."

Sam Arbesman:
Oh, okay. I like that. Right, because he did the Hieronymus Bosch, I did the Rodin statue-

Chris Gates:
Yeah, totally. Which I love.

Sam Arbesman:
The difference there is I still haven't done the statue. Everything is a remix. Related to this also, I was showing this morning... my daughter and I were watching, you know those Kurzgesagt videos where they explain something and so... Do you know this piece of it?

Chris Gates:
No.

Sam Arbesman:
Anyway, they're like explainer videos, but on big ideas. This one was about the long-term history of humanity, of going millions and millions... It was very interesting. As we start watching it, and I guess my daughter is now... has a heightened awareness of styles. She's like, "I really like the style that this video is in. What is that style?" And I said, "I think it's just their style. Their specific style." Now she's already thinking of how can I adopt that style? Obviously it's that person's style, but at the same time though, I feel like that's how artists learn. You don't ape a style, but you combine different styles and, "Oh yeah." And they figure out, "Okay, my style is a combination of," I don't know, "Monet and Hieronymus Bosch," or whatever. And writing it-

Chris Gates:
You may never put it out there, but you do get influenced by other artists and the things that are in your world.

Sam Arbesman:
Yeah, writing is the same thing.

Chris Gates:
Do you think that there's a conversation there? Is that worth talking about? Is the ethical implications using someone's name in the incantation?

Sam Arbesman:
Yeah. I guess there is, because there's the ethics of, "Okay, is the person dead? Is their work already in the public domain? Is their style basically emblematic of a certain style anyway?" So like van Gogh. That's a thing, versus you're choosing someone who, it's like a really deep cut, and most people are not familiar with it, and suddenly you're just generating things in that person's style. You don't have to mention an artist's name, but there are certain styles that just-

Chris Gates:
Like Impressionist.

Sam Arbesman:
... are more generic.

Chris Gates:
Yeah.

Sam Arbesman:
Impressionist, or watercolor and pencil drawings, or whatever it is.

Chris Gates:
Right. So then it's the question, if you use someone's name, do you need to put the prompt with the art?

Sam Arbesman:
The prompt with the art is the more interesting thing. A friend of mine did... he did a Hieronymus Bosch, The Muppets according to Hieronymus Bosch. It was freaky, but it was really funny. But if I hadn't known the prompt, it was almost like that was the caption on the joke. You needed the caption on that comic, otherwise it wouldn't have had the same power. So for me, it's the like the Rodin's Overwhelmed Man. Without the prompt, it's just a whole bunch of people who just look overwhelmed, which is, that is exactly what it is. But once it's like, "Oh, Rodin," then it's kind of funny.

Chris Gates:
Yeah, it's like a meme. It's like you have to know the context of the meme to really actually get it.

Sam Arbesman:
If you're trying to adopt that style, which is not a well known style, as your own, and then not let people know that's where it's coming from, that feels wrong. But there's a lot of other things you can do, which is all entirely separate from the IP issues of, "Okay, this thing is being trained on huge amounts... Shouldn't the artist be paid for it?" I don't know. But at the same time, are they going to be paid 1/1,000,000th of a penny? I don't know.

Chris Gates:
Yeah, but also it helps bring exposure to that artist.

Sam Arbesman:
Exactly.

Chris Gates:
There's another piece in my research. There's the question of, "Okay, what is going to happen to creativity? Is the tool creative or is the person creative?" And the comparison, that everyone used the example of the camera. You're not actually creating the art, it's just the machine. And then also, the giving increased access to photography and watching that and how it doesn't change, it just makes actually photography more ubiquitous and gives more people access to have creative tools. So I think maybe that is the right comparison, that it's a tool for creativity rather than a creativity killer or whatever.

Sam Arbesman:
Right. It's a collaborative tool. When I first got access to DALL-E, I went crazy and I was doing a million different things, and quickly bumped up against its limits, and then started having to be more careful about the different things I wanted to do. Then at a certain point I was like, "I think I got it out of my system."

Chris Gates:
Okay. The last piece that I was thinking about is signaling of value and how this could potentially affect the value of a drawn image. So, I think a lot about-

Sam Arbesman:
[inaudible 00:11:00].

Chris Gates:
... web design. Web design that when photography used to be a signal of value, it was like, "Okay, this is something we needed,"-

Sam Arbesman:
Right, and then we had-

Chris Gates:
... and then there were all these libraries came-

Sam Arbesman:
... stock photos.

Chris Gates:
... and stock photos, and then suddenly that became passe, and it's like now you need an artist to draw you something because it insignifies value over the stock photography. Well now with DALL-E and having access to unique drawn images so much easier, I would imagine that would then change the equation once again.

Sam Arbesman:
I've read some people think about that kind of thing. I definitely think there's something there. I don't know what that next thing is going to be, of what is the true mark? And maybe it's going to be some sort of video thing, which is even harder to do.

Chris Gates:
Yeah. Harder to do, yeah.

Sam Arbesman:
Or it's going to be the kind of... it'll have to be some very high resolution thing that looks real no matter how you... or not real, but it looks slick even when you zoom in, because a lot of the DALL-E stuff doesn't look slick. It'll actually still require a lot of effort. You can make something that looks kind of good, and then once you dig in you're like, "Oh, this is garbage and I need to fix it up." Which is why a lot of things I tweak them out, or at the low resolution, because if you saw just one, it's not good.

Chris Gates:
Yeah, like, "That's not good." Yeah, totally.

Sam Arbesman:
Yeah. I'm not sure we're quite there yet of like, we don't need the artist, but the artist will have a somewhat different role in terms of making things look good, as opposed to just making the things themself. I don't know.

Chris Gates:
It will change. Who knows?

Sam Arbesman:
Oh, yeah. My daughter, she loves art. She's almost nine and just loves drawing and doing all this kind of stuff and really interested in all this. I don't think at any point when we've been playing with DALL-E she's ever said, "Oh, what is the point of me drawing anymore, because now this thing can..." There was never once been that concern, which means she's viewing it as this is a tool for collaboratively creating more and enhancing creativity as opposed to diminishing things.

Chris Gates:
Totally. Okay, cool. All right, I think there's a lot there. I think we-

Sam Arbesman:
Okay, awesome.

Chris Gates:
... got a lot of stuff here, so I'm going to...

continue
listening