Can you copyright artwork made using AI?

0:01NPR.

0:11This is The Indicator from Planet Money. I'm Waylon Wong. You've probably heard about all the lawsuits against AI companies by groups of writers, the New York Times, and Getty Images. They've all alleged the big AI firms violated their copyright by scraping up their material to train AI systems. But there's another issue at hand too. And joining me to talk about that today is Emma Jacobs. Welcome. Hi. Hi, Emma. And you are a reporter in beautiful Montreal, Canada. I am. Thank you for having me. And that other copyright issue being raised by generative AI tools like ChatGPT and Dali is who owns the stuff they produce.

0:51Or even more complicated, Waylon, what if you use ChatGPT to partly create something and then a human creates the rest? Is it eligible for copyright? And if so, who holds it? You are breaking my brain, Emma. So today on the show, we look at these issues through the story of one guy who mashed up a photo he created with Van Gogh's Starry Night.

1:17This message comes from Vanguard. Capturing value in the bond market is not easy. That's why Vanguard offers a suite of over 80 institutional quality bond funds, actively managed by a 200 person global team of sector specialists, analysts, and traders. They're designed for financial advisors looking to give their clients consistent results year in and year out. See the record at vanguard.com slash audio. That's vanguard.com slash audio. All investing is subject to risk. Vanguard Marketing Corporation Distributor. This message comes from Dell Technologies. Your new Dell PC with Intel Core Ultra helps you handle a lot when your holiday to-dos get to be a lot.

2:02Luckily, the Dell PC helps you get it all done. Get yours at dell.com slash holiday. This message comes from Warby Parker. Prescription eyewear that's expertly crafted and unexpectedly affordable. Glasses designed in-house from premium materials starting at just$95, including prescription lenses. Stop by a Warby Parker store near you. A lot of the broad strokes of copyright law have actually been set by international agreements, which countries then implement. But international rules haven't been worked out yet when it comes to AI. That means countries must try and figure things out on their own, at least for now.

2:39I'll confess. I mean, I've got the best job in intellectual property law in Canada. David Feuer works on big challenges like this in Canada. He's general counsel at the Canadian Internet Policy and Public Interest Clinic at the University of Ottawa. SIPC for short. He's one of those people trying to figure out what the rules around AI and copyright should be. For a long time, intellectual property lawyers like David have thought of copyright as a system that creates incentives and rewards for humans to make things. There's lots of stuff out there that's beautiful. You know, the Grand Canyon is extraordinarily gorgeous, but there's no copyright in it.

3:13There's no author of that natural formation. It's got to be a work produced by a human, we would say, for there to be copyright in it. But he says works involving AI can follow along more of a spectrum of human involvement. Probably the easiest of the cases that we're going to get. I mean, the absolute easiest is just a text prompt that says, show me a cat. You know, and the AI pops up an image of a cat. In this situation, David and most other IP lawyers think you just won't get copyright. OK, so I won't be launching my illustrious art career that way. But, you know, it's like the human contribution, the human effort involved is just so minimal, right?

3:52Yeah. And an AI application, it doesn't really need incentives to make stuff. But at the other end of the spectrum, you might see more of a human AI collaboration. The question then becomes, how much human involvement would you need to get copyright? As David was mulling all this over, he was looking for a test case that might help clarify these difficult issues. And in 2022, he found one in Canada's copyright registry. The registration listed a human author, but also an artificial intelligence application as the authors. The guy who filed this registration was also an IP lawyer looking to test the rules on copyright.

4:32His name's Ankit Sani, and he lives in New Delhi in India. Now, Waylon, I don't know if you had a pandemic hobby. I just played Animal Crossing on Nintendo Switch 24 hours a day. That sounds terrific. Well, Sani's pandemic hobby was to run a series of international experiments around artificial intelligence and copyright law. Okay, so we relax in very different ways. Yeah, totally. But Sani started out by commissioning his own AI app, which can take one image and recreate it in the style of a second image. Next, Sani chose a photo he'd taken with his phone of a sunset. And using the app, he regenerated it in the style of a very old, famous painting in the public domain, Vincent van Gogh's The Starry Night.

5:19That's the one with all the blue swirls in the sky. I obviously tinkered around with the variables, which was how much style was to be transferred. This is Sani speaking on a podcast called Warfare of Art and Law. And then the output that got generated is what we ended up naming Suryast, which is the Hindi word for sunset. Sani then tried to register the new image with three copyright offices around the world, India, the U.S., and Canada. In his applications, he listed himself and the AI app as co-authors. How each of these countries responded shows just how complicated this matter is. In India, his application was granted, then withdrawn, but eventually upheld.

6:06In the U.S., it was rejected. And that rejection was upheld on appeal, which brings us back to Canada. Unlike the U.S. and India, registration is basically automated in Canada. There is no human review. You fill out a form, pay a fee, and click. You're in the registry. And it's important to note here, registration isn't actually what gives someone copyright. It functions more like proof if a conflict arises. Now, in the case of the Sunset Starry Night mashup, no one had challenged ownership of the work in Canada. But David had a problem with the registration itself. He said it was sending a message that AI work could be copyrighted.

6:47He and his colleagues first tried to reach out to Sani. To bring our concerns to their attention, to ask for a correction of the registration. And we just didn't get an answer. We, by the way, also tried to get in touch with Ankit Sani through his lawyer and a registered letter and did not hear back. But when David didn't hear back, he asked Canada's federal court to expunge the registration. And that seemed to get Sani's attention. Sani hired a Toronto law firm to fight back. One person who was pleased to see this go to court is Karis Craig, a law professor at York University in Toronto. When she heard Canada had let this image into the registry.

7:25I think I mostly rolled my eyes. But she also knew Ankitsani wasn't the only person running this kind of experiment. Past registry decisions have actually provided some of the first indicators of what AI work qualifies for copyright. There are lots of people who agree that these decisions could ultimately prove kind of critical in terms of how we see generative AI evolve and change our cultural landscape. Now, as we said, the U.S. office rejected Sani's application for having a non-human author. But overall, experiments in the United States have had mixed results. One applicant who gave an image generator more than 600 prompts also got rejected.

8:12But someone else who used an AI program called MidJourney to illustrate their graphic novel did get some protection. Most recently, I think there was an application to register a work that was called A Single Piece of American Cheese. I don't know if you heard about that one. I had not, but I'm definitely intrigued. This image looks like a mosaic of a woman's face with sort of melting cheese-looking hair. The U.S. Copyright Office would only grant the artist copyright protection for his arrangement of the AI-generated pieces of the mosaic. And if this all sounds pretty messy, that's because it is.

8:54But in general, the U.S. has tried to carve out the AI-made parts of works, which can't be copyrighted from the human-made parts, which can. And the difficulty is going to be whether it's possible or not, is to kind of parse what did the AI generate versus what did the human author or user create. Karis predicts this is only going to get harder as companies integrate AI directly into software we all use to write or draw or code. This summer, in the Canadian case, you'll have experts and lawyers all getting paid to dissect how Ankit Sani's Starry Night mashup got made and how much he contributed.

9:35But trying to figure out what's eligible for copyright day to day as the AI tools available also keep evolving, that's going to be much harder and really a moving target. This episode was produced by Julia Ritchie with engineering by Kweisi Lee. It was fact-checked by Sierra Juarez. Kicking Cannon is our show's editor, and The Indicator is a production of NPR.

10:15Learn more at capella.edu. This message comes from Solidigm. Yesterday's approach to storage can't meet the demands of today's AI ambitions. Bigger, faster, and more energy efficient, Solidigm solid-state storage solutions are optimized for AI. Learn more at storageforai.com. This message comes from EasyCater, a business tool for food. Helping organizations order food for meetings and events from favorite restaurants, set up meal programs for their employees, and manage food spend all in one place at easycater.com.