Latest / CLEs You Actually Want to Hear / AI and Fair Use in the Wake of Warhol and the Copyright Office Draft Report
Transcript
- 0:03Hello, everyone, and welcome to AI and Fair Use
- 0:07in the Wake of Warhol and Copyright Office Draft
- 0:10Report. I am Samantha Sevis, and I am one of
- 0:13the membership committee chairs of the BHBA's
- 0:15barrister section. And I am delighted to introduce
- 0:18our speaker, Amy Scala. Amy is a partner in the
- 0:21Los Angeles office of Mozilla Lindstrom LLP,
- 0:24where she practices litigation and transactional
- 0:27law in fine art, intellectual property, media,
- 0:30and entertainment. As a part -time lecturer at
- 0:33UCLA School of Law and past chair of the Los
- 0:36Angeles County Bar Association's Entertainment
- 0:38Law and Intellectual Property Committee, she
- 0:41frequently speaks on art law, copyright, and
- 0:44fair use. And before practicing law, Amy was
- 0:47a photo editor of Artform Magazine, and she currently
- 0:50serves on the board of Folkroom Arts, a nonprofit
- 0:53advancing collaboration between art and science.
- 0:56Now, let's get started. Thanks, Samantha, for
- 1:00that lovely introduction and thank you to the
- 1:04BHBA. and the Art Law Committee and our lovely
- 1:09chair, Sarah Conley -Odenkirk. Her leadership
- 1:12has been really wonderful, and I'm so happy to
- 1:16be here with you all. I also want to thank our
- 1:18membership. I am greatly indebted to all of the
- 1:23members, the practitioners, the professors who
- 1:26are part of this community, who have shared their
- 1:28knowledge about AI and copyright and these issues.
- 1:31I don't know that I would be nearly as well informed
- 1:34on these issues if it wasn't for the Beverly
- 1:37Hills Bar. And I'm just really happy to be able
- 1:39to pay it forward a little bit. So without too
- 1:45much more, I will start our presentation. So
- 1:56we're going to be talking about the Warhol case,
- 2:00the Copyright Office Draft Report, and the Fair
- 2:04Use Doctrine in the context of the AI and copyright
- 2:08litigation that we're seeing coming down from
- 2:12the Northern District of California and a Delaware
- 2:14District Court decision in one case. And so before
- 2:18we get too far along, I think we need to understand
- 2:22a little bit about the fair use doctrine in general
- 2:25and what the defense is. So it's a defense. So
- 2:28it's an affirmative defense. The burden of proof
- 2:31is on the defendant. And I think that that comes
- 2:35up a little bit in the Cadry case, as we'll get
- 2:38there in a little bit. But I do think that the
- 2:41burden of proof with respect to the fourth fair
- 2:44use factor in Cadry is an interesting question.
- 2:48The fair use defense is thorny. It has been called
- 2:52the most troublesome in the whole law of copyright
- 2:56by Judge Leonard Hand in 1945. And I think that,
- 3:01you know, troublesome is a great way to describe
- 3:04it. And I think that remains the case today,
- 3:06perhaps even more than at that point. You know,
- 3:11so here are the factors here. You know, I imagine
- 3:13that a lot of you know this, but just, you know,
- 3:15in case I want to make sure that we're all on
- 3:16the same page, While the factors are there and
- 3:20they need to be balanced, fair use really is
- 3:23a flexible concept, as we learn from the Warhol
- 3:26case and from Oracle. While one factor may prove
- 3:30to be more important in other contexts than in
- 3:33others, and application of the factors requires
- 3:37judicial balancing, depending on relevant circumstances
- 3:41and including significant changes in technology.
- 3:44And that is, again, the Oracle case, quoting
- 3:47Sony versus Universal Studios or the Betamax
- 3:51case. So the first fair use factor asks about
- 3:55the purpose and character of the use. Is it commercial?
- 3:58Is it nonprofit? Is it educational? The first
- 4:02fair use factor also asks whether or not the
- 4:05secondary use is transformative. And we're going
- 4:07to talk a little bit more about that. The second
- 4:10factor asks about the nature of the copyrighted
- 4:13work. Is it published? Is it unpublished? Is
- 4:15it fact -based? Is it more creative? If a work
- 4:19is unpublished, it's less likely that a secondary
- 4:22user can make fair use of it. We learn from the
- 4:26courts that the right to release a work to the
- 4:30public for the first time is a right that the
- 4:33author holds that isn't easily usurped via the
- 4:38fair use doctrine. factual versus more creative
- 4:42component, we're asking if there's a thin copyright.
- 4:45Facts are not copyrightable, ideas are not copyrightable.
- 4:48How much of the creative expression is really
- 4:50there? When we're talking about the works at
- 4:55issue in the AI cases, we really are talking
- 4:57about the heart of what the copyright act is
- 5:02intended to protect, these original expressive
- 5:04creative works. Then the third one, asks about
- 5:11the amount and substantiality of the portion
- 5:14used in relation to the whole. So how much is
- 5:16used? And what is used? It's not really a mechanical
- 5:20calculation. It's more about, we see that if
- 5:26you take the essence of someone else's creative
- 5:28work and use it for similar or the same purposes
- 5:30and compete with them. it's probably not gonna
- 5:34be a fair use, right? And so I think in the Harper
- 5:36and Row case, we saw the Supreme Court tells
- 5:40us that you can't take the heart of Gerald Ford's
- 5:43memoir where he's like deciding to pardon Nixon,
- 5:46I think was the moment that the nation decided
- 5:49to publish, right? And so they really did take
- 5:51the heart of it and that wasn't a fair use. The
- 5:54last factor asks about the effect of the use
- 5:57on the potential market for or the value of the
- 6:00copyrighted work. Is there substitution here?
- 6:03Can somebody use the secondary work as a replacement
- 6:06for the first? The fourth fair use factor, as
- 6:10we will see, is really crucial in understanding
- 6:13the decisions that are coming down about generative
- 6:18AI. The first and the fourth factor, spoiler
- 6:23alert, they're the most important. The purpose
- 6:27of the use is discussed a lot in Warhol. And,
- 6:32you know, this question of whether or not the
- 6:34secondary use is transformative. And this transformativeness
- 6:38test, you know, really it comes from a law review
- 6:42article written by Judge LaValle called Toward
- 6:45a Fair Use Standard and was really cemented in
- 6:48copyright jurisprudence in Campbell versus Acuff
- 6:50-Rose. which you'll all remember as the case
- 6:53involving two live crews parody of Roy Orbison's
- 6:57Pretty Woman. In Campbell, the Supreme Court
- 7:00interpreted the first very use factor to include
- 7:02this question of transformativeness very clearly.
- 7:07They tell us to ask whether or not the new work
- 7:12merely supersedes the objects of the original,
- 7:16which comes from the Folsom versus Marsh case,
- 7:18or if it instead adds something new. with a further
- 7:21purpose or different character, altering the
- 7:24first with new expression, meaning, or message.
- 7:28And that's a pretty broad statement. And I think
- 7:34that the Campbell case doesn't really articulate
- 7:36whether or not we are thinking about the purpose
- 7:41of the works. or the purpose of the secondary
- 7:45user. And I think that that's an issue that really
- 7:47comes up in Warhol. And then the fourth fair
- 7:51use factor is undoubtedly the single most important
- 7:54element of the analysis, according to the Supreme
- 7:56Court. And I think we're really seeing this factor
- 7:59do a lot of the work with respect to how fair
- 8:04use is being applied in the AI cases. This factor's
- 8:11gonna look like, as we talked about, the extent
- 8:13of market harm caused by the particular actions
- 8:15of the alleged infringer. And it asks whether
- 8:18unrestricted and widespread conduct of this sort
- 8:21would result in a substantially adverse impact
- 8:24on the potential market for the original. So
- 8:26that's coming from Campbell, and they're quoting
- 8:28Nimr. And it's important when we think about
- 8:32this factor to remember that there may be all
- 8:35kinds of market harm to a copyright holder from
- 8:37a secondary use. like a withering criticism of
- 8:42a movie that just kills the demand for it. That's
- 8:47market harm for sure, but it's not something
- 8:49that is actionable for purposes of copyright
- 8:53and undermining the fair use defense. The only
- 8:58relevant harm here is market substitution, and
- 9:00that's an important thing to remember. In approaching
- 9:06fair use, you know, and new technologies. The
- 9:11courts have sought to further copyright's basic
- 9:15purpose of promoting progress by striking a balance
- 9:19between protecting authors' exclusive rights
- 9:21on the one hand and enabling others to build
- 9:24on those works on the other. And in the Grokester
- 9:28case from 2005, the Supreme Court tells us that
- 9:32the more artistic protection is favored, the
- 9:34more technological innovation is going to be
- 9:37discouraged, and that the administration of copyright
- 9:39law is an exercise in managing the trade -off.
- 9:42Now, that is an incredibly difficult thing to
- 9:46do, and I don't envy the courts in their... having
- 9:51to deal with these questions and grapple with
- 9:53these questions. It's incredibly interesting,
- 9:55but it's also really, really, really difficult
- 9:58to balance these competing interests. How much
- 10:00protection is enough to let innovation flourish
- 10:04but not kill incentive to create new works in
- 10:08the future? How do we strike that balance? It's
- 10:11a real challenge. an art law committee panel
- 10:17here. And I am a practitioner in the art law
- 10:20space for sure. So outside of technology, in
- 10:23the arts, we see that this tension has really
- 10:26been embodied in the various legal disputes involving
- 10:29appropriation artists. And while I wouldn't categorize
- 10:33Warhol as an appropriation artist, pop art for
- 10:35sure, I do think that the Prince case kind of
- 10:39dovetails with those. But Jeff Koons, for example,
- 10:42he was a you know, sued a number of times, and
- 10:45he's relied on the fair use defense in those
- 10:47contexts. And somewhat more recently, about 10
- 10:50years ago now, appropriation artist Richard Prince
- 10:54was involved in litigation involving his appropriation
- 10:59of other artists' works and his own work. And
- 11:02I think that, you know, before diving into Warhol,
- 11:06we really need to understand the context in which
- 11:09Warhol developed, right? So Warhol is a second
- 11:11circuit, decision that was appealed to the Supreme
- 11:14Court. It was affirmed. But the Second Circuit
- 11:17has played a large role in interpreting the transformativeness
- 11:22test and applying the transformativeness test
- 11:25in its first factor analysis. And so you might
- 11:29remember in 2000, photographer Patrick Carey
- 11:35published this book, Yes Rasta. of black and
- 11:39white photographs of the Rastafarian community
- 11:41in Jamaica. And Richard Prince, being Richard
- 11:44Prince, created a series of works called the
- 11:46Canal Zone series in 2007 and exhibited them
- 11:49at Gagosian Gallery in 2008. So the Canal Zone
- 11:54series incorporates the images from Yes Rasta
- 11:57in more or less extreme ways. So this is just
- 12:04one of the works from that series. And so we
- 12:06look at that case, you know, in 2009, Carey U.
- 12:12sues for infringement. And in 2011, the district
- 12:15court, Judge Batts and the SDNY grants summary
- 12:21judgment for Carey U. in favor of infringement.
- 12:24Prince's works are not transformative. They don't
- 12:26comment on or criticize Carey U.'s originals.
- 12:28So they're not entitled to the First Amendment's
- 12:31offense. And, you know, What was kind of extreme
- 12:36in my mind, as somebody who is viewing this at
- 12:38the time, was that the STNY ordered that all
- 12:42of Prince's unsold works from the Canal Zone
- 12:44series and all of the catalogs that were produced
- 12:47on the occasion of the exhibition at Cogosian
- 12:49be impounded and destroyed. That's not, you know,
- 12:53I think that that was the remedy that the plaintiff
- 12:55asked for, but as you'll see in the Cadry case,
- 12:58you know, there, um the court noted specifically
- 13:02they you know the country court is telling us
- 13:03like look you know you don't need to you don't
- 13:06need to stop doing what you're doing you just
- 13:08need to pay for it right and so that's a little
- 13:10bit in in contrast with what we saw here and
- 13:13so i think that the second circuit in ruling
- 13:16uh you know in in reversing and and vacating
- 13:19uh the carry you decision was responding in part
- 13:24to the extraordinary remedy that was ordered
- 13:30in the district court. The Second Circuit reverses
- 13:35and remands, and they say that, look, we need
- 13:38to look at all these works separately. We can't
- 13:41lump them all into one box and decide whether
- 13:44the entire series is fair use or not. It's not
- 13:46what we're going to do. We need to look at each
- 13:48individually. Once we've looked at each individually,
- 13:51then we see that all except for five of them
- 13:54are fair use as a matter of law. These five are
- 13:57going to be remanded to the lower court to decide
- 13:59whether or not they're fair use. The Second Circuit
- 14:03also tells us that the appropriate standard to
- 14:06apply the transformativeness test under the first
- 14:09fair use factor, is you don't need to be commenting
- 14:12on the original work. You don't need to be criticizing
- 14:15it. It can be raw material for your new work.
- 14:18That's totally fine. And so the case settles
- 14:22in 2014. So we never know if, for example, graduation,
- 14:26this one in the top with the figure with the
- 14:30guitar or the bass, that's Prince's work. And
- 14:33then on the other side is Carrie's, right? So
- 14:35this is one of the works that was remanded for
- 14:41a determination of whether or not it was fair
- 14:43use. But in the aftermath of Carry You, the Second
- 14:47Circuit, there was a lot of critics. Nimr criticized
- 14:52the decision, and another appellate court did
- 14:55too, and criticized Carry You in particular.
- 14:57And the Second Circuit in general is giving too
- 15:00great a weight to transformative use and determining
- 15:03fair use. And so it's against that backdrop that
- 15:07we get the Warhol decision. And the Warhol decision,
- 15:10I think, really does need to be understood in
- 15:12the context of Kerry because it's the pendulum
- 15:15swinging, right? Sorry, the Second Circuit was
- 15:19interpreting the transformativeness test incredibly
- 15:22broadly to certain commentators and decided to
- 15:26use this, the Warhol case, as an opportunity
- 15:28to walk it back. So there you'll recall Lynn
- 15:33Goldsmith, she's a rock and roll photographer,
- 15:36you know, doing great work in the 80s and had
- 15:40prints in her studio in December of 1981. She
- 15:43took a series of photographs of him and a couple
- 15:46years later, Vanity Fair asks her to license
- 15:50her picture, one of her pictures from that session.
- 15:53to give to an illustrator to create an illustration
- 15:56for the magazine. And Warhol was that illustrator.
- 16:01He created this silk screen, this purple print
- 16:05silk screen. Goldsmith was credited, and she
- 16:09was paid a license fee. Importantly for this,
- 16:13the license was one time only. So it was limited.
- 16:17It was a limited license, and the scope was very
- 16:21clear. And Warhol ultimately created, I think,
- 16:2416 works based on this photograph. So he plainly
- 16:28exceeded the scope of the license. I think that's
- 16:30also another piece of this that we really do
- 16:33need to keep in mind. So 2016 comes around, Prince
- 16:39passes away, and Vanity Fair asks the Andy Warhol
- 16:44Foundation to license an illustration. for the
- 16:49cover one of Warhol's print series. Goldsmith
- 16:55doesn't get credit, she doesn't get a licensing
- 16:56fee, she's not involved at all. She sees the
- 16:59magazine and is like, hey, that's my picture,
- 17:02and asks the Warhol Foundation for a license.
- 17:06Instead of licensing the picture, the Warhol
- 17:09Foundation sought a declaratory judgment in the
- 17:11SDNY for a determination that Warhol's use was
- 17:17fair use. So in the SDNY, we see that summary
- 17:21judgment for the foundation, Andy Warhol, is
- 17:24granted three of the four factors, favor Warhol,
- 17:28and Goldsmith appeals, and the Second Circuit
- 17:32reverses. And they even mentioned, Carrie, you
- 17:34in this decision. They're reversing, they're
- 17:36taking this as an opportunity to really walk
- 17:38back the broad interpretation of the transformativeness
- 17:42test that we saw. And I think maybe that was
- 17:46unexpected for for the Warhol Foundation, I don't
- 17:50know, but I speculate. Anyway, the Warhol Foundation
- 17:55appeals and the Supreme Court affirms the Second
- 18:00Circuit and says, look, the Andy Warhol Foundation's
- 18:03use in this licensing context is not transformative.
- 18:08And what we do see is that when reading the decision,
- 18:14the scope of the Warhol decision is really narrow.
- 18:20The decision is limited to a licensing purpose
- 18:24for an illustration on the cover of a magazine.
- 18:28So crucially, we don't get a determination of
- 18:32whether Prince's series of I think 16 silk screens
- 18:38of Prince, sorry Warhol's series of 16 silk screens
- 18:42of Prince are themselves fair use, right? The
- 18:47Supreme Court never answers that question. So
- 18:49we really don't know whether the underlying works
- 18:52are fair use or not. What we do know is that
- 18:56in the context of a magazine cover and licensing
- 19:00an image for purposes of illustrating an article
- 19:04on the musician, we know that that is not fair
- 19:07use. But in the Supreme Court's decision, They
- 19:12tell us that different uses have to be considered
- 19:15separately and that the same copying may be fair
- 19:17when used for one purpose but not for another.
- 19:21And so, you know, it really does beg the question,
- 19:25was Warhol's original creation of these paintings
- 19:27fair use or not? And I think that sometimes that,
- 19:30you know, that that's not, you know, considered
- 19:34when Warhol's being applied. In addition to,
- 19:39you know, a very narrow purpose and evaluating
- 19:42a very, very, very narrow purpose in this licensing
- 19:44context, we also see that the Warhol decision
- 19:48applies, you know, they're only evaluating the
- 19:50first very use factor. And so I think that that's
- 19:53playing into this as well. Some language that's
- 19:56really interesting from the Warhol decision is
- 19:58that we get a clear statement that uses don't
- 20:02turn on the subjective intent of the user. So
- 20:06what Prince was thinking when creating the work
- 20:09or what the Andy Warhol Foundation was thinking
- 20:11when licensing the work is irrelevant to this
- 20:14determination. What matters is an objective inquiry
- 20:17into what use was actually made, what the user
- 20:20did with the original work. And so when we think
- 20:24about that, it's very clear that when we look
- 20:29at the magazine covers side by side, the purpose
- 20:34there is almost the same as the Supreme Court
- 20:37defined the purpose. So we can debate about whether
- 20:43that's the right call, but it's the decision.
- 20:49And another thing that we learned from Warhol
- 20:53is that The opinion makes clear that it's concerned
- 20:57with the standard for transformativeness under
- 20:59the first fair use factor subsuming the derivative
- 21:02work right. And that broad interpretations or
- 21:06an over broad concept of transformative use is
- 21:10going to subsume the derivative work right. And
- 21:16it's necessary for the new work to comment on
- 21:19or criticize the original in order to be transformative
- 21:23as that term is used in the first fair use factor
- 21:27analysis. Even significant alterations are not
- 21:31going to be enough if the use ultimately serves
- 21:33a similar purpose to the original and may produce
- 21:36instead a derivative work and demonstrate the
- 21:39need for licensing. So this is a piece of the
- 21:43Warhol decision that I think is worth drilling
- 21:48down on. and is an important part of it going
- 21:52forward. Given all that background with the very
- 21:58narrow scope of the purpose of the use that was
- 22:00analyzed and the fact that we're only getting
- 22:03a decision regarding the first fair use factor,
- 22:07not all of them, we see that Warhol going forward
- 22:12is being applied fairly broadly. And I think
- 22:18that that's an important thing to remember because
- 22:21as practitioners, we want to keep in mind that
- 22:26maybe we should be keeping Warhol in its narrow
- 22:32lane instead of jumping to the language that's
- 22:36in there and then applying it more broadly, which
- 22:38I think is what is happening. Shifting gears
- 22:43now to the Copyright Office pre -publication
- 22:45report. You know, the Copyright Office has been
- 22:50releasing reports. It's released three reports
- 22:53since July of 2024 on issues stemming from AI
- 22:56and copyright. So the first part was digital
- 23:00replicas. The second part was copyright ability.
- 23:02And the third is generative AI training. In large
- 23:07part, the reports evaluate and analyze the comments
- 23:10that the Copyright Office received in response
- 23:12to its notice of inquiry. In August of 2023,
- 23:17the copyright published in the Federal Register,
- 23:20a broad series of questions that it wanted comments
- 23:23on in order to evaluate this novel issue and
- 23:29develop its perspective. is a really interesting
- 23:39read if you have some time. It's definitely worth
- 23:43taking a look at. But it's also worth pointing
- 23:47out the sort of interesting circumstances surrounding
- 23:51its release. It's a pre -publication report.
- 23:55which as highlighted at the time by Aaron Moss
- 23:59of Copyright Lately, issuing pre -publication
- 24:02or draft reports is totally not standard for
- 24:05the Copyright Office. And the release of this
- 24:07report was between two pretty extraordinary firings
- 24:11that happened. So on a Thursday, the... long
- 24:16time librarian of Congress, Dr. Carla Hayden
- 24:19was dismissed. The report is released that Friday
- 24:23and then that Saturday, the register of copyrights,
- 24:28Shira Perlmutter was fired by the administration.
- 24:32And so, you know, given that the Copyright Office
- 24:37is a part of the Library of Congress, and Congress
- 24:42is obviously part of the legislature, is the
- 24:44legislative branch, not the executive. There
- 24:46is a tension there between the authority to actually
- 24:49do that by the executive branch, but that is
- 24:52a separate, fascinating question that I cannot
- 24:55delve into here. But for our purposes, it's important
- 25:00to have that context because we don't know the
- 25:04weight of this report is going to have going
- 25:07forward. It's really not clear at all. It could
- 25:10be pulled from the website tomorrow. It could
- 25:14be dramatically edited and revised. But the report,
- 25:18as it was published in May of 2025, tells us
- 25:22a lot. And I can't get into all of it. Again,
- 25:26it is really interesting reading, especially
- 25:29because it describes you know, so concisely the
- 25:34various positions of various stakeholders, you
- 25:38know, who really have skin in this game, you
- 25:41know, and their positions with respect to these
- 25:43issues. And I think that that's just so helpful
- 25:46to understand these issues on a practical level.
- 25:50And with respect to, you know, fair use in particular,
- 25:54right? So the report is divided up into five
- 25:56parts. Fair use is just one, but when we look
- 26:00at what the Copyright Office is saying there,
- 26:03it's kind of no surprise in order to evaluate
- 26:07whether a particular model is transformative,
- 26:10you know, depends on its circumstances and depends
- 26:13on the facts of the case, just like the fair
- 26:15use defense itself, right? We're never going
- 26:17to have, you know, a boilerplate answer of, you
- 26:21know, all models are fair use or none are. They
- 26:24all exist on a spectrum. and some systems are
- 26:27going to undoubtedly be transformative while
- 26:29others are not going to rise to that level. With
- 26:33respect to the market harm, as I was reading
- 26:38the report, this idea of statistical similarity
- 26:42caught my attention. I don't know that the report
- 26:44uses that particular phrase, but it does encapsulate
- 26:48something that's raised a few times. I think
- 26:54that maybe going forward, and I haven't fleshed
- 26:57this out yet in my own mind, but I think statistical
- 26:59similarity may be worth thinking about as a way
- 27:01to understand and maybe bridge copyright principles
- 27:04and standards of infringement with the complexity
- 27:07of AI models, right? So if you, this picture,
- 27:11which I saw actually in a Beverly Hills Bar presentation
- 27:15on AI a while back, it really helped me visualize
- 27:21you know, the process in a diffusion model where
- 27:24we're going from, you know, static into an actual
- 27:29image. And so, you know, the statistical similarities
- 27:32between original works and copyrighted work,
- 27:36sorry, and secondary works and ones that are
- 27:40output from AI systems, it is an interesting
- 27:43way of maybe thinking about this. In the cadre
- 27:48case, for example, the court does note the relationship
- 27:52between copying statistical elements of plaintiff's
- 27:55books and how caught and how. this is copying
- 27:59the creative expression embodied within them.
- 28:02The court tells us that even though LLMs may
- 28:04only learn about statistical relationships, those
- 28:07relationships are the product of creative expression.
- 28:10The statistical relationships in the organization
- 28:13of words in a book, that is the creative essence
- 28:18of that book and is on a certain level. I think
- 28:22it's an interesting question to consider. Another
- 28:25aspect of the Copyright Office report that is
- 28:29really interesting is that the report really
- 28:33articulates this broader market dilution harm.
- 28:38We see that it's a novel harm. It's not something
- 28:40that's really been evaluated in case law before,
- 28:44but it could very well be an actionable harm.
- 28:48At least in the Copyright Office's view, seems
- 28:52plainly relevant. to the questions raised by
- 28:57the fourth fair use factor. So moving along to
- 29:02the lawsuits that have been filed. And so to
- 29:07give us a little bit of context here, fair use
- 29:10has been... the way that disputes between technological
- 29:18innovations and rights holders have been resolved
- 29:20throughout our history. When the camera came
- 29:24out, although the Cerrone case was really questioning
- 29:28whether or not photographs were copyrightable
- 29:31at all, I think it is a case that arises because
- 29:37of technological innovation and these tensions.
- 29:41At the turn of the 20th century, we saw the player
- 29:44piano issue arise, and we get a case law on it,
- 29:47and then we get the 1909 act, which addresses
- 29:51the issue squarely. Closer to our own time, we
- 29:56see in the 80s, we had the Betamax case, the
- 29:59Sony case, and there, the VCR, the Betamax was
- 30:06fair use, and using it to record television programming
- 30:09was was a fair use. And the court there told
- 30:15us that the technology was capable of commercially
- 30:18significant non -infringing uses. And that question,
- 30:24I think Sony's, it hasn't been overruled obviously,
- 30:27but it's been distinguished. But I do think that
- 30:33this capable of commercially significant non
- 30:36-infringing uses is still a concern by the courts.
- 30:40It's just being captured in the discussion of
- 30:43whether or not there's a public benefit to the
- 30:47use, which is something that's taken up in the
- 30:49fourth fair use factor analysis. In 2005, we
- 30:55saw the Grokester case where Groster was liable
- 30:59for copyright infringement via the contributory
- 31:01and vicarious infringement doctrines, distinguished
- 31:04Sony again. There, though I think the Groster
- 31:07case is important in the cases that we're seeing
- 31:10come down because of its strong position with
- 31:15respect to piracy and theft. Particularly in
- 31:18the Bart's case, the issue of piracy was paramount
- 31:22to that court. The issue of good faith or bad
- 31:28faith is an unsettled area in fair use doctrine,
- 31:33because there's a question of whether or not
- 31:35it should be encapsulated within the character
- 31:38of the use. Did you use it in good faith or not?
- 31:44And we've gone, at least in the Ninth Circuit,
- 31:47here in the Ninth Circuit, this is an unsettled
- 31:49question. But obtaining a work illegally from
- 31:56the outset and having your copy be an illegal
- 32:00copy and then using it to create a work. That
- 32:04was a very important fact to the Bart's case.
- 32:08It's also incidentally an important fact in international
- 32:12regulation with respect to AI in the EU directive
- 32:17on this issue. For example, we see that the safe
- 32:19harbor that's created for data mining, using
- 32:25a lawful copy and making a lawful copy of something
- 32:28as a prerequisite to being able to take advantage
- 32:31of this safe harbor. So I do think that the court's
- 32:36understanding of piracy is an important piece
- 32:40of this going forward. And then in the Oracle
- 32:43case, sorry, in Authors Guild versus Google,
- 32:49which is another case in the Second Circuit,
- 32:53There was a fair use defense that was unanimously
- 32:56affirmed. The courts were in agreement, which
- 32:58is often not the case in fair use decisions.
- 33:01You see a flip -flop all the way up, which shows
- 33:05just how hard these questions really are to grapple
- 33:07with. For Google Books, the creation of the search
- 33:12functionality and the display of snippets is
- 33:14non -infringing fair use, and it's highly transformative.
- 33:18We see this highly transformative spectacularly,
- 33:21so we see that language coming out in the BARTs
- 33:25case and in the cadre case as well. Using training,
- 33:31LLMs or image generation models, whatever. using
- 33:37copyrighted works is a highly transformative
- 33:39use. And so what's going on there is we're seeing
- 33:42the market harm factor take center stage. And
- 33:46then in Oracle, it was another of these flip
- 33:49-flop situations because the Supreme Court reverses
- 33:53the federal circuit and finds that Google's use
- 33:56of Oracle's code was fair. It was, you know,
- 34:00the re -implementation of copied code was transformative
- 34:03because it, quote, furthered the development
- 34:05of new computer programs by enabling programmers
- 34:08to use existing skills in a new mobile platform.
- 34:12You know, and then we get some language about
- 34:15fair use that is, you know, it permits the courts
- 34:18to avoid the rigid application of the copyright
- 34:21statute when on occasion it would stifle the
- 34:24very creativity for which the law is designed
- 34:26to foster. And so, you know, there we see, you
- 34:31know, strong statement that technological innovation
- 34:35really does have a place, you know, but is it
- 34:39tipping the scales too far? I think the Kadrian
- 34:43-Bart's cases are telling us that, you know,
- 34:45perhaps. In any event... The litigation landscape
- 34:50with respect to AI is changing all the time.
- 34:52I, like many of you, I'm deeply indebted to Professor
- 34:57Lee and his incredibly helpful and insightful
- 35:00website and newsletter that keeps me up to date
- 35:05on all of this. But this map, which he updated
- 35:08September 11th, shows that we've hit 51 AI cases
- 35:12here in the US. And there is another tracker
- 35:16that he's created for cases on abroad that is
- 35:21too much for us to discuss here. So only three
- 35:26of these cases that I'm aware of have issued
- 35:28a ruling on fair use. And so the first was the
- 35:32Thompson Reuters case, where the district court
- 35:35of Delaware found that Ross infringed. Thompson
- 35:45Reuters copyrights and that the fair use defense
- 35:48does not apply. Reuters who owns Westlaw sued
- 35:52Ross for using all of Westlaw's headnotes. The
- 35:56summaries of key points of law and case holdings
- 35:58that exist to train a competing AI driven legal
- 36:01research search engine. Importantly, this new
- 36:05product that Ross has developed is not generative
- 36:08AI. So it's not creating new material. from what
- 36:13it was trained on. It operates a little bit differently
- 36:16on a technical level, so then that's an important
- 36:19distinction here. So with the first fair use
- 36:23factor, the court found that Ross's use was for
- 36:25a really similar purpose. It was a commercial
- 36:27use, and it wasn't transformative. It's not generative,
- 36:31right? So the first factor went to Westlaw. The
- 36:34fourth factor also went to Westlaw. The derivative
- 36:39market for AI training purposes is harmed. Ross
- 36:43also meant to compete with Westlaw by creating
- 36:45a direct market substitute, which is the harm
- 36:48that is protectable under the Copyright Act here,
- 36:54even in the context of a fair use defense. This
- 36:58case I mentioned is distinguishable. Ross's AI
- 37:01is not generative. It's reproducing more of Westlaw's
- 37:05original. And, you know, the headnotes themselves
- 37:08were considered original and therefore warranting
- 37:10of copyright protection. That was an issue in
- 37:12that case that the copyright registrations weren't
- 37:15necessarily valid, but the court found they were.
- 37:18So the Bart's case involves Anthropix, LLM, Claude,
- 37:27and its various incarnations. And as I've alluded
- 37:32to, this case makes a strong distinction between
- 37:36the library copies that Anthropic purchased.
- 37:39They actually bought a bunch of books and took
- 37:42them apart and scanned them. Then they also pirated
- 37:45a bunch of libraries. The court evaluates the
- 37:48fair use defense for these groups differently.
- 37:51And I think that that's really key. This theft
- 37:56from the outset issue is, even in the first line
- 37:59of the decision, you'll see it. It is very important
- 38:02to the court if that happens. Procedurally, this
- 38:07case is interesting because Anthropic moved for
- 38:10early summary judgment on its fair use defense
- 38:12before class certification. The decision is only
- 38:17going to bind those parties that are involved
- 38:22in litigation, but not any of the members of
- 38:24the class that have yet to be certified. Strategically,
- 38:29that's an interesting maneuver, one that we see
- 38:32also done in the Kadri case. where both parties
- 38:36move for summary judgment and before class certification.
- 38:40With respect to the first fair use factor, the
- 38:45first fair use factor favors anthropics lawful
- 38:47copying. The purpose and character of using the
- 38:50works to train the LLMs and to train Claude was
- 38:53transformative spectacularly so. But this doesn't
- 38:58excuse the pirated library copies. It's pretty
- 39:02important to the Bard's court. And it's interesting
- 39:06because this issue is not important to the cadre
- 39:09court. And you know, it's just, you know, it's,
- 39:12it's, fair use is thorny, like I said. But so
- 39:18the use of the books that issue to train Claude
- 39:20was exceedingly transformative and, and was a
- 39:24fair use under the Copyright Act. But piracy,
- 39:28but piracy of otherwise available copies is inherently
- 39:32irredeemably infringing, the court tells us,
- 39:35even if the pirated copies are immediately used
- 39:38for the transformative use and immediately discarded.
- 39:43So you can't get around a finding of infringement
- 39:47if you've started with pirated copies, no matter
- 39:50how transformative your use is, according to
- 39:53this court. the BART court goes on to distinguish
- 40:01the Google Books case, where they're saying,
- 40:04look, the university libraries and Google went
- 40:06to exceedingly great lengths to ensure that all
- 40:09copies were secured against unauthorized uses.
- 40:12And that didn't happen here. So the first fair
- 40:15use factor points against fair use for the central
- 40:19library copies made from pirated sources. and
- 40:22no damages from pirating copies could be undone
- 40:25by later paying for copies of the same works.
- 40:28I think that this distinction between the pirated
- 40:31versus lawfully obtained copies in the Bart's
- 40:36case really drove the settlement there. That's
- 40:39my take because irredeemably infringing, the
- 40:46court tells us that. If you were anthropic, You
- 40:50know, anyway, it's an interesting question. So
- 40:53with respect to the fourth fair use factor, it
- 40:57also favors anthropic. And the Barts Court's
- 41:01not concerned with this market dilution, you
- 41:03know, indirect harm that the Copyright Office
- 41:06report raises. They're just not concerned about
- 41:08it at all and tell us that that's not the kind
- 41:11of competitive or creative displacement that
- 41:13concerns the Copyright Act. And so there, The
- 41:17court seems to really be going back to the idea
- 41:20that not all harms are actionable harms. It's
- 41:24only this market substitution issue. That brings
- 41:28us to the Kadri case, which was judges right
- 41:33down the hall from each other, issued a couple
- 41:35of days apart. This case comes out in 2023, a
- 41:40group of book authors, I think 13 of them, Suze
- 41:43Mehta. and it's open source LLM Llama in Llama's
- 41:48incarnations. So again, they use this procedural
- 41:52device of making summary judgment motions early
- 41:56before class certifications. So the decision
- 41:58only binds the 13 authors who are the named plaintiffs.
- 42:04So the Kadri case, is really different. The decision
- 42:08is really different than the Bart's case. And
- 42:10I think they're really helpful to read together,
- 42:13if a little confusing, to see what we can learn
- 42:17for moving forward. But the cadre court emphasizes
- 42:21that if a fair use defense fails, the consequence
- 42:23isn't an injunction. So not like carry you. But
- 42:26instead of payment, the defendant will only be
- 42:29forced to stop what they're doing if they're
- 42:31unwilling or unable to pay for the right to do
- 42:33it. So The cadre court also doesn't really think
- 42:37that it's reasonable. I think that the court
- 42:39calls it nonsense that AI companies wouldn't
- 42:43be able to develop the technology if they were
- 42:48required to license all of the copyrighted works.
- 42:51The court thinks that's silly. They're making
- 42:52billions of dollars. Come on, some of it can
- 42:54go to these rights holders if their contributions
- 42:57are really as crucial as you say. you know, the
- 43:06burden of proof, the court, as we'll see when
- 43:09we talk about the market harm factor, the fourth
- 43:12fair use factor, the court... spends a significant
- 43:17amount of time criticizing the plaintiffs for
- 43:20failing to develop this adequate market harm
- 43:22claim and provides essentially a blueprint for
- 43:25the next plaintiff. It seems to acknowledge the
- 43:28criticism though, and goes on about the burden
- 43:31of proof and identifies that the rights holder
- 43:36doesn't have to prove or present evidence of
- 43:38market harm, but they do bear some initial burden
- 43:40of identifying relevant markets. And that's emphasized
- 43:44in the case. And so I think at root, the cadre
- 43:49court is looking at the underlying incentive
- 43:51to create and looking at the purpose of copyright
- 43:54under the IP clause of the constitution and concludes
- 43:56that flooding the market with gen AI created
- 43:59works is going to dramatically undermine the
- 44:01incentive for human beings to create the old
- 44:04fashioned way is what the court says. You know,
- 44:07and to me, Gen. AI tools are tools and people
- 44:12who are creators are gonna be using these tools
- 44:14to create. So I don't know that this ruling really
- 44:19encapsulates that nuance, but with respect to
- 44:23the first fair use factor, again, no serious
- 44:26question, highly transformative. And that makes
- 44:30sense. The court also addresses the shadow library
- 44:33issue here and raises the unsettled, you know,
- 44:36good faith, bad faith distinction, but doesn't
- 44:39really find it super relevant that a lot of this
- 44:41stuff is pirated, you know, based on case law
- 44:46because, you know, the whole point of the fair
- 44:48use analysis is to determine whether a given
- 44:51act of copying was unlawful. So it doesn't really
- 44:53matter where you got it. I don't know if I agree.
- 44:56You know, I mean, the parts, I don't know. I
- 44:58leave it to you all to think about that. With
- 45:01respect to the fourth fair use factor, the court
- 45:03plainly sees that this factor should go to the
- 45:06authors, but finds for meta anyway. The court
- 45:13emphasizes that the relevant market harm includes
- 45:15indirect market substitution, which is this idea
- 45:18from the Copyright Office Draft Report, flooding
- 45:21the market, indirectly substitutes for the originals,
- 45:24and then people are going to be disincentivized
- 45:26to create. Right? The example used in the cases,
- 45:33Agatha Christie, the next Agatha Christie, I'll
- 45:35never bother to pick up a pen, I think the judge
- 45:37says. So the plaintiffs in Kadri focused their
- 45:41arguments on the harm to the author's licensing
- 45:45market. And don't really touch this totally novel
- 45:48indirect harm because of market dilution issue
- 45:53that's raised. And so to my mind, this ultimately
- 45:56almost seems like an advisory opinion to a certain
- 45:59extent, at least this portion of it, because
- 46:01the court's talking about a hypothetical plaintiff
- 46:03who could raise this, and this is how you should
- 46:06do it. So I think that's really, really, really
- 46:12interesting. And the other members of this class
- 46:16who have not you know, been certified yet and
- 46:22who can bring claims against Metta will be able
- 46:25to take this and go running with it. But too
- 46:28bad, so sad for the 13 who lost in summary judgment
- 46:31because they, I guess, didn't identify the relevant
- 46:34market with sufficient specificity to the court.
- 46:38So this, you know, in terms of going forward
- 46:42is a very strong statement that, you know, market
- 46:47dilution will cause plaintiffs to decisively
- 46:50win the fourth factor and thus win the fair use
- 46:54question overall. That tells us which way this
- 46:59court believes this issue is going to go. To
- 47:03me, I will leave you with a quote by Sam Altman.
- 47:09who tells us, I think he did a Ted talk in the
- 47:12spring that I watched and I pulled this from
- 47:15there, that we need to figure out some kind of
- 47:17new model around economics of creative output.
- 47:21I think to a certain extent that's true. Is fair
- 47:24use the best place to be resolving these crucial
- 47:27issues? Is regulation in order? Is some kind
- 47:31of industry agreement, professional associations?
- 47:35I don't know the answer. I just leave you with
- 47:37the question because I don't, you know, it seems
- 47:40like a very thorny area of law, a very unsettled
- 47:44area of law to leave these big questions. So
- 47:47with that, if you have any questions, I would
- 47:51love to hear from you. And in the couple of minutes
- 47:53that we have, I'd love to answer any questions
- 47:55that are in the chat. All right. We currently
- 48:04don't have any questions just yet, but we'll
- 48:06allow a minute or two for that to go about. Do
- 48:10you have any more closing remarks or important
- 48:14tips to remember? I just, you know, I'm thankful
- 48:18to be here and I appreciate all of your attention
- 48:21and, you know, I would love to spend some time
- 48:23talking about these issues with you. You know,
- 48:26in the future, it's something that's developing
- 48:28and changing all the time. And I think we're
- 48:31really going to be seeing a lot more activity
- 48:33in this area. And hopefully, the more we, you
- 48:37know, talk with each other and learn about these
- 48:39issues, you know, more answers will come. All
- 48:44right. It looks like Marina Bagured is asking,
- 48:48can you comment on the settlement and the flood
- 48:50case? Yeah, you know, I do think that The settlement
- 48:55there, I think it's related to this piracy issue,
- 48:59this irredeemable infringement perspective that
- 49:02the court seems to have. I think it's interesting
- 49:05that the court's not approving the settlement
- 49:07because it's going to bind the class members
- 49:09who were not involved in the summary judgment
- 49:12motion. I haven't checked into it today, but
- 49:19I do think it's going to be very interesting
- 49:21to see how that plays out. I don't know that
- 49:25Meadow would have made the same decision if,
- 49:28you know, because the decision in Kadri is just
- 49:30so divergent from the Bart's case and the role
- 49:34of the pirated libraries or the shadow libraries
- 49:37is so crucial in the one and not very relevant
- 49:41in the other. Thank you. So Sean is asking further
- 49:50to the economic model, the license deals are
- 49:52using digital proxies to allocate license fees
- 49:54to works. Do you have thoughts on appropriate
- 49:56digital proxy methodology? You know, I don't
- 50:01because I'm not, I will have to drill down on
- 50:03this. But if you want to reach out and tell me
- 50:06more about it, I would love to give you an informed
- 50:09opinion when I form one, but I don't have one
- 50:11right at this moment. Okay, let's see. Oh, Bianca,
- 50:16the Sam Altman quote is from his TED talk. You
- 50:18can find it on YouTube. But I think my slides
- 50:22will be available sometime soon. I didn't make
- 50:24them quick enough for them to be posted before
- 50:27the presentation, but you should be able to find
- 50:29it there too. But it is on YouTube if you look
- 50:32for his TED talk from the spring. I think April,
- 50:35maybe. Am I familiar with a case similar to the
- 50:42Meta case where authors have successfully articulated
- 50:46the relevant market that was harmed. No, we haven't
- 50:49seen that yet. At least as far as I know, and
- 50:52maybe Professor Lee could jump in and tell us
- 50:55if there's another case. I know that the studios
- 50:57have just jumped into the fray in bringing AI
- 51:01cases, but this. This market harm issue, this
- 51:07indirect market harm via dilution, is not something
- 51:11that we've seen in case law before. It's something
- 51:13that I think comes up for the first time really
- 51:16clearly in the copyright office draft report
- 51:20and was expounded upon at length in cadre. Plaintiffs
- 51:25who are suing Metta after the cadre decision
- 51:28will obviously make this argument, but we haven't
- 51:30seen it yet because we're not too far in those
- 51:34cases yet. How am I advising my artist clients
- 51:47with respect to this issue? Watermarks. If you
- 51:51put something online that's your creative work,
- 51:55especially if it's a visual work, watermarks
- 51:57are like incredibly thorny problems for training
- 52:02purposes, and your work will probably get...
- 52:05filtered out because of watermarks. I would use
- 52:09watermarks, I would share sparingly, and I would
- 52:13make robust use of contract law wherever possible
- 52:18to make sure that you're prohibiting distribution
- 52:21beyond what you want. That's not the greatest
- 52:26answer because things get stolen as we see, but
- 52:31those are some broad strokes. recommendations.
- 52:42Suzanne, are there other questions that you see?
- 52:47Let's see. I do see some. Oh, here's one about
- 52:56laws, regulations. So I don't know. I think we're
- 52:59over time. So Suzanne, please tell me if we should
- 53:02shut it down. No, we're good on time. We have
- 53:05a couple minutes, actually. We have about seven
- 53:06minutes left, so we can dig into things. OK,
- 53:08great. So what kinds of laws are being drafted
- 53:12to control how AI is developed? So the regulatory
- 53:15side of this is very interesting. And so a number,
- 53:19I mean, a huge amount of laws have been introduced
- 53:23on the state level. Some of them are passed and
- 53:29some of them are not. But when you compare the
- 53:31number of bills that have been introduced to
- 53:33the actually the number of laws have been enacted,
- 53:35it's very different. There's a lot of ideas,
- 53:38but there's not a lot of concrete determination
- 53:40on what we want to do. There's also been a statement
- 53:45that was in a bill on the federal level, and
- 53:49I honestly don't know what happened with it because
- 53:51there's so many things to follow lately. The
- 53:57idea that maybe... We shouldn't pass any laws
- 53:59about AI for the next 10 years, is what I recall.
- 54:03And that was coming from the federal side of
- 54:06things, which would prohibit, obviously under
- 54:08the Supremacy Clause, the ability for states
- 54:10to legislate in this area if they wanted to.
- 54:13And so I don't think that that was made law,
- 54:16but it's been an idea that's tossed around. We
- 54:19see a lot more regulatory action in the international
- 54:23context, in the EU, in the UK. And in Japan,
- 54:30we see that there's a lot more activity on the
- 54:33regulatory side. And there, there's kind of safe
- 54:37harbor carve -outs that if you adhere to certain
- 54:41things, then as an AI company, you can take advantage
- 54:44of those safe harbors. But lawful copying is
- 54:48one requirement to taking advantage of those
- 54:52safe harbors. At the end of the copyright office
- 54:55draft report, is an overview of the international
- 55:01landscape, which includes a lot of information
- 55:03about regulatory issues. Let's see. There's another
- 55:14question about a licensing of content for uses,
- 55:20AI training data. Yes. The courts are... At least
- 55:27the cadre and the Barts court diverge about the
- 55:31availability of licensing content for AI training
- 55:36purposes. There is case law in the fair use context
- 55:41that says that, look, if someone's making a transformative
- 55:44use of your work, a plaintiff is always going
- 55:47to lose the right to license to that particular
- 55:50defendant if a fair use decision is found. So
- 55:53in order to avoid the market harm factor as becoming
- 55:57circular, it's necessary to acknowledge that
- 56:01the plaintiff is always going to lose that licensing
- 56:03revenue, but that's not the relevant harm here.
- 56:06And that's really interesting. And I think that
- 56:09that's discussed more at length in the cadre
- 56:11case, I think. But it is an interesting question.
- 56:16It's also clear that there is a market for this
- 56:20developing. It's been developing, and there's
- 56:24a lot of interest in licensing, and there's AI
- 56:27models that use clean data. Getty Images has
- 56:30entered into a licensing agreement for its content.
- 56:35I think the Atlantic entered into a licensing
- 56:38agreement to license its content for AI training
- 56:42purposes. The Copyright Office's report makes
- 56:49clear that they really do feel like we should
- 56:53let this market develop. There is definitely
- 56:56the opinion in the report that as stated by all
- 57:01of the stakeholders who responded to the notice
- 57:03of inquiry that compulsory licenses are a terrible
- 57:07idea, and we shouldn't do that, and that there
- 57:10is this growing market for licensing copyrighted
- 57:17content for AI training purposes. In the image
- 57:20space, it seems clear that copying all the images
- 57:25on the internet and doesn't give you as good
- 57:29of a generative model as highly curated data
- 57:34sets of images. With respect to images, it seems
- 57:38like the licensing market will develop on its
- 57:41own because the quality of your model is really
- 57:44going to be dependent upon what you're putting
- 57:46into it in this particular space. I don't know
- 57:50that that's the case with large language models
- 57:53where It does seem clear that the more stuff
- 57:56you feed the trainer, the model for training
- 58:00purposes, the better output you get. But for
- 58:04images, that doesn't seem to be the case right
- 58:06now. Curious if I predict a circuit split. I
- 58:15think the Second Circuit has been pretty slapped
- 58:18down when it comes to its fair use jurisprudence
- 58:22and really tried to course correct with Warhol.
- 58:28The Ninth Circuit is kind of always a crapshoot.
- 58:30I can never predict what's gonna happen in our
- 58:32courts. California is just so wonky. I imagine
- 58:37that we're gonna see, I mean, I guess I'll say
- 58:40this, given the, difference of opinion with two
- 58:45judges who are literally right down the hall
- 58:47from each other in the Northern District of California.
- 58:50I think the chances are pretty good that there
- 58:52will be a circuit split on these issues and how
- 58:55the Supreme Court chooses to resolve the questions
- 58:58here is anybody's guess. Let's see. Can we create
- 59:09laws so that AI technology must be developed
- 59:12in such a way as to create tags for all copyrighted
- 59:14materials used so as to provide protection? That's
- 59:18such an interesting idea. I love that. In the
- 59:21art context, that's something that had been attempted
- 59:24to be done using blockchain technology for authentication
- 59:28and provenance purposes, like the chain of title
- 59:31of an artwork. The market itself was really resistant
- 59:35to that idea. I think this will impact anonymity
- 59:42with respect to creating works. And people might
- 59:48not want to be associated with particular content.
- 59:51Sometimes yes, and sometimes no. So I do think,
- 59:53you know, that's an interesting question. And
- 59:57there is ways to create tags in a certain sense,
- 1:00:00right? You have metadata, you have watermarks,
- 1:00:02you have other ways to identify material as yours.
- 1:00:05But the models are getting better and better
- 1:00:08at filtering all of that out. One of the issues
- 1:00:12that we see in the copyright cases, which claims
- 1:00:16that have not been dismissed and were not at
- 1:00:18issue on summary judgment are the, you know,
- 1:00:21removal, the automated removal of content management
- 1:00:24information. So I think if we created tags to
- 1:00:29provide protection, the models would be, you
- 1:00:32know, moving in a direction of trying to automate
- 1:00:35and remove those tags, unless there was some
- 1:00:40kind of agreement in place as to a revenue sharing
- 1:00:44method, kind of like what Sam Altman seemed to
- 1:00:46be talking about in the TED Talk. But let's see.
- 1:00:55If technology was regulated to the point that
- 1:00:57copyrighted materials used to feed AI algorithms
- 1:01:00is developed, then it may help solve the problem.
- 1:01:02Yes, maybe, maybe. And this is where we need
- 1:01:05a data scientist to help us because art law only
- 1:01:09takes me so far with respect to this. And I'm
- 1:01:12curious. I love learning about all this stuff,
- 1:01:14but I am certainly not an expert when it comes
- 1:01:16to the mechanics of the models. Should AI companies
- 1:01:20be treated as publishers or social media platforms
- 1:01:23under Section 230 of the Communications Decency
- 1:01:26Act? You know, that's a good question too. My
- 1:01:32initial thought, I've never thought about this
- 1:01:34before, so don't hold me to this, anyone who's
- 1:01:38listening, but my initial thought is that the
- 1:01:42models are creating new content, right? The safe
- 1:01:45harbor of Section 230 is really for people uploading
- 1:01:51copyrighted works and you as an ISP, you're not
- 1:01:55aware of that, right? So you can't necessarily
- 1:01:58be held liable for reproducing copyrighted content
- 1:02:02and infringing because you're not the one who's
- 1:02:05actually uploading it. But here the models that
- 1:02:08the product that's... providing content to the
- 1:02:12public, the generative information that's coming
- 1:02:16from the model to a user is not going to be that
- 1:02:19copyrighted content. At the beginning of these
- 1:02:23cases and when this first became an issue in
- 1:02:26the law, there was a lot of discussion of infringement
- 1:02:29on the input side and infringement on the output
- 1:02:31side. With respect to infringement on the input
- 1:02:35side, the wholesale copying for training purposes,
- 1:02:39what we're looking at now, because it seems like
- 1:02:42plaintiffs have, I won't say largely abandoned,
- 1:02:46but have put the issue of the output to the side
- 1:02:50and I think in tandem. the technology is developing
- 1:02:55to include more significant guardrails than there
- 1:02:58were in the beginning. I think there was that
- 1:03:00Getty case where the output of, I don't remember
- 1:03:05if it was Mid Journey or Dali, which has now
- 1:03:08been folded into Chad GPT, but the output was
- 1:03:11including Getty's watermark on a picture of a
- 1:03:13soccer player. It was in the complaint. There,
- 1:03:16I think that that's not happening anymore to
- 1:03:19a certain extent. the rationale behind treating
- 1:03:23publishers or ISPs and giving them a safe harbor
- 1:03:28under Section 230 is being mitigated because
- 1:03:34none of the copyrighted content is being reproduced
- 1:03:36on the output side. Let's see. I think many judges
- 1:03:43see AI scraping to be no different than what
- 1:03:45artists have done for centuries, but now at an
- 1:03:48unprecedented level. Yes, I think you're right.
- 1:03:53And there's a debate between the judges in the
- 1:03:56Bard's case and the Kadri case on precisely that
- 1:03:59issue, right? Like the Bard's judge is saying,
- 1:04:01look, this is no different than like school children
- 1:04:04reading all these books and an explosion of new
- 1:04:07works that are competing with the ones that they
- 1:04:10learned on. But I think that the cadre court
- 1:04:13makes a lot of good points that this is really
- 1:04:16distinguishable and the way that a model is learning
- 1:04:20is totally different than the way that the human
- 1:04:22brain is learning and that it's apples and oranges
- 1:04:26to a certain extent. I do, however, keep in mind
- 1:04:31that I think it was I think it was Picasso, but
- 1:04:35it could have been Bob Dylan. I really don't
- 1:04:37recall, but you know, you know, good artists
- 1:04:40quote and great artists steal, right? So I do
- 1:04:43think that that's true. That, you know, for centuries
- 1:04:47we've built on the innovations of others, just
- 1:04:49like I've built on the innovations of all of
- 1:04:52the, you know, contributors to the Beverly Hills
- 1:04:55bars, you know, you know, learning on this subject,
- 1:04:59right? We're all learning from each other, but.
- 1:05:01But the models are not learning like that. And
- 1:05:05then that brings me back to that kind of statistical
- 1:05:08similarity issue, right? Where the copying on
- 1:05:11that level is really the granular actual expression,
- 1:05:16right? How you're choosing to put these things
- 1:05:18together. Artists, you know... in large part
- 1:05:23are copying ideas, which are not protectable
- 1:05:26anyway under the idea expression dichotomy. What
- 1:05:30the models are doing to a certain degree is copying
- 1:05:35the actual expression, the original expression.
- 1:05:38And so I do think that it is a little bit more
- 1:05:43nuanced than maybe the judge in the Bart's case,
- 1:05:46for example, was thinking about it. And then
- 1:05:50I think this is the last question. So that law
- 1:05:55is developed so that mandates that AI technology
- 1:05:57be developed so that all copyrighted materials
- 1:06:00that are used to feed AI algorithms have unique
- 1:06:02tags attached to them or that are automatically
- 1:06:04attached by the AI technology or that. Oh, I
- 1:06:11see. You're just clarifying. So I don't understand
- 1:06:13what you mean, Miriam. I'm so sorry. I wish I
- 1:06:16did because I'm not totally following this. And
- 1:06:19I wish that I could in this moment because I
- 1:06:23would love to answer and clarify for you to the
- 1:06:26extent that I have a position on that. But I
- 1:06:32don't know that we could ever require authors
- 1:06:35to include specific content with their works.
- 1:06:39I think that that might be abhorrent to the First
- 1:06:44Amendment. But in any event, okay, Miriam, you're
- 1:06:47telling me that I already answered it. Okay,
- 1:06:49great, great, great. So perfect. Suzanne, I will
- 1:06:52leave it to you if there's any other questions
- 1:06:54or issues that we should talk about before we
- 1:06:57all part ways. All right, well it looks like
- 1:07:01this concludes today's presentation. Thank you
- 1:07:03so much Amy for your time and thank you so much
- 1:07:05to our audience for joining us today. Have a
- 1:07:08great day everybody.