The human in the machine: What is ‘authorship’? [Fireside chat]
James Gibson; Matthew Jockers; and Sarah Hartman-Caverly
Fireside chat featuring James Gibson and Matthew Jockers, facilitated by Sarah Hartman-Caverly.
- Jump to the recording
- About James Gibson and Matthew Jockers
[Hartman-Caverly] Please join me in welcoming James Gibson, who’s the sesquicentennial professor of law and director of the Intellectual Property Institute at the University of Richmond Law School. And with him, joining him is Matthew Jockers, who’s the distinguished research scientist and senior engineering research manager working in the world of machine learning and AI at Apple. I’m going to get them both started with our question, which honors Heterodox Academy’s tradition of intellectual charity and intellectual humility: I’ll invite you both to share a view regarding human authorship, regarding the world of copyright, regarding our traditional economic incentives around making your way as a creative in the United States, with which you actually disagree. Steelman or present the strongest case for that perspective, and then share us the alternate view that you ascribe to and give us some reasons for why you ascribe to that view.
And I think, because I’m assuming with his background in law that James Gibson is very adept at steelmanning opposing views, maybe we’ll throw this question to you first, Jim, and then we’ll invite Matt to respond and we’ll see where things go from there. Thanks so much to you both for joining us.
[Gibson] Yeah, thank you, Sarah. and Craig and everybody else who helped organize it. I enjoyed the first session. Hope we can live up to the promise of that.
I don’t know, lawyers might be better at strawmanning than steelmanning, but I’ll give it a try.
I would say that the view about the intersection of copyright and AI that I disagree with, but I can make the case for, is the notion that copyright law in the first instance is something that significantly pushes human creativity forward. In other words, it creates incentives not just to create, but to sort of paradigm bust and push the envelope. And that therefore, the perhaps loss of the importance of the copyright system that AI might precipitate is something that will probably materially affect the output of sort of human creativity and particularly not quantity, but quality. And I think that is, to steelman a little bit, a view that is conceptually consistent with what we think of as copyright law. It’s supposed to set up incentives for people to choose the creative arts over alternative things they could do to make money. And obviously, human culture is pushed forward very strongly by people thinking about new ways of expression, whether it’s in film or literature or art or what have you.
I guess the reason I don’t necessarily buy into that argument is that that’s sort of conceptually promising, but in operation, copyright law, when we look at the actual requirements to get something protected by copyright, for example, when we look at the length of time that copyright protects a work, is a lot more about protecting sort of the everyday and the sort of quotidian rather than the true paradigm-busting sorts of innovation. Not that it doesn’t capture those, but most of the mill-run works that we see created and frankly valued in the marketplace that copyright creates tend to be relatively recursive, tend to take existing IP and recast it rather than really bust out something new. Certainly the most profitable forms of copyright—big Hollywood feature films for example—are usually not the ones that are considered the most innovative and artistic. So it’s not that copyright doesn’t have room to embrace the notion of paradigm-busting and bleeding-edge creativity, but I don’t think in operation that’s really its main purpose.
And so I don’t know, Matt [Jockers] has thoughts on the cost to paradigm-busting that AI might pose, which I agree is a really important issue and a really important challenge that AI presents, but I’m a little less confident that copyright law—maybe the law in general can help, but copyright law in particular—is going to provide much of a backstop against that AI threat.
[Jockers] Do you want me to jump in, Sarah?
[Hartman-Caverly] Yeah, please do. Thanks so much.
[Jockers] Awesome. Good provocative setup, Jim. Thank you.
Before I start, though, I want to add a little caveat, since you introduced me in my Apple capacity. I am not here in my Apple capacity today. I am a former academic, former dean, founder of a company that uses AIML to help authors develop better manuscripts. So that’s who I’m representing here today and not my day job.
Jim, I love the conversation that we’ve had in advance of this meeting as well. Really interesting to get your particular legal take on some of these things. If I were to summarize what I think you were saying right there is that copyright really isn’t an incentive for authors to take risk. Am I getting that right? Yeah. And in my write-up, thinking about this, I made the argument that the opportunity for copyrighting gives the author a certain willingness, perhaps, to take that risk because there’s this promise, at least, that it’s going to be protected. But I think I’m supposed to take the opposite stance, which is easy for me to do here, because I don’t actually think I have the answers to any of the questions we’re talking about today. It’s a fascinating, fascinating time, what’s going on. And it is scary on the one hand and exciting on the other.
I’m going to take a really provocative stance then on this copyright thing and say that even human- generated authorship is not unique and original. But I’m going to borrow from Robert Sapolsky, who’s recently written this unbelievable, mind-altering kind of book called Determined,[1] where he’s basically making the case, and this is a very drastic simplification, but making the case that we are effectively biological machines and everything we do is programmed, and that would include our output. So I’ll take that stance and say, so in that sense, nothing’s copyrightable.
[Hartman-Caverly] So let’s jump in on that idea, because human authorship is kind of one of the bellwether standards of what is copyright eligible. But this isn’t the first time that technology’s forced us to confront these questions about the human capacity for creativity, for expression. So what are some other examples that you’ve both thought about or come across in your own work? To what degree are those useful for thinking about the current moment of generative AI and other, you know, artificial general intelligence, all those kinds of things? And then where maybe is there some divergence? Do you think of generative AI as really a departure from things like the printing press, the internet, the smartphone, those kinds of things? Where’s your thinking on that?
[Gibson] I’ll jump in first. I do think that this is a very, very different technological challenge than we’ve encountered before in the creative space.
I think, first of all, the speed at which it’s arrived, I think, is a little bit different. You know, the printing press was perhaps the most disruptive technology in terms of human expression that preceded AI and had huge consequences and arguably is the reason copyright law was invented. And then obviously the rise of digital technology and then digital networks technology and sort of Web 2.0 were huge disruptions as well. But to me, those strike me as disruptions in degree more than disruptions in kind, because they didn’t necessarily implicate the scarcity issue as much as AI does in the sense that books became less scarce because of the printing press. The distributional costs of getting one’s creative works out into society changed dramatically because of the Internet, essentially.
But neither of those, I think, affected the costs of creativity in as dramatic a fashion as AI has, and I don’t think either of those were complete one-on-one or close to one-on-one substitutes for the product. And so really the only kind of comparisons I can draw when it comes to a question of human authorship, one’s just a little quirky and the other is sort of copyright law nerdy.
The quirky one is the famous monkey selfie[2] from about, what was it, 15 years ago?, where someone put a camera that could essentially be be triggered— this has happened a couple times—triggered by wildlife, and a lot of wildlife came up and sort of stared into the camera and triggered the sensor and pictures were taken. And the question was, is that copyrightable, or should it be copyrightable? Which was an interesting thought experiment about how central human authorship is to copyright, but was hardly, you know, an existential threat to established industries. It was useful as a thought experiment, but not particularly consequential regardless of which side you came out on. So that’s the quirky story.
The other sort of nerdy copyright way to think about the challenge of human authorship is that when we think about copyright law—and most of the conversations, including the conversations that Matt and I were having in print before this session—we tend to think about human authorship as capturing human creativity and human originality. And copyright law does, in fact, think in terms of creativity and originality, although the threshold for what counts as creative and original is very low. But the much more technical move that copyright makes with the concept of authorship is not actually about the highfalutin’ concept of originating in the human brain and reflecting the sort of humanity of the source of the creativity. It’s that authorship primarily in copyright law is a concept we use to decide who owns the copyright in the first place. It’s just a very sort of ministerial way of saying, we have this word ‘author’ because someone has to start using the copyright, so we’re going to use the term ‘author’ to describe that person.
But when I say ‘person’, that’s actually inaccurate, because authorship in copyright law could be corporate.[3] If you work for an employer and it is part of your job to create amazing original expression, it turns out that you’re not the author under copyright law: it’s the corporation that’s the author. They start out owning the copyright and you have no interest whatsoever under copyright law in whatever it is you produce. You just get paid a salary, presumably. And I know that’s super technical kind of boring copyright stuff, but it does, I think, illustrate that the law actually is already perfectly comfortable with situating authorship in something other than a human. In the same way that we situate free speech rights in something other than a human, in the same way that lots of areas of the law actually accord these things that we think as being essentially human rights or attributes, in fact, in non-humans and particularly in corporate entities. So I don’t know, Sarah, if that’s exactly responsive to your question, but that’s where my mind goes.
[Jockers] I’d like to follow on with a question for Jim here. Based on what I think you were just saying, the artifact—I’m not even sure I’m going to be able to express this right, but what I was thinking is—what if I’m the one doing the prompting? I am creating through that prompting, and my materials that I’m creating my art with are somewhat already pre-contrived, but that’s always the case. The colors were picked for me. I mix them in my painting. So is there a world you can imagine where this, it doesn’t matter? It’s whoever claims ownership of the prompted result?
[Gibson] That’s sort of what the copyright law is playing with right now. In other words, the Copyright Office won’t let you register a copyright if it’s, I don’t know exactly what the language is, but if it’s primarily an output of AI. But if you can say, “No, I actually iterated this through a series of prompts to get to the result I wanted,” then they will let you sort of claim authorship and register it.
And you’re absolutely right, Matt. This is not new. I mean, the way that it happens and the efficiency with which it takes place is new. But, you know, all art, all expression is derivative. And all art is essentially selecting components and arranging them in a particular way. Even writing a novel, you’re not making up the words, you’re not making up the syntax or the grammar, you’re selecting words and putting them in a particular order. So it’s definitely true that one can claim human authorship and creativity even though what you’re really doing is essentially riffing on something that came before, and so I think that’s where the the Copyright Office is now.
But let’s keep in mind that with the decision that the Copyright Office is making—the Copyright Office is actually not the final authority, I think, the courts are, but the Copyright Office gets the first bite at it because that’s where people to go to get their copyright registered—what we’re really doing is we’re deciding, at a legal level, should anyone own it or should no one own it? The ‘should anyone own it?’ is, there’s enough human input into the results through the prompting that you describe, that we’re going to decide it’s a product of human ingenuity and therefore it’s copyrightable. But the alternative is, it’s essentially public domain from the beginning. Because if copyright doesn’t attach to it, then it simply means that people are free to copy it and riff on it themselves without worrying about copyright at all. And in a world in which we’re actually perhaps getting rid of the scarcity of copyrighted goods, it could be that there’s really no instrumentalist reason anyway to assign ownership.
In other words, copyright—speaking of friction, which was something that came up last talk—copyright is friction in the system that prevents people from using expressive works in ways that they would otherwise be free to do. And one of the central kind of conceptual challenges that AI is bringing to the fore is, is this the end of the necessity for friction in the system? Because the friction as currently constituted is supposed to be creating an incentive to create. The friction allows you to charge licensing fees and make a living as an artist or author. But in a world where the incentive is no longer needed to have an adequate supply of creative works, then the friction is all cost and no benefit.
[Jockers] I want to come back, if I may, to probe a little deeper something that you were just saying about the manner in which the creative goes about their work and the selection. I’m going to stick to writing here to keep it in my world. There’s this famous story about James Joyce when he was working on Ulysses and Frank Budgen and his friend shows up to check in on how the how he’s progressing.[4] And he says, “How’s it going, Jim?” And Joyce shows him this sentence and Budgen and reads it. And he says, “Yeah, it looks good to me.” And Joyce says, “Yeah, they’re the right words. They’re just not in the right order yet.”
That to me gets at the heart of something I was trying to articulate in our pre-conversation about that intentionality. And you called me out on this beautifully, because I’ve always been against the intentionality argument as a literary scholar. You can never know exactly what the author was … but if we remove the intentionality component from this, where are we left?
[Gibson] Just to bring everybody up to speed on what Matt and I were talking about before this session, if we’re trying to decide what should be protected, what what sorts of human output should be protected, one of the ways that we were riffing on it was, well, what does the art world consider art? And there is a strain—and art, not just painting, but the creative arts generally—and there’s one, and I’m vastly oversimplifying here, and I’m sure Matt, you know a lot more about this than I do, but one of the strains is: it’s art if the author intends it to be art. And so we don’t actually, that’s a very egalitarian way of thinking about it. Your kid’s picture on the refrigerator is every bit as much of art as a Picasso is. But of course, post-literary criticism—and again, Matt, I’m treading much further into your world than I probably should, but basically, has strayed away from thinking about intent of the artist and brings the audience into the discussion a lot more.
But there’s other theories of how do we define art and value expression? Another is formalism: it has certain characteristics. Music has rhythm and tempo and notes. Or just institutionalism, which is a way of saying, the received wisdom of the art world decides what is art. And to the extent that that’s relevant to the question of AI creativity and whether we shouldn’t value it because it’s not artistic, it really depends on our decision as to which of those—and I’m sure there’s others—definitions of art we’re going to use.
And we have to ask these philosophical questions. And I’m glad the last panel was asking the philosophical questions of, can AI have intent or does it not matter as long as what it produces looks like the kind of thing that has the formal attributes of human expression? Or are we actually making essentially a deontological decision that, as the institutionalists who control the definition of art or as the lawyers who control the definition of what’s copyrightable, we’re simply saying we don’t care about its characteristics, if AI produced it, it’s not. And that’s just a policy decision and not an instrumentalist or consequentialist decision that’s more about human dignity than it is necessarily about maximizing the output of quality expression, which is what we say copyright is supposed to be about.
[Jockers] I love that comment because isn’t the rub, that we have no way of policing it? I worked on authorship attribution for a long time, so I feel very comfortable in this space. It was sort of a “solved problem” in certain unique kind of circumstances: you have a set of potential candidates and you want to figure out which is the most likely among those candidates to have written document X.
This is a whole different ballgame, what we’re what we’re dealing with now. Detecting—people are working on this problem very hard. But I’ve gone back and put things that I wrote 30 years ago through these tools and, “oh, 20% AI” or whatever it tells me. So if that’s the job of the legal establishment to determine whether it’s authored, they’re never going to win. So now what?
[Gibson] I think you’re exactly right, which is why I think that copyright law is not going to save us. Copyright is not going to answer this question. Because if it’s not already—and you know better than I, because you’re the one who’s using AI in the authorship space—the idea that we could somehow detect what’s created by AI and what’s not, if it’s even possible now, is not going to be possible in five years. Again, which is why this is a values and human dignity question, not a question of, consequentialist outcomes.
But Matt, I’m interested in, you’re currently using AI to aid human authors: is this a short-term, medium-term, or long-term business proposition? What’s your anticipation about what human authorship is going to look like with AI assistance in the next three, five years?
[Jockers] All right. Well, thank you for that question. So the company that I founded seven years ago, it’s called Authors AI.[5] This was pre-LLM days, pre-generative AI when I founded this company, and the idea was to leverage machine learning—and many of the things that I had worked on as a scholar in my academic life—leverage some of that to help authors see where their writing exists in the ecosystem of other published work, to help them see how, for example, they were utilizing a particular theme or emotional language and so on. The point is, none of what Authors AI provides is generative. It’s all about an analysis. It’s giving you the kind of feedback that you might get from an editor or a reader who looked at the book and said, “Oh, I really think this character dies out too soon” or whatever. And I left the company when I joined Apple. That would have been a big conflict of interest. But the company continues to maintain that very principled stand that they are not a generative—they are not helping authors to generate content.
The challenge then becomes what happens when the authors start feeding generative content to the machine to get feedback on the generated content. I think that’s where you were going, Jim, right? Right now, obviously, nothing’s stopping that from happening, and arguably, if I were enterprising and wanted to figure out how to make money as an author, I might do exactly that and generate the thing, send it to the tools that I developed to get the feedback to then either reprompt or hand edit. And then it becomes a hybrid effort.
The really scary thing, though, is what I’ve always worried about with the work that I did in predicting bestsellers, we’re identifying the themes that are the mean, the not the real outliers in in the data, and that is also what generative AI is doing under the hood, is homogenizing and avoiding the less likely sorts of phrasings and wordings. A former student of mine, Ryan Heuser, has written this fantastic research paper on cultural collapse,[6] where he does a very rigorous experiment to show that this is happening already. This is my biggest worry.
Sarah asked the question earlier, I think, about “the revolution” kind of question, and you talked about the printing press and so on. And there’s a part of me that wants to say, “Yeah, it’s no different. It’s just another Industrial Revolution kind of thing. We’ll get over it.” But it does feel different. There was that famous anecdote about the guy who invented the crossbow takes it to the king and the king says, “Oh, with this weapon, we’ll end all wars.” And then we get the the bomb and you think, “Oh, if they’d only known.” It feels like that kind of moment to me.
I don’t know exactly what’s going to happen here, but I do think, at least for me—and I’m in a good place because I’m going to retire sometime soon and not have to worry about this, I’m going to go fishing—but I think we better get used to it. This is not going away. I don’t think the legal establishment has the means to stop it. I don’t think the companies are incentivized to stop it. My utopian world is one where—and people in the author world are already doing this. They’re creating badging systems, “guaranteed organic prose” kind of thing. But then it’s going to become, sort of a farmer’s market-kind of world for authors. And nostalgia makes me sad about that, but at the same time we can’t stop it, so we better get used to it.
[Gibson] I think that’s exactly right, I am relatively convinced that no matter how good it gets there will be—I hesitate to say ‘market,’ but at least an audience for truly human-authored stuff. I advise one of our journals here at the law school, and I was working with the editor-in-chief on their publication agreement and she said, “We want to introduce something that asks whether you used AI in the writing of the piece,” and I said, “Fine, but what if they did a Google search?” Because then they’ve used AI. And it’s not obviously what she meant by it and we could play with the language a little bit, but it is just going to become so fully ingrained in our everyday lives that I think it will be difficult to avoid using it.
That said, I completely agree that there will still be some value to saying, “This is a ‘fully organic work,'” to use a label from a related field. The question I don’t have the answer to is, is that going to be a robust market? Or is that going to be a quaint little cultural artifact? The rest of us view as maybe cute, but a little other, like the Amish or something—respected, but very not mainstream. I don’t know how big that market is going to be, and copyright is essentially a market mechanism. It’s creating private rights, but then it’s leaving it to the marketplace to value that, and a lot of law does that.
I have a little bit more hope there than maybe I have in some of the other topics we’ve been talking about, because I do think that humans value authenticity in lots of different ways. The most valuable art is original art, and not original in the creative expression sense, but original in the sense that it’s the actual canvas that the artist painted on. I think the most expensive piece of artwork ever sold was Salvator Mundi, a Da Vinci painting, an oil painting on wood. And it’s not expensive because copyright set up some sort of exclusive rights on Da Vinci’s part, it’s expensive because it’s the actual wood on which Da Vinci’s own hand put the oils. That’s artwork, that’s technically copyrightable, although I think it actually predated the invention of copyright law, but it’s not copyright law that’s mediating the value there.
Or to make it a little bit more accessible, people will pay a premium for the official Steph Curry jersey, because they want the same one that Steph Curry wears, even though they might be able to buy a knockoff much cheaper. And I think that’s because they value the actual human connection, the authenticity that comes with that. And so my question is, is there going to be a robust enough valuation in the marketplace for that sort of authentic human connection that copyright will, or some sort of legal regulatory regime will, still have meaning.
And the last thing, Matt, because I think I’ve been pretty negative here, but I don’t think copyright law is going to save us, I think there’s a possibility that law in general might help, but it would have to be direct regulation of the development of AI, full stop. We’re going to have to introduce some very broad guardrails into the marketplace because AI, to the extent it poses a threat to various industries or an existential threat to the species, is a collective action problem, and you can only solve collective action problems through collective action, and our governmental system, when it’s working well, is a way of addressing those problems. But I don’t think it’s going to be private law, market-based regimes like copyright that are going to really pump the brakes in any meaningful way.
[Jockers] I’d like to follow up on two things that you said there. One is around provenance, and everything that you said there about the value we place in
the originality and the Da Vinci example and so on. Practically, we’re now in a space where I’ll just, for the sake of a pointed argument: it’s no longer practical to assure the provenance. The opportunity for fraud now is thousand times greater. And Da Vinci could have prompted his AI to give him whatever it is that he then transferred to that wood block—by hand, but the originality of the idea might not have been his. So, to implement some sort of “authenticity” seal of approval requires that somebody stands behind me every hour that I’m working and ensures that every word emanates from the tips of my fingers. So it sounds good, but in practice, I don’t think it’s tractable, is it?
[Gibson] I guess there’s two ways it might be tractable, but I don’t think I essentially disagree with you. One is maybe it’s something that could happen to the same extent that we believe a label that says this is not a GMO product. In other words, there’s always room for fraud in the background, but maybe there is some pressure we can exert, you know, through the regulatory state to have some degree of reliability.
But the other I think is exactly the example you said, which is, I think live shared human experiences are going to matter a lot more. Why do we pay a premium to go see our favorite band in concert? It’s not necessarily because they sound better in concert than they did in the studio version. It’s because of the actual shared authentic human experience there—and again, maybe that form of entertainment is also going to be a quaint little sideshow in our culture going forward, or maybe it actually becomes more important, at least until we can’t tell the humans from the robots, and that’s probably a different panel.
[Jockers] I am really pleased by that I, hadn’t thought of that, and it’s very attractive. Of course, Milli Vanilli immediately popped to mind when you said that. But that’s interesting, performance …
[Gibson] You’re dating yourself, Matt! You and I are both close to retirement if you’re pulling Milli Vanilli samples out.
[Jockers] The other thing I wanted to come back to was your comment about regulating the AI, and I’m curious to hear what you think, what direction that goes. What needs to be regulated?
[Gibson] I’m going outside my area of expertise here, so I don’t know that my insights are any more valuable than the next guy’s. One could simply try to introduce some sort of regulatory approval process that would at least introduce some friction and allow us to understand the dangers and deal with them at a pace that actually would allow for more legislative or regulatory interference.
Part of the collective action problem is the international collective action problem. Even if the U.S. does the right thing and tries to slow this down, then, who knows whether other countries are going to follow suit. One could try to disincentive agentic AI over non-agentic AI. In other words, AI that can actually make decisions and go out in the real world and cause things to happen is a different problem than AI that is essentially advisory.
But I’m not holding out hope. It’s more my conviction that copyright law is certainly not going to make a big difference. Copyright law could make could change who makes the money. For example, if we decide that, some of the current litigation that you’ve probably all heard of, ends up favoring copyright owners so that feeding copyrighted materials into an AI engine is actually considered something for which you need a license from the copyright owners, that would change things, but it would it would change who gets the money. For instance, then Getty Images, which has the copyright to billions of images, suddenly becomes the source for you know generative AI engines that want to produce images, because they can get them all through some licensing deal in a one-stop shop. But I don’t think it actually slows down the production, it just shifts who makes money. The best hope for copyright law is introducing a few frictions that move the money around without actually stopping the train. I think it has to be some other regulatory effort to do more, and I’m not an expert tech law or the administrative state, unfortunately.
[Jockers] Got it. Every second that we’re talking, I just can’t keep Pandora’s box out of my head. It’s out. Sarah, I see lots of things popping up in the chat, I don’t know if you want to…
[Hartman-Caverly] Absolutely. I can pop in a question from our Q&A queue, actually. So this goes back to the discussion about kind of operationalizing copyright in protecting a market rather than really protecting the artistic expression itself. So the questioner asks, “Art and copyright is really not about art, but about commerce,” which is what you’ve all been talking about. “Art used to be deeply connected to a religion and to community ritual and to connecting us to something that’s larger than us. So in what ways are we actually devaluing art by focusing on who owns it and sort of the copyright implications of generative AI?”
[Gibson] It’s a really good question. I think the assumption is correct that art used to have a very different sort of valence in the culture than it does now. I do think that train started leaving the station back in the early 1700s, when we essentially shifted in Anglo-American law from control of artistic and particularly printed materials being within the remit of the crown, essentially, that the British government would essentially decide who got charters for operating printing presses, and then in 1709, it essentially became a decentralized sort of market determined system. Ever since then, it’s been something that certainly allows for whatever kind of connections between author and audience mediated through art that the market wants to determine. There’s certainly room still for personal attachment to art, it’s just not the assumption on which the copyright law has been been built for some time. I don’t know, Matt, what do you think?
[Jockers] I’m not sure what I think, but what came to mind is the patron model, which I think is what we’re talking about under the hood here. And it just got me to thinking about, I wonder if that’s the new direction, back to something like that. If the marketplace is flood—that gets to the provenance and authenticity problem, if you take that approach. It also probably disappoints a lot of teenagers wanting to be novelists.
[Gibson] That was, the patronage system was the primary driver of what we would think of as great art. Michelangelo’s David was the Florentine Wool Guild paying him to do it, or his, Pietà, I think, was Pope Julius II [Cardinal Jean de Bilhères-Lagraulas] giving him some money, and maybe that does reemerge. We have some parallels for that, we—I don’t know if we still do but until recently—we had the National Endowment for the Arts, in the scientific area we have government grants for researchers, so there is some sort of direct payment to be an innovator, but the beauty of the dispersed market-based system is that you don’t have the taste of whoever has the money determining what gets produced, you have the taste of the public (and not that the taste of the public are unerring, but at least they’re a little bit more decentralized and diverse).
[Jockers] The other model we have is what’s happened with content creators on the Internet. Youtube has set up financial mechanisms based on views and so forth, but others get Patreon accounts. And of course, I’m thinking of the really horrific one that’s made the most money. We won’t talk about that right now. But that then begs the question of whether there’s a market for an artist, a writer, to “do a live performance”, as it were, to paid subscribers. I don’t know.
[Gibson] If it’s digital, we’re already having problems deciding what’s actually a real performance and what’s AI slop.
[Jockers] Oh, my God, right.
[Gibson] It would have to be live in person. I want to take a step back because copyright in its current form, and really in almost every form that’s existed since it really became a thing in the early 1700s, has never been particularly good for the actual creators. In other words, it’s not that AI is the first development technologically that disadvantages the creative class. The fact that it’s a market-based system means that it’s going to be the people with money who end up controlling the expression and determining what’s valuable and what’s not. I have a brother and sister-in-law who are both screenwriters in Hollywood. They’re obviously worried sick about what AI is doing, but—and I hope they don’t mind me saying this in public— they’ve been worried sick about their careers ever since that became their careers. It’s always been somewhat exploitative. It’s always trying to balance your own desire to be expressive and artistic and true to your own vision with the need to get paid for whatever it is the local producer values. I’m not saying AI isn’t a sea change in the way we approach that, but the change is not ‘artists used to have it good and now they don’t,’ it’s ‘they don’t have it good in a new way,’ I think, from when they used to not have it good.
[Hartman-Caverly] If you don’t mind, I want to jump in there with shifting the perspective from the artist to the audience. Some of the metaphors you’ve already brought up in the conversation are these things like, we might have some kind of authentic organic media badge that appears on content, and we might have these, characterized as quaint, a little outside the mainstream, farmers market-style places or modes of exchange for human-created or human-authored works, so one of—as a person coming at this from the library world—one of my areas of concern is what that might do for the public’s access to expression and information. Are we going to end up in a situation where we have this tiered premium access to human-authored content and kind of authentic or organic media that’s super expensive, pricing most of us out of the market, and the rest of us are going to be stuck in a world of AI slop, with all the downstream implications for cognition and our understanding of reality? Any thoughts that you might have about that?
[Jockers] Maybe.
[Gibson] If it’s about shifting from author to audience, I’m going to let the the literature professor take a first crack at that one.
[Jockers] That’s the other perspective, is that the market determines what is art and what isn’t, and the literary—my field for many years—was the arbiter of good art. That changed pretty radically probably in the ’80s, especially, it started before then, but we began to to look at things differently, from different perspectives. I remember a class as an undergraduate on folk art which is very close to the farmer’s market model we were just talking about, and finding ways to appreciate that that were independent or sometimes dependent on what the author’s intention was, but it was all about the audience response to it.
This gets to the banana thing that you [Gibson] brought up in the pre-write.[7] That one kind of blows me away. If we find something and put a frame around it and people say, “Oh, I want that on my wall,” is it art? If the answer is yes, then it doesn’t really matter if it was a generated piece of “art” or not.
[Gibson] If that’s the approach, then we might want to say “yay AI,” because it is probably going to increase the choices available to the public when it comes to expressive works, because it has eliminated a lot of the front-end costs. That is arguably consistent with the way that our legal system thinks about copyright law and the incentive for creative works. Copyright and patent law are actually constitutional law. There’s Article 1, Section 8[8] has a clause that gives Congress the power to create copyright and patent regimes, and it’s one of the very, very few provisions in the Constitution that explains why it’s doing what it’s doing. Saying you could coin money, it doesn’t then explain capitalism, it just says you can coin money, you figure it out. But with the patent and copyright clause, it says “to promote progress in science and the useful arts,” we are giving Congress this thing.
So the overall goal, and again, this is just copyright as currently conceived and constitutionalized, this isn’t necessarily normatively the right answer, but the purpose of copyright law as currently conceived is to essentially maximize the amount of expressive works available to the public. Providing an incentive through private exclusive ownership of your own work is not the goal, it’s just the means to the end. That’s the way that, in 1787, the framers thought, given their inheritance of Anglo-American law, was the best way to try to ensure a dynamic, growing culture with creative works.
I’m playing a little bit of devil advocate here, but one could say that if we find some new system by which we can maximize or increase the amount of expressive works available to the public, that’s better than the ‘give rights to human author’ system, fine! We’re just getting to progress better than we were getting to progress through the traditional means. It’s not, I don’t know if I agree with it, but it’s not crazy to say that AI is actually serving that goal, which again is why I think that if we’re going to regulate AI qua AI for purposes of mediating the relationship between humans and creative expression, it can’t be an instrumentalist goal, it can’t be a consequentialist equation. It’s got to be because we just inherently value something about humans as authors, because we value something about human dignity itself, rather than because it’s going to get us better to some sort of shared cultural goal than the alternatives. That I do actually agree with.
[Hartman-Caverly] So Jim, I know you mentioned that you’re collaborating on, I don’t know if you want to call it a legal argument or at least a paper in the realm of law, that to my reading, creatively applies some copyright conventions to some of these problems. I don’t know if you’re in a position where you want to preview that now or if you want to tell people that’s a coming attraction.
[Gibson] I can give you the 60 second version. As we’ve been talking about, IP rights both in patent and copyright were intended as incentives, a short-term cost for a long-term gain. We’re going to bind up this otherwise very shareable attribute, whether it’s how to invent something or a book or what have you, with rights that actually restrict its dissemination in order to incentivize the innovator to innovate in the first place, because those exclusive rights allow them to get licensing fees and so forth. And so usually, there are costs to IP, but the long-term benefits are greater, because we think we’ll have more innovation with IP than without it.
But if we’re dealing with a form of innovation that actually is very low cost, then suddenly the costs of the intellectual property system end up predominating over the benefits. In other words, we wouldn’t want to give rights to something, again, if we’re doing the sort of consequentialist, instrumentalist view of the law, that would be created without the rights, because then we’re binding them up with no long-term gain. And so there’s a number of fields that scholars have called “low IP” fields, where we think we’d actually have a lot of innovation even in the absence of IP protection. Fashion is one of them. Fashion actually doesn’t have particular IP protection, but we think that people would continue to buy the latest fashion, maybe even because of the rapid copying and turnover, because that makes the next trend emerge even more quickly. Apparently, pornography is something that gets produced without a whole lot of worries about whether it’s copyrightable. And our theory is that artificial intelligence engines are clearly low IP in the sense that the Anthropics and Open AIs of the world created these massive, incredibly expensive engines without any understanding of how and whether IP would actually protect either the engines themselves or the engine’s output.
And so we have, I think, a pretty good example of something where granting IP rights might actually impede innovation rather than further innovation. And if that’s the case, and we want to actually impede innovation in AI, if we want to slow it down, then one clever way of doing that is to load it up with the exclusivity that IP rights create. So you need the IP licenses on the front end to load it into the engine, and everything that the AI produces on the back end is actually fully protected by copyright, because that actually slows down the AI engine. And this is not a turnkey solution to the Skynet problem of AI or anything, but it might—again, referring to what it was the last speakers were talking about—introducing friction into the system can at least maybe give us a little time to get our heads around what it would look like to try to introduce some other more direct longstanding regulatory means. That’s the idea anyway. We’ll see if it flies when we finish writing the paper.
[Jockers] Could you unpack for me a little more the comment, “Everything that AI produces is protected?” How does that look?
[Gibson] That was a little too simplistic. If we apply the same sort of formalist requirements for AI production that we do to human production—so, it looks significantly different from things that were produced before, that it has the modicum of originality we require to say copyright should protect it, then obviously AI is, you know, gazillions of things a day are going to be bound up through exclusive copyright rights that wouldn’t be if we take the intentionalist view and say, “No, AI does not have intent to be artists and therefore they can’t get a copyright.” You know, the former example means that exclusive rights are actually impeding the dissemination of AI works in a way that the latter possibility isn’t, and maybe that would be a good thing if we’re trying to slow the whole engine down. Does that make sense?
[Jockers] I think so. Let me give you a scenario to make sure. So if in that model, you’ve got two people independently inserting the same prompt and producing two outputs that are highly similar to each other, the first one who got it done seconds before gets the protection and the second one is deemed too similar to the original?
[Gibson] Current copyright doctrine would say we deny protection to the second guy only if they copied the first guy because it’s copyright, right? Patent would actually say, “Yes, you can’t get a patent because you weren’t first,” but copyright is actually about whether you were a copy. It’s not about, you know, whether it’s by coincidence, similar to something else. So in fact, because the answer to your example is ‘no’, they would both get rights, and that would gum up the system even more! We’d actually have more exclusive rights, perhaps impeding dissemination and causing people to pay licensing fees and introducing friction into the system.
[Hartman-Caverly] So I’m going to jump in because we are crossing over into the break that I promised everyone, attendees and presenters alike, but I did hope to go out on maybe a hopeful or optimistic note. So as you grapple with what is really emergent for all of us, I think, thinking in this space, what’s making you hopeful or optimistic about the future? Or what do you hope to see develop in the future with respect to human authorship and generative AI?
[Jockers] Wow, such a tricky question, but I’ll take a stab, because I would say that even for what I’ll describe more as utilitarian authorship—I want to separate that from the creative piece—I have to admit, I’m already a user of these tools for the most mundane sorts of tasks that it has just made so much better. I also do a lot of coding in my day job, and the coding skills of these tools is just phenomenal. I can do in two hours what used to take me two weeks. So there is incredible utility in that. But at the same time, I’m cautiously optimistic, I work with many people who are super optimistic and I’m always the doubter in the room.
[Gibson] I would say that perhaps this technology is so disruptive and asks such foundational questions about the creative process that perhaps we’ll be a little bit more attentive to the distributive consequences of it than we usually are when a new technology comes along, that because it’s so quick in its disruption, because it’s going to affect so many different fields—and not just the creative industries, but I train attorneys for a living! It’s going to be very difficult to figure out what law practice looks like in a world of AI. The sheer scale of its disruption maybe will finally allow us to think about doing something within the economy for those who are going to suffer most from its introduction.
But usually it’s the technologist-takes-all when it comes to the financial benefits of something new. And we say, “Well, you know, the 55-year-old drill press operator in the car plant should just go back to school and learn to code.” And I’m not predicting this would happen, but I do think the sheer scale of disruption may actually cause us to ask sort of foundational questions about what we’re going to look like going forward in a way that less disruptive technologies to date haven’t.
[Hartman-Caverly] Well, thank you both so much. This has been extraordinarily thought-provoking. Great follow-up on our initial fireside chat. Thanks for your time and for sharing so generously of your expertise. Thanks to everybody for hanging around for a few extra moments. We are going to go into break and at 10 after the hour, we’ll come back for our spark talks. Thanks so much, everybody.
[Gibson] Thank you.
[Jockers] Thank you very much.
Recording: The human in the machine: What is ‘authorship’?
Embed not working? Access the recording via Penn State Mediaspace.
About James Gibson
James Gibson is Sesquicentennial Professor of Law and Director, Intellectual Property Institute at University of Richmond School of Law.
Professor James Gibson teaches and writes in the areas of intellectual property, contracts, professional responsibility, and law and technology. He is also the founder and former director of the School of Law’s Intellectual Property Institute.
Gibson’s scholarship has appeared in the Yale Law Journal, Virginia Law Review, Georgetown Law Journal, Texas Law Review, and UCLA Law Review, among other venues. Professor Gibson is also a frequent commentator in the media and has been quoted in the New York Times, Wall Street Journal, National Law Journal, Slate, and Chronicle of Higher Education. He has been a visiting professor at the University of Virginia School of Law and is a recipient of the University of Richmond’s Distinguished Educator Award.
About Matthew Jockers
Matthew Jockers is Distinguished Research Scientist and Senior Engineering Research Manager at Apple, where he leads applied AI and machine learning research focused on discovery and recommendations across apps, books, podcasts, and video/tv. Prior to joining Apple, he served most recently as Dean of the College of Arts and Sciences and Professor of Literature and Data Analytics at Washington State University.
A scholar at the intersection of the humanities, data science, and artificial intelligence, Jockers is widely known for his contributions to digital humanities, including foundational work on authorship attribution, large-scale literary analysis, and cultural analytics. He is the author of Macroanalysis: Digital Methods and Literary History (University of Illinois Press, 2013), Text Analysis with R for Students of Literature (Springer, 2014), and (with Jodie Archer) The Bestseller Code (Penguin, 2017). His research and writing have consistently explored how computational methods reshape our understanding of creativity, culture, and human authorship.
In addition to his academic career, Jockers has founded and led several companies operating at the intersections of publishing and AI. The most recent of these, Authors A.I., is a company built by human authors, for human authors. His recent writing critically examines the cultural and psychological appeal of generative AI for storytelling, raising questions about creativity, agency, and what it means to be an author in an age of machine-generated text.
- Robert Sapolsky, Determined: A science of life without free will (Penguin, 2024). ↵
- Christopher Bavitz and Kendra Albert, "The 'Monkey Selfie' case: Can Non-humans hold copyright?", Berkman Klein Center for Internet & Society at Harvard University, January 30, 2018. ↵
- Stacey M. Lantagne, "Goncharov (1973), Internet Folklore, and Corporate Copyright," Vanderbilt Journal of Entertainment and Technology Law 27, no. 1 (2025). ↵
- Frank Budgen, James Joyce and the Making of Ulysses (Indiana, 1960). ↵
- Authors AI ↵
- Ryan Heuser, "Cultural Collapse: Toward a Generative Formalism for AI Cultural Production," Anthology of Computers and the Humanities 3 (2025), 10.63744/USvuyzSIapvy. ↵
- Matthew Jockers, James Gibson, and Sarah Hartman-Caverly, "The human in the machine: Copyright, generative AI, and the 'ecology of creativity'," HUMAN AUTHORSHIP, Heterodox Libraries Symposium (Spring 2026). ↵
- Article I, Section 8: Enumerated Powers, Clause 8: Intellectual Property, Constitution Annotated, accessed May 22, 2026. ↵