AI-generated art is disrupting lawmakers’ understanding of who a true copyright holder is; is it the human who prompted it? The company behind the generator? Or the artists who are part of the data set the generators use? Senior policy editor, Adi Robertson, explores the recent lawsuits between concerned creators and the U.S. copyright office.
Category
🤖
TechTranscript
00:00 I'm not an artist. I can't paint. But I can type, and in 2023, if I want to make a picture,
00:05 that's all I have to do. Computer chips, in the style of Salvador Dali.
00:12 Oh, so Dali just on a microchip. Neat. Not exactly what I was expecting, but still cool.
00:18 Well that was easy. If only I knew if it was legal.
00:28 Almost every piece of art you'll ever see was built on something else. Maybe it was made in
00:33 a tradition like Expressionism or Surrealism. Maybe it used a photograph for reference or
00:38 parodied another artist. Maybe it was created directly from other people's art, like a collage.
00:45 Whatever it is, if it was made in much of the last century, in the US, that art is likely also
00:50 protected by copyright. When the artist created it, they earned the exclusive temporary right
00:55 to control who makes money from it. The Constitution offers these protections
00:59 to promote the progress of science and useful arts. There's a paradox in this system. On one hand,
01:06 letting people control and profit from their art helps encourage them to make more of it.
01:09 On the other, culture is a common good. People can build on it in ways the original artist never
01:15 intended, and the law includes exceptions to make that possible. So copyright law is a constant,
01:21 delicate balancing act. And when a new technology comes along, it can throw that balance into flux.
01:27 Artificial intelligence is one of those technologies. And right now,
01:30 it's throwing a monkey wrench into just about every step of the copyright process.
01:34 So in October of 2022, someone actually tagged me and they had made an AI-generated image using my
01:44 name. I remember it being black and white, which almost all of my work is, and it kind of reminded
01:51 me of a style I use for a series called Fangs. So I felt like I recognized the contours of my style
01:58 in that image. This is Sarah Anderson, an illustrator, cartoonist, and one of the people
02:03 on the vanguard of the fight over AI and copyright. In January, Anderson and two other artists,
02:08 Carla Ortiz and Kelly McKernan, sued the AI image generator company's Midjourney and Stability AI
02:13 for copyright infringement. That incident kind of forced me to directly confront the issue because
02:19 I was then involved. Their suit took aim at one of the central elements of generative AI,
02:24 training its models. AI image, text, and audio generators, like stable diffusion,
02:28 the system I was just using, are basically computer-generated sets of rules for creating
02:33 work similar to a huge set of previous examples. Most AI companies are cagey about where they get
02:38 their data. But for an image generator, the collection might look something like this.
02:43 This is Lion5b. It's a database composed of 5.85 billion web addresses for images,
02:49 scraped from across the internet, plus captions describing the images. Generative AI companies
02:55 use or build databases like this one to train their systems, and many likely contain lots of
03:00 copyrighted work. The evidence? Among other things, they can often reproduce famous photos
03:05 or book snippets. Creators can also search for their work in datasets thanks to sites like
03:10 Have I Been Trained? And some have seen their names get used as prompts. For some of them,
03:15 the experience felt violating. If a generator is trained on your work and then can imitate your
03:22 style, someone could create images you don't approve of. They could create sexualized images,
03:27 for example, and that can have real and unfortunate impacts on artists' careers,
03:32 not to mention just the fact that people are going to people already sell merchandise using
03:39 artists' names in their work. And lastly, there's definitely a general malaise in the art world.
03:45 I never saw this technology coming, and a lot of other artists didn't, and it has definitely
03:51 dampened the spirit, although not killed it, certainly.
03:56 In other cases, they worried they were losing commissions to AI copies of their own work.
04:01 That's raised the obvious question, is this legal? The answer is, right now, nobody knows for sure.
04:08 More on that in a second. First, here's a word from our sponsor.
04:11 More and more, we're seeing AI tools be integrated into our daily lives,
04:15 from generating quick, inspiring art to capturing notes from important meetings.
04:19 But with SAP Business AI, their tools are designed to deliver real-world results,
04:24 helping your business become stronger and helping you make decisions faster.
04:28 This revolutionary AI technology allows you to be ready for anything that is thrown at you.
04:32 Okay, that's it for me, but before we go, SAP doesn't influence the editorial of this video,
04:38 but they do help make videos like this possible.
04:41 Okay, you're probably watching this on YouTube. You've probably heard the term "fair use."
04:45 It's a set of exceptions under which you can use copyrighted work without paying or getting
04:50 permission from the original creator. It's a tool for striking that balance between
04:54 giving individual artists control of their work and letting people build on common culture.
04:59 AI companies like Stability and OpenAI call model training a classic case of fair use.
05:04 The process takes something that was created for one purpose, like a picture designed to
05:08 be admired in an art gallery, and uses it for the completely different purpose of analyzing art
05:14 or producing new work. And there is precedent for this argument.
05:18 In the 2000s, for instance, an adult photography site called Perfect 10
05:22 sued over Google Image Search, saying it was "scooping up their subscriber-only pictures
05:27 and delivering thumbnail versions for free." Meanwhile, the Authors Guild filed its own
05:32 suit over Google Books, which was scanning millions of books and letting people search
05:36 through short snippets online. Google was clearly using copyrighted work without permission,
05:42 but courts found it provided significant, helpful new services that outweighed the
05:47 potential financial damage to authors and photo sellers. It had the legal right to
05:51 build on millions of copyrighted works. It's possible AI companies do, too.
05:56 That said, Google Image Search wasn't trying to replace adult photography sites,
06:01 and Google Books wasn't trying to produce its own books. Generative AI is a new technology,
06:06 and when copyright confronts a new technology, anything can happen.
06:10 One possible analogy would be the development of movies. In an old case involving the novel
06:16 Ben-Hur and a movie version, the courts said yes, although movies didn't exist
06:22 and were hard to imagine when the book was written. Nonetheless, the copyright includes
06:29 control over this new space. With other technologies, we've been more skeptical.
06:35 The courts at first said that player pianos didn't invade the copyright on music,
06:42 and then Congress wrote a compromise into the Copyright Act where you could make player piano
06:48 roles out of popular songs if you paid a royalty. And both of those are on the table for generative AI.
06:54 Even if a company could be generally in the clear for training in copyrighted work,
06:58 a user could infringe copyright with a specific creation, like say an almost exact
07:04 imitation of a very famous photo or a sequel to a best-selling book. Long story short,
07:10 AI companies are arguing that they've found an innovative new fair use of art.
07:14 Some artists are pushing back. But a lot of these services users are wondering a very different
07:19 question. Did I make this? And if I want to protect it with copyright law, can I?
07:24 Computer scientist Stephen Thaler didn't start his work hoping to become a pioneer of AI copyright,
07:32 but in 2018 he asked to register an image titled "A Recent Entrance to Paradise."
07:37 The creator? A system he called the Creativity Machine, part of a decades-long project to
07:43 demonstrate computers can be sentient. My case is that we have a machine,
07:48 machine intelligence in fact, that is building up complex concepts totally on its own.
07:56 In a series of decisions and a recent lawsuit, courts and the U.S. Copyright Office have
08:01 disagreed on very simple grounds. Their ultimatum? No human, no copyright, period.
08:07 The Copyright Office's current position is that an AI is not a human and that only humans can be
08:13 authors. So if you type a short prompt into an AI, there is no copyright on the resulting output
08:21 because it's not the result of human authorship. The legal precedents underlying this are honestly
08:26 weirder sounding than a debate over whether computers can make art. Have you ever wondered
08:30 whether God could get a copyright? Well, the answer is no. Can you copyright the way a garden
08:34 grows? Sorry. And a monkey that started playing with a camera and took a selfie? His name was
08:40 Naruto, and a hard-fought battle determined that it wasn't his. Thaler is still fighting his case,
08:45 and it's been hugely influential on the issue of AI copyright. But it's not directly addressing
08:50 the question lots of people care about. The Creativity Machine doesn't work like a prompt-based
08:55 AI image generator. It produces art pretty much autonomously, and Thaler's goal is to establish
09:01 that it can earn a copyright with almost no human involvement whatsoever. For most people who are
09:06 using systems like stable diffusion, the question isn't whether that program can own an image.
09:11 It's whether they can create an image through it and own whatever comes out.
09:14 This is Teatro de Opera Espacial, or Space Opera Theater. In 2022, a man named Jason Allen created
09:22 it with the image generator Midjourney and submitted it to an art show at the Colorado
09:26 State Fair. He won, and in order to protect his claim to it, he applied to register the copyright.
09:32 The Copyright Office saw news coverage of the story and turned him down. It applied the same
09:38 basic principle. No human, no copyright. The problem is, Allen isn't a hands-off computer
09:44 scientist. He's a Midjourney prompt engineering obsessive. Allen estimates he used about 600
09:50 prompt variations to get the output that he wanted. He sees Midjourney not as a program
09:55 creating its own art, but as a tool similar to Photoshop. And while I feel safe guessing that
10:01 the creativity machine isn't personally sad about being denied its copyright claim, Allen says not
10:06 having his pretty famous work registered has created the same problems it would for any artist.
10:11 Allen's not the only person in this boat. The Copyright Office has denied at least one other
10:16 AI-generated work, a comic book created by Christina Cashtenova, also using Midjourney.
10:22 The Office agreed Cashtenova could own the human-written script, and even the way she
10:27 arranged AI-generated images on a page, but not the images themselves. I think their position
10:33 is also completely unsustainable, and this cannot last more than a couple years. We are simply going
10:39 to have far too many people making far too much that's interesting and valuable for us to say
10:46 this is a copyright blank space. There is no copyright in any of this. And on top of all
10:51 these concerns, there's a pragmatic problem. In all three cases I've mentioned, the Copyright
10:56 Office rejected the works because they learned they were AI-generated. But the lines between
11:00 digital art and AI are blurring as programs like Photoshop introduce AI elements. And if artists
11:06 don't mention somewhere that they use AI, it might be hard for the Office to detect its use.
11:11 The debate over AI copyright is moving fast. What probably won't change is the stakes.
11:17 Behind most art, there's a human who cares about its creation, who might depend on it to live.
11:22 Ahead, there's a world of unexpected possibilities for adapting it to make something new.
11:27 And somewhere in the middle, there's a few hundred years of lawyers and judges writing about
11:32 monkeys, gods, and machines. And as for this?
11:46 Never mind.
11:46 So what do you think about the balance between artists and AI creators?
11:51 If you want to learn more, go to theverge.com or stay tuned to this channel.
11:55 We'll have more coverage about AI soon.