The Copyright Office takes on the sticky issue of artificial intelligence

Tom Temin: And let’s begin with how you went about research on this because you got quite a bit of input from the public.

Emily Chapuis: We definitely did. So the office, couple of years ago, in 2023, undertook this multipart study about artificial intelligence and what it means for copyright law and policy. We initiated it by issuing a notice of inquiry seeking comment on a broad range of questions. And we got a huge response, as you said. We got over 10,000 comments, I think closer to 11,000 written comments, by members of the public, stakeholders, copyright owners, people from tech communities. Just a really wide range of commenters.

Tom Temin: And by the way, these comments weren’t what sometimes come in with comments, 10,000 copied and pasted emails of the same thing from one or two sources. You really got a diverse input here?

Emily Chapuis: We really did. It was very interesting. We got a lot of individual creators wanting to tell us about how advances in artificial intelligence technology were influencing their process or their livelihoods, both in positive and negative ways. And we also got input from governmental stakeholders from, I mentioned, tech companies, trade organizations, unions, everyone that you can think of, sort of who touches on creative industries or industries that rely heavily on copyright.

Tom Temin: And I guess you did it now, I’m guessing, because of the emergence of the generative AI, which is what is being used to create stuff both visual and written.

Emily Chapuis: Yeah, that’s absolutely right. I think everyone knows that over the past several years, generative AI technology has become more accessible and more ubiquitous in ways that mean people are using it, who’ve never maybe used this technology before. And the technology is changing really quickly. So in addition to this sort of understanding that societal and technological change, our office became aware of some of the different uses of generative AI in creating expressive works because we started getting registration applications or content that’s created using some of these technologies in all variety of different ways. And so our registration division was really on the front lines in looking at each of these applications and thinking about how the technology was used and in what context and what that means for questions of copyright.

Tom Temin: Right. So you had a practical interest in this in knowing whether and when to grant a copyright. Is it fair to say that if an algorithm is fed the precise same input by two different people, you’ll still get identical outputs from both efforts? If I load something into ChatGPT and you load the exact same thing, it’ll put out the exact same thing.

Emily Chapuis: So it can. It depends on the system. So most widely commercially available systems like ChatGPT or Midjourney use what’s called random seeds, such that if I put in the same prompt that you put in, we would get some sort of randomization process that would mean we would get different outputs. But that’s not a necessary feature of generative AI technology. So you can have a nondeterministic system where every time you put in the same prompt you get the same outcome.

Tom Temin: Right. So then the question then might be is it the person’s work or is it just the random work of something that is a machine and therefore not copyrightable?

Emily Chapuis: Yeah. And that’s the key question that we addressed in the second part of the report. We focused on copyright ability, meaning whether outputs created using generative AI are copyrightable and the sort of key inquiry there is one of human authorship. So it was the creative expression that of the human user or did the machine to the AI technology really stand in for the human in that circumstance. And so what the office says here and what we’ve said before is that human authorship is a core, bedrock principle of copyright ability.

Tom Temin: Right. So many questions. Otherwise you could get a billion copyright applications every time someone hits the return, so to speak, on a generative AI algorithm.

Emily Chapuis: We could, I mean, I suppose we could still get the applications, but part of what we’re trying to do in this report is provide some guidance to the public and courts and all kinds of stakeholders, including people looking to register their work.

Tom Temin: And by the way, was there congressional interest in this from members? I mean, you’re not GAO and you’re not the Congressional Research Service. Did anyone ask you to look in on this on the Hill or the members that do care about this?

Emily Chapuis: So this is a study that our office initiated ourselves. So without a specific request from Congress. However, a big part of our role is advising Congress on important issues that touch on copyright law and policy. And there has been a considerable amount of congressional interest in what we’re doing in these first parts of our reports. And we’ve seen that both in Congress and much more broadly.

Tom Temin: Sure. We’re speaking with Emily Chapuis. She’s deputy general counsel of the U.S. Copyright Office. And generally, you found initially that there doesn’t seem to be any legislation required, but that there is a rule-based story, an interpretive based way of deciding what is copyrightable.

Emily Chapuis: Yeah. That’s right. So we don’t recommend in the report that Congress take any action. And the reason for this is we think that copyright law is sufficiently flexible to deal with changes in technology. And that’s not just based on AI, but on the entire history of copyright law, has had to deal with these questions, whether it’s the development of the camera or the internet. The questions about copyright ability are always on a case-by-case basis. And the technology that’s used and how it’s used and what it’s used for are important elements of that. But the sort of defining legal principles aren’t different in this context than in those other ones.

Tom Temin: Right. So the human input idea then is kind of an eternal for copyright. How do you decide that? Is it a percentage of human input? Because the machine does a lot here. But you could say, ‘Well, the camera did a lot when it opened and closed the shutter and exposed silver halide. And then there was a machine process to produce that image. But it was the selection, the timing, the decisive moment.’ To quote Henri Cartier-Bresson, another French photographer. That’s really the issue here. The human input and not the machine input.

Emily Chapuis: Yeah, that’s right. And it’s hard to parse. I mean, we’ve had people ask, so what’s the percentage that has to be human created? And there’s not a clear answer to that, again, because it’s case by case. But also the question isn’t really amount as much as it is control. So who’s controlling the expression. And so one of the things that we try to explain is that even the same technology can be used in a variety of different ways. So you can use generative AI technology as a tool assistive to enhance the human expression or you can use it as a substitute for human expression. And so control is sort of the bottom line in terms of what we’re looking at to draw that distinction.

Tom Temin: And for the Copyright Office, the people that examine and in issue copyrights, what do they need then to know that something created with AI wasn’t fact having human input, etc. because it’s hard to tell from what it is you’re handed?

Emily Chapuis: Absolutely. So just about two years ago, the office came out with guidance for people seeking to register works created using artificial intelligence. And one of the things that we say in that guidance is you should disclose your use of that technology. When you file an application with our office, even if you’re not sure exactly what to say or where parts of your work fall in terms of one side of line or the other, just say what you did, right? Say I used AI to do X and it doesn’t have to be detailed. It doesn’t have to be comprehensive. It just has to be enough to put our office on notice. And usually what happens in those cases because this is a developing area, is that the copyright examiner, the registration specialist, will reach back out to the applicant if there’s questions and try to clarify exactly how this work came to be. So I mean, one of the points that we really want to drive home to people out there thinking about registering this kind of work is that our office is in the business of granting registrations, right? We want to encourage people to use the system and we want to be helpful in this process to the extent that we can, right? So more communication is better. And if we have questions, we’ll ask and if you have questions we want us to come and we want you to come to us as well.

Tom Temin: All right. And by the way, have you had any collaboration with the executive branch agency that does patents and trademarks? Some of the issues may impinge a little bit here.

Emily Chapuis: Yeah, we have. I mean, our office and the Patent and Trademark Office are in pretty regular communication. We collaborate in a lot of different ways in terms of copyright ability versus patent ability. There’s some different legal issues but looking at how they’re addressing these questions is of course very interesting and informative for us, too.