SCOTT DETROW, HOST:
We wanted to spend some time talking and thinking about artificial intelligence, which is the process of using machines to mimic human intelligence. By now, you may have heard about the debates that AI technology like ChatGPT have brought up when it comes to plagiarism, but it's also raised questions in another area.
SHIRA PERLMUTTER: Copyright law protects works of authorship. And that includes books or movies, music, sound recordings and works of art. The copyright protection gives the author exclusive rights to make certain uses of those works.
DETROW: That's Shira Perlmutter, who directs the U.S. Copyright Office. She says authorship is central to securing a copyright in the U.S., and right now, that means human authorship. But Perlmutter is seeing more artists trying to protect AI-generated work.
PERLMUTTER: One of the early cases claimed copyright protection in a work that the applicant said was generated entirely by a computer called the Creativity Machine. And we rejected the application. We said it could not be registered because, in the applicant's own representations to us, there was no human authorship involved in the creation process.
DETROW: Perlmutter admits that this brings up tricky questions, and that has real effects on artists looking to copyright their intellectual property. That's why her office has launched an initiative to learn more about AI-generated art and use that to issue new guidance.
PERLMUTTER: We want to learn as much as we can as quickly as we can, and to help people along the way to ensure that they're able to protect their rights in the new environment and also able to enjoy the benefits that the new technology offers to them.
DETROW: We wanted to get perspective from someone who actually uses AI technology to create art and how they think about these types of questions. So we called the anonymous digital artist, who goes by the name Claire Silver. She's been pretty successful. She has used AI technology to help create her piece called "Blood In The Streets, Late To The Ball." The physical piece of art was sold at auction in London for more than 40,000 pounds, with an exclusive digital NFT going to the buyer. And she joins me now to share her thoughts and perspectives on art, AI and what counts as authorship. Claire Silver, welcome to ALL THINGS CONSIDERED.
CLAIRE SILVER: Thank you. I'm so happy to be here.
DETROW: I guess I've always wanted to say this - if that's your real name - which (laughter)...
SILVER: It's - it might be.
DETROW: What can you tell us about yourself?
SILVER: Well, I can tell you that I am from a small cornfield in a flyover state, that I like being anonymous. I remember I watched "Harry Potter" when I was a kid, and I was so mad that Daniel Radcliffe did not look like how I pictured him in my head...
DETROW: Yeah.
SILVER: ...As Harry Potter. I would like to spare anyone that is a fan by just letting them imagine me how they would like to.
DETROW: Listen, as a radio person, I relate to that. So many people are like, You don't look like I thought you looked like. I'm sorry. Tell us about how you started your career as an artist and when you started working with AI.
SILVER: I had a prior career in something unrelated. I got hit with a chronic, life-changing illness, and that ended that prior career. Kind of sulked for a couple of years, chronically online and started searching for art and AI. Found a website called Ganbreeder, now called Artbreeder, which was one of the early, accessible, no-code AI tools anyone could use. I made maybe 40,000 images in - I don't know - a week, just constantly, 14 hours a day - wouldn't sleep, wouldn't eat, just obsessed - and then curated those down to maybe 20 and then shared a few of them on social media, unsure if anyone would see them as art like I did. And slowly, people started to. So I've just kind of been on a journey since then.
DETROW: Can you explain - I know it's probably different each time you make a new thing, but can you explain how the AI comes into it and either a hypothetical artwork or maybe a specific piece of artwork that you recently put together to help us understand what exactly we're talking about here when we talk about collaborating with the AI, as you call it.
SILVER: Yeah. It's sort of different each time because it's dependent on the tools, and the tools are constantly evolving.
DETROW: Yeah.
SILVER: So in the beginning, it was all curation. It was choose an image that AI has made that you like, and then sort of crossbreed it, like gardening plants or something, with another image to make something you like. And then kind of make a lineage down from there until you get something you love. Then it was painting into it by hand with transparent layers and using text. So you would type what you would want the part of the image that you highlighted to change into, and I would kind of do this glazing technique that way, low-opacity layers.
Now I've worked a lot with training my own models. So that means you feed it your own images or images that you've made with AI, and it learns what art is based on what you've given it. And so everything that that model makes looks like work that you've given it in some way.
DETROW: I'd love to get your response to some of the criticism that's out there. And I think by explaining in intricate detail how you create some of this artwork, I think you partially answered this first question. But what do you say to people who say if you're using technology, if you're using AI, you're not an author, the coding is doing the work for you?
SILVER: Yeah. Well, the first thing that needs to be understood is that there's a fundamental misconception on how AI works. People think that it searches the internet - say Art Station, for example, which is a very big website for artists - and kind of pulls pieces - like, collages pieces together by stealing to create an image that's kind of a collage, a photo bash of those things. That's not how it works.
How it works is it learns traits about things the same way that our minds do. And then it kind of combines all those traits together to make something new. So, for example, if I were to type that, I wanted, I don't know, a cyber-noir landscape with figures by John Singer Sargent, it doesn't take pieces of Sargent paintings and put them together to make something new. It learns that Sargent often paints figures, that figures have two hands with five fingers, that fingers bend at joints, that joints work like this and that Sargent often paints them with this sort of brushstroke or that quality of lighting.
And it takes all of those things that it's learned, and it creates something new, which I think is how our minds work. I think that's how influences work in general. So maybe that's a fine line of distinction for a lot of people. For me, it makes the morality of things very clear. I think it just mimics our brain. But because it's so efficient, people assume that it's stealing.
DETROW: And the Copyright Office is seeking public comment right now as it tries to come up with a guidance for issuing copyrights. Is there anything else you would want to tell somebody who's trying to come up with the rules here?
SILVER: I would say to please speak with the people who are making these systems - Emad of Stability AI is a good example - to understand exactly how they work and exactly the role of data in the process, and understand that we're at the point where AI can read images from MRI brain scans and reinterpret them as similar images. I don't see how traditional copyright will be able to hold up in the coming decade. With technology like that, I think you need to rethink how we look at influence, how we look at authorship. And there are a lot of people that are very passionate about helping you do that in order to not fall behind other countries. I think that it's important that you do.
DETROW: That was the anonymous digital artist who goes by the name Claire Silver talking to us about her use of AI technology in the development of her art. Claire Silver, thank you so much.
SILVER: Thank you so much. It was an honor.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.
