Webinar Video: When Should We Recognize Something as a Property Right?
Updated: 21 minutes ago
Jeffrey E. Depp, Senior Counsel for Law and Policy at the Committee for Justice, joined a Federalist Society panel that returned to first principles to ask a deceptively simple question: when should the law recognize something as a property right?
Moderated by Satya Marar of the Mercatus Center, the discussion starts with the Intellectual Property Clause and the right to exclude and moves to today’s debates: rights of publicity in the age of AI, European-style rules that treat data as a quasi-property right, whether training generative AI on copyrighted works is fair use, and the patent eligibility of medical diagnostics and other subject matter.
Depp argued that using copyrighted works to train large language models without compensation should be treated as infringement, that the judge-made “transformative use” doctrine has swallowed the statutory fair use factors, and that Congress, not the courts, should set clear property rules. He made the same separation-of-powers critique of the judicial exceptions to patent eligibility and argued that because AI is a tool, the human using it should be treated as the author or inventor.
Speakers
Senior Research Fellow, Mercatus Center, George Mason University; Former General Counsel, Federal Trade Commission
Professor of Law and Journalism, University of Florida
Senior Counsel for Law and Policy, The Committee for Justice
Postgraduate Research Fellow, Mercatus Center, George Mason University
Video courtesy of the Federalist Society. As always, the Federalist Society takes no position on particular legal or public policy issues; all expressions of opinion are those of the speaker.
Related CFJ work: “A Patent, a Blood Test, and 20 Years of Waiting” and “Don’t Tax Innovation: The Patent ‘Wealth Tax’ Is a Terrible Idea”.
Transcript
Libby Dickinson: Good afternoon, everyone, and welcome to FedSoc Forums. My name is Libby Dickinson, and I’m an Associate Director with the Federalist Society’s Practice Groups. As a reminder, all opinions expressed are those of our speakers and not of the Federalist Society.
Today’s program will be a discussion of “When Should We Recognize Something as a Property Right?” Today’s moderator is Satya Marar, postgraduate research fellow with the Mercatus Center at George Mason University. If our audience has any questions throughout the webinar that pertain to the topic at hand, please submit them through Zoom’s Q&A feature located at the bottom of your screen and not the chat, so that we’ll be able to access them later.
With that note, I will turn things over to our speakers to start the conversation. Satya, thank you so much for joining us today and for moderating today’s discussion.
Satya Marar: Hey, thanks for that. As we approach the 250th anniversary of the American Republic, we recall that America’s constitutional experiment began with one of many striking propositions: that the new republic would protect not only land, ships, tools, and goods, but also products of the mind. Article I, Section 8, Clause 8 of the Constitution gives Congress the power to promote the progress of science and the useful arts by securing for limited times exclusive rights for authors and inventors in their writings and discoveries.
This is a remarkable clause, not merely because it authorized copyright and patents, but because it embeds IP not simply as a policy proposition, but within a constitutional theory of ordered liberty, republican self-government, and economic dynamism. It treats creativity and invention as public goods worth encouraging, while also limiting the duration and purpose of exclusive rights. The idea isn’t that ideas themselves should be owned in the way you own a farm or a house. It’s a subtler premise: that a society that wants learning, invention, expressive culture, industrial progress, and technological diffusion must sometimes grant legally enforceable rights over intangible assets.
Those rights channel private initiative towards public ends, giving authors, inventors, firms, and investors enough control to risk time and capital. They allow trade, licensing, specialization, and markets to flourish. They create incentives to disclose rather than to conceal useful ideas. But they also carry dangers: monopoly, rent-seeking, suppression of follow-on innovation, and over-claiming of what belongs within the common stock of knowledge.
The modern debates over AI, data markets, and so on have returned us to a deeper question in IP: When should the law recognize something as a property right? The answer can’t simply be whenever something has value. Facts, ideas, language, scientific truths, human style: all have value, but they can’t all be enclosed. Nor can the answer be only when exclusion costs nothing, because every serious property regime imposes some social costs.
The four main questions these situations seem to come down to are these. Is the right to exclude necessary or useful to induce valuable creation, investment, coordination, disclosure, or the protection of personal interests? Are the boundaries of the property right we’re going to create administrable by courts, tribunals, and regulators? Will a new right increase or reduce transaction costs in net terms, or shift the transaction cost burden unduly in favor of one side or the other? And if we give the rights holder this degree of control, what would the implications be for free speech, competition, science, democratic culture, and any number of other policy considerations?
To help us answer and clarify current debates around expanding or scaling back the scope of property rights, we have today three great experts. First, we have Alden Abbott, senior research fellow at the Mercatus Center at George Mason University and a former general counsel at the FTC. Professor Jane Bambauer of the University of Florida, where she teaches law and journalism. And Jeffrey Depp, senior counsel for law and policy at the Committee for Justice. I’m your moderator, postgraduate research fellow at Mercatus, Satya Marar.
So, with that, let’s kick things off. Jane, what makes property rights different from other kinds of rights, such as under tort, contract, or statute?
Jane Bambauer: Yeah, I’ve given this a lot of thought. Even though I cannot claim to be a property theorist or really even a property law scholar, and I’m going to rely on Jeffrey and Alden to help elevate the discussion on that side, I have found by writing about privacy law and about free speech that the conception of information as property has for decades now been creeping in and causing some tension and some intellectual fuzziness in these areas.
So I think what’s key to property, and this is debatable among property scholars, but I think it’s clear that the main key right involved with property is the right to exclude. It doesn’t have to be an absolute right to exclude. But if someone has the power in most circumstances to decide who gets to use a resource and who doesn’t, and if the courts will give them compensation when their wishes are violated irrespective of having to prove fault or societal harm, then you know that you have a property right rather than some other type of right.
Contrast that with things like contract law, where a violation requires an agreement in advance between people, so it’s more relational. And contrast that with tort law, as you said, or risk management law. It doesn’t necessarily have to be common law tort, but whether it’s statutes or torts, if the idea is that we’re setting in advance some expectations about types of risks we don’t want to be imposed on society, then we would ask: Did someone else’s use of this resource cause harm? Was it blameworthy for the other person to have used it?
So I think this applies very easily to things that have a thinginess to them. If I have a car, it’s my car, and other people can’t use my car, because then I wouldn’t be able to use it and I wouldn’t be able to get the best value out of it. But obviously with non-rivalrous intangible goods like information or ideas or content of any sort, it gets a little trickier.
There are still some good reasons to assign property rights, but there are also some good reasons, as you said in your introductory remarks, to be cautious. By assigning property rights, you’re setting this default that a single person gets to decide what they think the best use of that item is. And it may be that that’s wrong. It may be that other people actually have a much better idea of what to do with that resource, and maybe their interests either conflict with or are orthogonal to the original rights holder’s.
Satya Marar: Cool. Alden and Jeff, do you have anything to add to that definition, those distinctions?
Alden Abbott: Yeah. Well, I think Jane summarized it pretty well. The key thing about property, and it sometimes gets tricky, is that there’s a moral dimension, which we may talk about a little bit later, as well as a purely utilitarian dimension. It’s not just what rules allow the maximum output of certain goods or services or expressions for society. It’s also: Are we able to protect the rights of individuals to protect what they have created, because they don’t want others to use it?
And the right to exclude is the same principle. A trademark owner may have a specific mark that identifies a good or service and prevents confusion, and thus may not want anyone else to use it. There may be an economic reason for that: if others use it but degrade the quality, it can cause confusion, and the mark may lose its value. But at the same time, there’s also a moral dimension: I came up with the idea and the thought to use this, and I want to control it completely.
That’s particularly the case in the area of so-called rights of personality, which have existed for about a century or half a century, mainly in state common law and some state statutory law, but they’re becoming very important right now. Think of the college athletes playing football and baseball. They are now being paid large amounts of money, usually by supporters of a particular school, to attend the school, and supposedly that’s to protect their name, image, and likeness. The idea is that if an individual has invested talents in developing a particular persona, whether it’s Michael Jackson or a particular athlete, they have some right to control it.
But there are always boundary questions. How far does this right extend? Just as in trademark, going back to trademark: how far does that extend? It may be right that you have a right to protect what you develop, but what if you don’t police it? For instance, the terms Kleenex and Thermos used to be trademarks that referred to specific inventions by particular parties, but they didn’t check. They didn’t try to prevent other people from using the terms generically, and then they lost control over them.
The idea is that these are rights you need to maintain control over, and if you don’t, the claim that you are preventing confusion, or that you have a right to prevent the benefits of a generic use, falls away. So it’s tricky, and these questions really differ based upon the type of right you’re dealing with. For example, copyright doesn’t protect ideas. There are many very similar ideas in a particular novel, whether it’s Gone with the Wind or Pride and Prejudice, but copyright protects particular manifestations, particular expressions, very precise expressions of those ideas.
So I think I’ll stop there. There are lots of tricky boundary questions in law, as I think you can already see, in deciding whether a property right is being properly applied in a case or not.
Satya Marar: Yeah, I think he’s also saying something really important there, right? We often talk about these things in very utilitarian terms: we need to have these rights so we can encourage investment and economic growth and all that good stuff. But there’s also a moral dimension, whether it’s fairness, sometimes an amorphous concept, or giving people control over their own reputations and the value that they bring to whatever gets generated. So yeah, fascinating stuff.
Jane, some of your research has focused on calls for the right of publicity: artists who want more control over their image or reputation, and those of us who want more control over the data we generate when we go onto social media platforms and so on, given how large and profitable these companies are. This bleeds into talk about data privacy rules, which sometimes seem to treat our data as a quasi-property right. So could you comment a little bit on legal movements in this area, whether here or in Europe? And if not through these sorts of quasi-property rights that give people control, how else might we address these concerns?
Jane Bambauer: Yeah, picking up on something Alden said, it’s true that copyright law does not cover ideas. There needs to be a particularized expression in order to be copyrightable. But these other areas of law are not quite so strict. So take the fact that I visited this particular website and clicked on this particular link. I’m thinking about web browsing data, right? You could treat that as a fact or an idea or something that wouldn’t be copyrightable.
And yet nevertheless, the instinct, whether we call it moral or something else, is for people to say, “Hey, yeah, it’s information, but it’s my information.” A lot of privacy law didn’t quite start out this way. But for the last few decades at least, especially with Europe, I’d say, at the forefront, there’s been an eagerness to try to recognize that very instinct through law.
You could treat the right of personality the same way: this is what my face looks like; I’m a celebrity, or I’m a college football player, and this is the sort of movement I make. It’s not like these games necessarily replicate exactly what the players did in a particular game, but the idea is taking some information that the privacy victim feels should have been under their control.
And so Europe, at least, through the GDPR and other ancillary laws, has tried to maximize that control model. There are other justifications for a business to use personal data, but those other justifications are interpreted extremely narrowly, which means that getting consent is, most of the time, the way that a company will be able to do something unusual or new with the data.
I’ve come out sort of critical of this, in part because every time someone makes a claim to a property interest in some piece of information, whether it’s sounding in moral or economic theory, it is carving out some liberty interest that someone else might claim, maybe even with the attribute of the moral high ground. So if I cannot observe or record or share or subsequently use some information about how people browse the internet in order to produce new knowledge about the world, then that is in some sense taking away an opportunity from the users of the data, who might claim to have a liberty interest.
Alden Abbott: Yeah, one good point Jane made is about data. For example, what if people voluntarily provide, say, their birthday, their address, or some other personal data? Lots of it actually may be discoverable online by third parties, but they’re just providing it to have access to a certain database. And let’s say the users collect that and data from many other individuals who are browsing, anonymize it, and use that to direct programs to people of a certain age or background, which may actually benefit them more. They can also direct ads in a certain way.
So, in other words, value has been created. Yes, it’s information that, quote unquote, belongs to a particular person, but it’s a third party that created the value. So then you get into the question of how you weigh this value creation against any privacy rights. That’s particularly true in the online world, where many examples of supposedly historically non-public information may be easily obtainable. If you create too much of a blockage, then valuable online applications, which could benefit a lot of people, consumers and businesses, can’t be developed because of threats of legal liability.
Jane Bambauer: Yeah, I think one problem is that the person described in the data does not have a good sense of the likely consequences, both positive and negative. And so this is where I think there is a deviation from at least the economic, utilitarian theory of property, where normally we think the person with the resource has a lot of inside information. This is like the Hayek model: they know the value, and they know the value of giving it away as well.
That’s not as true here. Or at least I think the transaction cost, the information cost, of figuring out what your life is going to be like if you do or do not let companies track you is high. It’s very hard for consumers, or really anyone, to know with great certainty. But people do tend to undervalue what they get back, as Alden said, from some of these tracking activities.
Satya Marar: Yeah, those are really interesting points. You touched on something important there, which is the implicit assumption that the property rights holder will know how best to use the property. I can think of a number of examples where maybe that isn’t literally true, where people who own property, or a share in it, hold it through an intermediary and pay for those services, for it to be managed by asset management firms and so on.
To your point, though, with regard to data, I think there is an argument to be made that the economic utility of data is in most cases not in the individual piece of data itself. Just as a drop of oil is one small share in a big barrel of oil, the value seems to come more from aggregating it and turning it into something that has utility, which in most cases is what the platform would normally be doing. And of course, there isn’t that same economic incentive for someone to pay for the best use of that data. That’s how it seems, anyway.
But thanks a lot for that. One of the interesting things you mentioned is that property rights often come down to whether there’s a thinginess to the thing, whether it’s tangible or intangible. And some intangible things have a thinginess that others don’t. I’m reading now some articles arguing that we should change our copyright law and recognize copyright in artistic style, because AI tools have now allowed us to create an algorithm that can arguably quantify the style of a Drake or a Taylor Swift. I find that interesting. I’m not personally convinced, but there are some interesting movements in the space.
With that in mind, I’d like to move on to a controversial topic that’s being widely discussed nowadays: the creation of generative AI tools and training them, often using the sum total of human knowledge, including copyrighted works. So, for Jeff: quote unquote, information yearns to be free, right? AI tools are often just, some would argue, more efficient at doing what humans do: learning patterns, assimilating info, and producing creative outputs. Now, we hold humans liable for violating copyright if they release art, music, or books that replicate existing works without permission, but not for studying and learning from those works. So why should we treat AI tools differently by extending copyright to the training?
Jeffrey Depp: Well, I struggle with a lot of the current discussion on this topic, because it always seems limited to a descriptive or positive perspective, and to me, the better discussion is from more of a normative perspective. In other words, I don’t necessarily disagree that under the current fair use regime you can certainly make a strong argument that this is fair use. But my question is from a different perspective: Should it be? Should we be utilizing copyrighted works? Should we be utilizing the works of authors without compensation?
And it implicates a number of concerns I have, particularly in the IP space, about separation of powers issues and whether we should be leaving so much of this to the courts. The Constitution, Article I, Section 8, Clause 8, gives plenary power to Congress to set these rules. And yet, to the point you brought up on copyright, transformative use, training data: it’s all on this transformative use, which is entirely a court-created construct. It goes back to the early ’90s, to a judge who wrote a law review article, and it was picked up by Justice Souter in the Campbell case, and all of a sudden, boom, there it is.
The word “transformative” does not appear in the statute anywhere. It’s entirely a judge-made construct, and what’s happened is what always happens when judges and the courts get involved where they shouldn’t be: it just expands, right? And now it’s consumed and subsumed the entire debate.
If we just go back to first principles, founding principles, which is my perspective, then my answer to your initial question, the title of the webinar here, is: early and often. Property rights should be found early and often. I’m a student of the Austrian school, and I think it is essential, for the market to create value, to create those property rights expansively and early and simply, and to allow the market process to do what it does and allow the parties to adapt. And I just don’t think we’re giving ourselves enough opportunity.
But getting back to the transformative copyright question, in my view the proper analysis is going back to the statute, where the statute talks about the various factors: the nature and the purpose of the use. Well, this is as commercial as you can possibly get. But we never seem to talk about that in the discussion of whether this is fair use. We always jump to the judge-made doctrine of whether it’s transformative or not. My view is that that’s not the way to go, and we’re depriving these authors, these innovators, these rights holders, of proper compensation for their work.
And frankly, I think in the long run we’re harming ourselves. If you find a way to compensate these people, based on the way the system is set up, you’re going to get more of what they do, and that’s going to benefit all of us, and that’s going to benefit the AI industry. And this applies in the patent space as well. You’re going to get more innovation, which is what we all say we want. If they’re properly compensated, they’re going to do more work, they’re going to generate more, and that’s going to benefit where we need to go.
And I have no desire to hold back the AI industry. I believe wholeheartedly that we need to win this race. We need to be the leader, and we need to innovate there. But there ought to be a way to respect these folks’ property rights in a way that incentivizes them to do more, which is going to feed a better outcome for the users of this data, the large language model creators.
And so I think if we set some simple categorical rules here, the parties will adapt. We have seen many license agreements take place for the use of this data, so it can be done. And as I always say, if we found a way to compensate a recording artist when their songs are streamed on the internet, we can figure out a way to provide some reasonable compensation to authors. They’re not going to get rich off of this, but it will show that we’re respecting their property rights, and it will certainly create no disincentive to do more, which I think is what we really want out of the system here. So that’s how I think about the use of copyrighted materials in the large language model training setting.
Jane Bambauer: Right. I—oh, sorry. Go ahead, Alden.
Alden Abbott: You know, in 2021, in a six-to-two decision, the Supreme Court said that Google’s copying of 11,500 lines of Java API code developed by Oracle was fair use because it was transformative: it allowed Google programmers to use familiar skills to create new smartphone apps. Well, Google could easily have entered into a licensing agreement with Oracle. And there were a number of commentators who said there’s a huge amount of copying here; this is nothing less than theft. It was called a transformative use, but this wasn’t just a tiny bit; it was a huge amount. And it created huge value for Google, but Oracle wasn’t allowed to share in that value.
Jane Bambauer: You know, I might come around eventually to agree with you both, and especially you, Jeff, but I’m still struggling a little bit with this conceptually. If you’re talking purely about cases that could be brought against an AI company based on output that is a copy, whether transformative or not, then everything you said makes sense to me. But a lot of these claims are not about the output; they’re about the input of training the model.
And in that sense, if we’re going to make our analogy to humans, it’s like saying that because I’ve read Shakespeare, if I later, in some way even unknown to me, processed that and wound up writing a totally different story about star-crossed lovers, but didn’t use that phrase, that itself is some sort of violation of the Copyright Act. And I don’t see that.
What I’m struggling with, though, is that you’re right: if there’s no recognition of the value of input training data for the rights holders, and I’m thinking about the New York Times case in particular, it means that all of this expensive work that they do with investigative journalism can just be dumped into an LLM, the knowledge is extracted in this mysterious way, and the output will never be a copy. So I see the incentive problem. But just as a straight-up application of the copyright statute, I don’t see how you get from the statute we have today to liability based on input data, input only.
Jeffrey Depp: I guess my thought is, they’re clearly making a copy of the work, which is one of the rights protected by the statute. And it’s certainly the expression that matters, because they create these vectors off of the actual specific expression that the author used. That’s how the vectors are created: word order. So to me, it’s essentially what copyright is designed to protect.
And again, I struggle with this whole thing. The law, as we know, always lags, right? Fair use has been around for almost 200 years. It was codified in ’76, and then the courts got involved and expanded it. But I just don’t think that applying a 50-year-old statute and a 200-year-old doctrine as they are works. It needs to be updated to adapt to the current use and to respect property rights, and I just don’t think it is. That’s why I really think this is more of a normative discussion: What should it be?
We should not be taking people’s work without compensation. That’s just against the American experiment, against our founding principles going back to the Declaration, which is how I look at this stuff. Those principles require that these rights be protected, be respected, and be compensated.
The other issue I want to point out is where I recoil over this transformative use analysis that we focus on so much today. Isn’t it implicit in that analysis that there’s always a specific second work, and we’re asking whether there was transformation? Well, what is that in the LLM training context? It just doesn’t fit very well. So I just don’t think that hanging this debate on a positive analysis of the current transformative use fair use regime makes a lot of sense, and it’s leading us down the wrong road, in my view.
Alden Abbott: Yeah, just to interject briefly. I think one area where fair use clearly makes the most sense is parodies, transformations to make fun of or critique other works. Those clearly have very strong First Amendment interests at stake, because if you don’t allow people to parody or to make political or other critiques of a work out there, you are really limiting their First Amendment rights very severely. And one could argue that copying thousands of lines of code doesn’t rise to the same First Amendment position. I’m not going to say anything more about the copying used in training, but clearly fair use has some strong First Amendment implications, and you should be careful about how those are bounded.
Satya Marar: Yeah, I think those are some great arguments going both ways. Jane, I’m sympathetic to many of the points you made, and I think similar arguments have been made to defend the rights of these companies to train their models on copyrighted works. But what I find is that a lot of these arguments rest on a proposition about these tools, because that’s what they are: machine tools. Do we perhaps tend to anthropomorphize them a little too much? The AI is hallucinating, the AI is learning. Humans learn too, but in a very different way, right?
In the recent cases that are winding their way through the courts on the subject, I thought it was interesting that a lot of the focus seems to be on the replication of the work. Are you creating perhaps a database that can then be used for future training, and have you given proper attribution for that purpose? I forget the name of the case, but that was one of the points made by the judge recently against Anthropic.
But I want to move on to another notion that frequently pops up in these cases. There’s this huge debate around what is or isn’t transformative. In a recent case where the judge ruled against the rights holders suing the AI company, the judge flagged this idea that market dilution is going to be a very significant consideration going forward. To what degree is allowing this use of copyrighted material going to dilute the market for those works? This is analogous to the idea of monopsony power: Are we suppressing the value of future output and innovation, which is one of the ideas behind having IP law?
So the question, for anyone who wants to jump in on it, is: Are we seeing any evidence one way or the other that these AI tools are diluting the market for original ideas through the production of what people derisively call slop?
Jane Bambauer: I mean, I don’t know. You’re asking an empirical question, and I haven’t kept up with it. It’s an interesting question. It’s the right question, I guess, is how I want to respond, because it takes seriously the idea that one of the core missions, maybe not the only one, but one of the main missions of copyright and other IP laws, is to make sure that the incentives for future generation are still there.
And I think one reason I don’t even want to speculate about what might be happening empirically is that I could see it going either way. I could see AI bringing a wealth of new creators who now have this tool that allows them to meet the maximum potential of their creativity. Or I could imagine the opposite: what’s the point?
Jeffrey Depp: Yeah, I think that’s the key way to think about this. Again, I recoil at the constant debate about whether AI is an author or AI is an inventor. To me, that’s the wrong question. The answer’s no. It’s never going to be an author. It’s never going to be an inventor. It always goes back to the closest human, the human that’s utilizing the tool, as you mentioned. They’re always going to be the inventor and the author.
And we should encourage that, because that will embrace the utilization of these tools in a way that’s low-risk. It’s not going to eliminate their livelihood, because that’s the concern, right? That AI is going to go and generate all these new novels, so that we won’t ever have to read our favorite author anymore, because we’ll have all this stuff available to us so cheaply and ubiquitously.
We only get there if we think about it in the wrong way, in my view. If we bring it back to “there’s always a human; they’re the key; they’re utilizing the tool,” we’ll get more and better works, we’ll get more and better inventions, and at the same time we’re not going to disincentivize or eliminate the livelihoods of the innovators that we rely on.
America was built on innovators. We must be respectful; we must provide an environment where they can grow and prosper and be fruitful and multiply. That’s what we want. It’s essential for national security, for economic security, for everything that we do. So we have to think about this stuff the right way, and thinking that an AI can by itself be an author or an inventor is, in my view, absurd and problematic.
Alden Abbott: Yeah. Those are excellent points everyone’s making. I think intelligent judges can parse through the copyright law. One can argue about amendments, but they can come to some intelligent decisions. I’m thinking of Bartz v. Anthropic, where a judge in the Northern District of California held that using books to train Claude was exceedingly transformative and a fair use. Why? Anthropic had scanned lawfully purchased books into a digital internal library, which was not multiplied or redistributed, but it did inform the development of Claude.
However, the court separated that from the acquisition and retention of books. For instance, its creation of a permanent library from pirated copies was not fair use. The court stated there is no AI company carve-out from the Copyright Act and set the pirated copy issue for trial on damages. So you could read that as saying training may be fair use when the model does not create infringing output copies, but stealing or retaining pirated source copies is a separate problem and is infringement.
So I’m not saying he necessarily got it right on the first one, but the courts can work through this. And of course, eventually some of these issues are going to get to the Supreme Court, perhaps. And from what Jeff has said, I know he has a concern, and I also might have a little concern: the problem is the Supreme Court has not been very great on property rights in general. We haven’t talked about patent, but I think in patent the Supreme Court has shockingly ignored the plain statutory language of Section 101 of the Patent Act to find that certain highly novel inventions were not patentable. It just made absolutely no sense, but there we are. So sometimes Congress does have to step in and clear things up.
Satya Marar: Thanks. That goes to my point, and I do want to move on to patents in a second, but before I do that, I want to quickly get responses on this question. The national AI framework was recently put out by the Trump administration. Obviously, their goal is to encourage American innovation and give American companies an edge in the AI race. One of the things they’ve quite diplomatically tried to do is sidestep these debates about the fair use question and say, “We’ll leave that answer up to the courts.” They’re open right now to amending the copyright law.
So my question for the panel would be: What amendments, if any, should we make to the copyright law in order to address these sorts of issues that are being brought up? Because, Jeff, to your point, while the economic question can be a little bit murky, based on evidence that might not exist yet and hard to argue in each separate case, I think the normative or moral case is probably going to be what dictates any future reform in this area. So what amendments would you make?
Jeffrey Depp: Again, I think the bigger question is this. I believe the democratic process should play out here, but Congress must set simple categorical rules. Whatever they are, that is most important. My personal view is that in the particular case we’ve been discussing here, it should be defined as infringement. It should be infringement to utilize copyrighted works to train large language models without compensation, in my view. But I don’t get to make that decision.
I think the more important thing, from a political economy perspective, from a property rights perspective, is that we just need simple categorical rules. Leaving it murky, leaving it complicated, leaving it up to the courts, as we’ve discussed here in copyright and as Alden talked about in the patent space, is a disaster. All it does is increase litigation and create confusion.
And again, coming home to my belief in the process of innovation, and the discovery process and the subjectivity that are at the core of Austrian economics: if we do that, the parties will adapt. We need to believe and understand that that will happen, if we just have the institutional courage to follow the Constitution, follow separation of powers, set the rules, and then get out of the way and let the market drive the adoption and the adaptation. And the country will be better off for it, in my view.
Alden Abbott: No, I really can’t disagree with Jeff. I think that’s important, particularly in the patent area. And maybe some clear language on what is fair use, saying that copying certain things is or isn’t, with very clear categories, perhaps could be helpful.
I think the problem in patents is even more important, because there’s a huge amount of innovation driven by patents. Medical diagnostics, for example, are patentable in China and Germany, but because the Court has had some weird interpretations about what abstract ideas are, applying them to that and a few other important areas, we’ve lost out. Some foreign investors have gone to China to do work on medical diagnostics because they thought it was too complicated and risky to try to get defensible patents in the US. That’s a terrible situation to be in.
And unfortunately, the public choice people would say you’ve got a rent-seeking coalition. You’ve got the people who are using this property without having to go through licensing and so on, and they’re going to lobby against that. But I think it is very, very important, in terms of international innovation and the role of the US, that patent law gets straightened out by Congress.
Satya Marar: So, with regard to that, one of the debates we’re having right now is around subject matter eligibility. It’s rare to find a court that has said, as a blanket matter, you cannot patent this technology or that technology, this innovation or that innovation. Rather, they do it on a case-by-case basis that in effect makes it very difficult to patent certain things. For example, it is possible but difficult to patent a business method. It is really very difficult, from what I’ve read, to patent a medical diagnostic tool. So the question is: If we do expand subject matter eligibility, perhaps through statute, would there be any negative implications in terms of holding up innovation?
Jeffrey Depp: In my view, no. It would be injecting gasoline into the innovation engine. Again, the subject matter eligibility issue is the poster child for courts meddling where they have no business meddling, and the problems it has created are a direct result of a violation, in my view, of the separation of powers. Congress has plenary power to set the rules for patenting. They did it with a simple categorical rule in the text of Section 101. And here come the courts, saying, “Well, wait a minute, those four statutory categories—we have to have some exceptions to that.” And look at the mess that has created.
So the statutory change that would be right, in my view, would be to eliminate, abrogate all those judicial exceptions and just go back to the statute as written. Everybody can understand that.
In the meantime, at the Federalist Society conference in November, I suggested to the director of the Patent Office, “Given this clear violation of separation of powers, where the judiciary is writing patent law in the subject matter eligibility context, have you considered treating their opinions as advisory?” As in: “Okay, thank you, we’ll look at that. But at the end of the day, we’re an Article II agency. We can interpret that statute, and we appreciate your opinion, but we’ll make the call as to whether to cancel the patent or not.” It just illustrates the problems that are created when courts go outside their lane and start to violate separation of powers, in my view.
Jane Bambauer: Maybe I’ll weigh in to try to find a little bit of the controversy between us. I’m highly sympathetic to some of the examples that have come up, like medical diagnosis devices, for example. And I mentioned news gathering earlier. There are certain types of even raw information that today isn’t even copyrightable where I think we have a serious incentive problem.
But what those things have in common is that they surface information that’s hard to extract otherwise. It takes effort or ingenuity, and it’s highly useful. And so the problem of free riding is right there, and I think we’re all interested in solving that.
My concern is on the other side: when I think about who would be lobbying Congress for these changes. If I think about what the RIAA does, it’s like, do we really need more incentive for people to become celebrities and recording artists? I actually don’t think we do. And perhaps Jeff and I can compromise: maybe if the terms of copyright were changed or modified to fit, in categorical terms, how much other psychic reward or other incentive there is for people to produce, then I think it might make sense to expand rights. But as a matter of political economy, I do worry a bit about who exactly is going to be there, ready to get their favored terms.
Satya Marar: Yeah, that’s interesting, right? Leaving aside the moral question of compensating people for their effort and their creativity, it seems to me that if an AI tool can ingest the sum total of your works and produce something generic that replicates them in a way that’s as good as what you yourself are continuing to put out, that’s probably a sign that you as an artist need to do something different, something more. Or perhaps your art or your style happens to be a particularly generic one.
And with that being said, for all the talk about AI slop, I do think there’s a lot of human slop out there. Mumble rap is an example, but I’m biased on this topic. So I do think there is an argument, at least on the economics side, leaving aside the question of compensation for the use of IP, that these tools will in the long run perhaps encourage more creativity. You now know what the generic version of something sounds like, and that makes it easy to get to: Okay, what’s innovative? What’s new? What can I add to all that now exists?
All right, so we’re going to move on to audience questions. There are some interesting questions here. This is a question from Dr. Jack David Gallup: “The opinion in Acuff-Rose moved the Court away from a purely commercial focus towards determining whether a new work supersedes the original and creates something of value for the public. Is this not analogous to patent law embodying the novelty doctrine?” Whoever wants to jump in, if anyone.
Alden Abbott: Well, I guess I’m just not exactly seeing what the specific analogy to patent law is here. Maybe I missed that.
Jane Bambauer: So, creating a new market, basically, because it’s doing something that goes beyond, maybe embedding, but going beyond? I don’t know. I don’t know enough about patent law, so I’m just trying to help translate.
Jeffrey Depp: Was the question about creating a new work that is sufficiently different from the previous work? Is it like the analogy to obviousness? Is that what I’m understanding?
Satya Marar: Yeah.
Jeffrey Depp: I mean, again, you have the substantial similarity doctrine in copyright, right, before you can infringe. I guess that’s something to think about in that context. But I don’t know. In copyright, it’s about the expression. If you express the idea in a different way, that’s a different work, and it’s entitled to copyright.
Jane Bambauer: The concept of derivative works might be too broad if we compare how patent law handles it, and maybe that’s the essence of the question. If we compare how patent law deals with new add-ons, does that imply that maybe the scope of derivative works is too broad?
Alden Abbott: Well, yeah. Clearly, as a copyright holder, you have a right to come up with derivative works. In patents, it’s somewhat different. The big thing about patent law is that the public can use the information disclosed. That is one of the arguments as to why patents may be socially superior to trade secrets: they put that information out there, and people are allowed to use it to come up with their own patents. But those have to be novel. They can’t be substantially based on prior art; they have to go beyond it. So you have to show your own form of novelty.
You may get your own patent. It may not be a derivative work of the original patent holder’s, but you have a follow-on patent that has to meet the conditions of patentability. Say you change the way a drug may be chemically injected. If it’s obvious, you can’t do that, but you may be able to come up with some changes that allow for new patents. So yes, there is sort of an analogy, but it doesn’t really work the same way as derivative works, because the two statutory schemes are fairly different.
Satya Marar: Cool. We have a question on copyright terms: “Any opinion on whether copyright terms are too long? Isn’t a ‘limited time’ properly determined as a portion of a human life? Otherwise, there is no quid pro quo: a person in their 20s today will never see fair use of the copyrighted materials they spend money on. In theory, someone in their 20s today should see fair use of today’s data by the time they’re 50.”
I want to expand the question a little bit. Does anyone have thoughts on the current duration of copyright terms, as well as patent terms?
Jeffrey Depp: Well, again, they’re currently simple and categorical. We know what they are. They’ve been a little bit muddled in the patent space with the extensions and adjustments and obviousness-type double patenting, which again is mostly judge-made meddling that has caused those problems. But again, I get back to this: as long as they’re simple and categorical, I can live with whatever they are. The market will adapt based on whatever we decide, as a body politic, they should be.
But that’s the key. Don’t complicate them. Don’t have a bunch of exceptions. Don’t have different terms for different kinds of works. Or if you do, they have to be very clear and very straightforward.
Alden Abbott: Yeah, the Supreme Court did address the issue: Can a term be too long? There was the infamous Mickey Mouse case, where I think Disney had lobbied to get an extension of like 70 or 75 years beyond death, because they wanted to preserve copyrights for some early Mickey Mouse cartoons.
Jane Bambauer: They went around the market with that one. They went around the market straight to Congress.
Alden Abbott: It was a matter of lobbying, but the Court did not say, “Oh, we can tell that this is no longer limited.” Now, if you had a thousand-year term or a 10,000-year term, that might be different. Correct me if I’m wrong, Jane: was it 75? Or was that the magic amount at which the Court refused to suggest that it was not limited?
Satya Marar: Right. So there’s a question about how much human contribution to a creative work should be required for someone using an AI tool to be named as the author of something original. “And on the patent side, will the test be ‘substantial contribution’ under the Pannu case, or conception of all claim limitations under the [unclear] case? It appears that in many cases no human will satisfy the criteria, and thus the property right will be invalidated as not being the product of a human.” Thoughts?
Jeffrey Depp: I remarked on this during the panel. I think this is the wrong way to think about it. We need to state unequivocally that AI tools cannot on their own be authors and inventors. They are always tools. There’s always a human, and you just have to find that human and make him or her the author or the inventor. And then this debate goes away, and we energize innovation, because we are allowing the tools we have available to us to be used in their most effective and maximized way.
It’s just like when we invented different laboratory equipment or other things to help the discovery of new pharmaceuticals, or when we invented the personal computer to help authors put their work on paper. These kinds of things are tools to assist the human innovator, and that’s the way it needs to be thought about. We should not be debating it. Just say no: this is not going to happen; this is not how we think of it. It’s always the human, and the AI is the tool.
Jane Bambauer: Jeff, do you think, though, that there might be some instances where the human contributions are so diffuse or indirect that we should just treat the output as immediately within the public domain? I’m thinking about some, I don’t know, hacker groups right now that are using AI to just generate a whole bunch of patent ideas or copyrightable expression, and they’re not using them. They’re just creating these giant bodies of material. So if something’s automated enough, do you think that maybe we should recognize no author and no inventor?
Jeffrey Depp: Well, yeah, there’s always the issue of nefarious actors getting in, and we do worry about that in the patent space: generating a bunch of prior art that could undermine the ability to get inventions patented. But I think there’s a better way to deal with that than claiming that the AI is the author and then it’s not eligible, because I think we’re seeing where that leads, and it’s just too messy. There are better ways to handle that kind of computer-generated art through the examination process, and that’s how I would address it.
Satya Marar: All right, we’re almost right at time. Thank you all so much for what’s been a really great and insightful discussion. Looking forward to the next one.
Libby Dickinson: Wonderful. Well, thank you all so much for presenting on today’s program and for sharing your expertise and opinions on this really important topic. And thank you to Satya for moderating. If our audience is interested in learning more about all of our programming here at the Federalist Society, please visit our website, fedsoc.org. Or feel free to send us your feedback; we always love to hear it, at fedsocforums@fedsoc.org. And with that, we are adjourned. Thank you so much.
