Latest / CLEs You Actually Want to Hear / The Evolving Legal Landscape of Generative AI in Entertainment and How it Impacts the Copyrightability of IP
Transcript
- 0:03I am Adrian Sihand, chair of the Beverly Hills
- 0:06Bar Association's intellectual property law section.
- 0:09And on this first day of autumn, I hope you all
- 0:11had a great summer and I'm pleased to welcome
- 0:13all of you to this Beverly Hills Bar Association
- 0:16presentation of the evolving legal landscape
- 0:18of generative AI in entertainment and how it
- 0:22impacts the copyright ability of IP. And today
- 0:25we're very fortunate to have our speaker, Lisa
- 0:28Califf, and I'm delighted to introduce her. Lisa
- 0:30is a founding partner of Donaldson Califf Perez
- 0:33and represents independent producers and production
- 0:36companies in equity financing, production, and
- 0:39distribution with deep expertise in fair use,
- 0:42copyright, and personal rights issues. Her clients
- 0:45include Richard Linklater, AGC Studios, Submarine
- 0:49Deluxe, Asteria, the Safdie Brothers Elara, and
- 0:54the documentary division of Artists' Equity.
- 0:56And she has also been honored as a power lawyer
- 0:59by The Hollywood Reporter, one of America's top
- 1:02women lawyers by Forbes, and a top entertainment
- 1:04lawyer by The Daily Journal, among other accolades.
- 1:08And now, let's get started. Hi, everybody. Adrienne,
- 1:13thank you for that introduction. I always feel
- 1:16weird having people talk about me, but thank
- 1:18you very much. And I'm excited to be here today.
- 1:21I'm excited to talk about AI and copyright. Adrian
- 1:27already told you a little about me, but just
- 1:28a little bit more. The reason I'm kind of involved
- 1:32in AI being an entertainment lawyer is that part
- 1:35of our practice, a key core focus of it is our
- 1:38clearance matters. And with clearance matters,
- 1:41analyzing materials for fair use, other copyright
- 1:44issues, personal rights, trademarks. And also
- 1:48just representing producers and production companies
- 1:51holistically, AI has become part of our everyday
- 1:55practice essentially. Most of our clients are
- 1:58using AI tools to help them and assist them in
- 2:02creating content. And it's just become a part
- 2:05of our lives. So we view it as important to stay
- 2:08abreast of what's happening in this area. And
- 2:10there is oh, so much going on. So today, I really
- 2:14limited the discussion to copyright, because
- 2:18AI, I think, is impacting all facets of all of
- 2:22our lives legally and ethically and otherwise.
- 2:26But this issue of copyright I find pretty interesting,
- 2:29because I think it's really we're really looking
- 2:31at copyright from a different perspective because
- 2:32we have this idea which I'll talk about of humans
- 2:36being authors and whether or not we need to maintain
- 2:38that. So anyways, so that's a little about me,
- 2:42a little roadmap for today. As I said, we are
- 2:44going to be talking just about copyright, but
- 2:47we'll go into the definition. We'll talk about
- 2:51outputs generated by GenAI platforms. We'll discuss
- 2:55inputs and whether LLMs and other AI training
- 2:58models can train their technology using other
- 3:01people's stuff without permission. We'll discuss
- 3:04whether tech companies are going to be responsible
- 3:06for infringing outputs. We'll discuss international
- 3:11issues a little bit, hopefully touch on some
- 3:13practical tips. And it's a lot. It's a lot of
- 3:17stuff to go over. Diving in, Jose, you can go
- 3:20to the first slide or the next slide. Looking
- 3:23at the copyright code, I know probably everybody
- 3:25on this call has looked at this before, has probably
- 3:28read it many, many times. And it seems like depending
- 3:32on what's happening in society or in life, we
- 3:35focus on different aspects of this code. Right
- 3:38now, we're focusing on authorship and what that
- 3:41means. It has been very well established for
- 3:45a very long time that authorship means human
- 3:48authorship, and that the Copyright Office has
- 3:50really had a long time policy of only granting
- 3:55copyright protection to humans. And actually
- 3:58in 1997, I actually was not aware of this before
- 4:01I started preparing for this panel, this seminar.
- 4:06But in 1997, there was a case that was brought
- 4:09in the Ninth Circuit by by a religious group
- 4:13because there was a writing that was created
- 4:15in the 1920s and 30s and the adherence of this
- 4:19group had said this was created by celestial
- 4:22beings. And so in the 1990s, a woman who was
- 4:27a follower and loved the book decided to create
- 4:29the study aid and she copied the entire book.
- 4:32And her claim was, yes, I copied the entire book.
- 4:35I did it without their permission, but it was
- 4:37created by celestial beings and therefore it
- 4:39shouldn't even have any copyright protection.
- 4:41And the court looked at that and they said, you
- 4:43know what, you're right. But there was a lot
- 4:46of human involvement in this creation of this
- 4:48book. The humans actually decided what questions
- 4:51to ask to the celestial beings. The humans created
- 4:55the book. They decided what to include, what
- 4:58not to include. So it was sort of a compilation
- 5:01of this work, regardless of whether you believe
- 5:04in such celestial beings or not. The court said.
- 5:07It doesn't go to other worldly entities, but
- 5:11there was sufficient amount of human involvement
- 5:13to grant this organization copyright protection,
- 5:15and therefore the work that this follower did
- 5:19was an infringement. So this goes about 30 years.
- 5:22This is not a new concept. And then if you go
- 5:26to the next slide, please. This again, this just
- 5:31goes into this again. Basically it doesn't say
- 5:33human authorship in the copyright code, but as
- 5:37I just explained with this Urantia case, this
- 5:41goes back many, many years and probably from
- 5:44the beginning when they implemented in the Constitution
- 5:47saying this is for human authors, we want to
- 5:50promote science, we want to promote the arts,
- 5:52and by doing that we want to give human beings
- 5:55the ability to protect their creations, to incentivize
- 5:59them to create more. So this is just a little
- 6:02bit more about that. We can go to the next slide.
- 6:06So you guys probably remember this case, our
- 6:09cute monkey, Naruto, who a wildlife photographer
- 6:13left his camera unattended. This cute monkey
- 6:15picked up the camera, took some selfies. The
- 6:19wildlife photographer actually published those
- 6:21photographs. Piva decided on behalf of Naruto
- 6:25to file a claim saying, Naruto should have this
- 6:28copyright. He's the one who took the pictures.
- 6:30And the court was very quick to dismiss this,
- 6:33to say he's not a human. There's no copyright
- 6:35protection afforded to these photographs, and
- 6:37there's no infringement here. So again, way before
- 6:41AI, we're talking about human authorship and
- 6:43not granting protection to works that are not
- 6:46created by humans. Okay, we can go to the next
- 6:49slide. So here we have... this nice statement
- 6:58of policy by the U .S. Copyright Office. And
- 7:00I'm just going to read this. I think it's a nice
- 7:01statement that kind of helps set the foundation
- 7:03for this idea that we'll be talking about a lot
- 7:06today regarding AI and using AI as a tool to
- 7:11help create, to help humans create works and
- 7:14not displace humans, not do all of the work,
- 7:18not generate the work entirely by the AI machine.
- 7:21So this is authors have long used such tools
- 7:23to create their works or to recast, transform,
- 7:25or adapt their expressive authorship, EG cameras,
- 7:28Photoshop, guitar pedals, et cetera. In each
- 7:30case, what matters is the extent to which the
- 7:32human had creative control over the work's expression
- 7:34and actually form the traditional elements of
- 7:37authorship. So that existed hundreds of years
- 7:40ago. It still exists today. This idea of human
- 7:43authorship is very important. And actually, I
- 7:47was talking to somebody about copyright and someone
- 7:50was saying that when cameras were invented, you
- 7:53know, we always talk about the VHS, we talk about
- 7:55things that happened more recently when cameras
- 7:56were invented. There was a whole uproar at that
- 7:59point. People saying, well, the photographer
- 8:01shouldn't have copyright protection in a photograph.
- 8:04He's just shooting that he doesn't have to paint
- 8:06anything. This is much less work, much less human
- 8:09involvement than somebody who has to create a
- 8:12portrait. So even back then, it's kind of the
- 8:15same debate was going on. But it did become clear
- 8:18that a photographer was exerting a lot of human
- 8:20involvement, a lot of human control over what
- 8:23that output at that point was. And that still
- 8:26exists today. We can go to the next one. So our
- 8:32works created using AI copyrightable. Of course,
- 8:36it's a case by case inquiry. How could it not
- 8:38be? If it wasn't, it wouldn't be a very interesting
- 8:39topic. And the work must contain a modicum of
- 8:44human creativity. same rules that we've been
- 8:46talking about since the Constitution was enacted.
- 8:50And that modicum of human creativity is what's
- 8:53really being explored by the courts now. And
- 8:55it really depends on how the AI tool functions
- 8:58and how it was used. So I think all three of
- 9:01these points really go hand in hand. If you go
- 9:06from one end of the extreme of using an AI platform,
- 9:11typing in a prompt, Give me a script in the form
- 9:14of Quentin Tarantino. And this platform is able
- 9:17to do that. And there's no human involvement
- 9:19in what that output is. That is not going to
- 9:22have copyright protection. And then you can go
- 9:24to the other side of the other extreme. Someone
- 9:26writes an original script. Someone writes an...
- 9:30Yeah, that's perfect. Someone writes an original
- 9:32script and then maybe ask ChatGBT to write them
- 9:37a joke in the style of Woody Allen, because this
- 9:39character wants to sound like Woody Allen in
- 9:41this one scene. and it kind of sparks an idea,
- 9:45and then he ends up writing an original joke
- 9:47based on what CHAP GPT came up with. That's something
- 9:52with so much human involvement, that's going
- 9:53to have copyright protection. So again, case
- 9:55by case increase, utilizing and understanding
- 9:59there has to be some human creativity and human
- 10:02involvement in it, and then going from there.
- 10:05So we can go to the next slide now. So these
- 10:10are some examples, and I think these are all
- 10:12great examples because of course the facts are
- 10:14different, the analysis was different, but I
- 10:17do think the outcomes are consistent with one
- 10:20another even though they are different outcomes.
- 10:22So this first one, a recent entrance to Paradise,
- 10:26the one on your left, this was not granted copyright
- 10:28protection. There's spoilers on here, sorry.
- 10:31But this was not granted copyright protection,
- 10:33and that is because the claimant, when he registered,
- 10:37for copyright protection, he told the Copyright
- 10:39Office, this was created entirely by AI. He had
- 10:43created this machine, I think it was called the
- 10:45Creativity Machine. The Creativity Machine fully
- 10:51generated this image without any human involvement.
- 10:54At least that's what he said to the Copyright
- 10:55Office. And the Copyright Office said, no, I'm
- 10:58sorry. There's no human involvement, no copyright
- 11:01protection. So Thaler, the claimant, he decided
- 11:04to... petitioned for reconsideration twice with
- 11:06the Copyright Office, but he didn't say anything
- 11:09new. He just kept saying, hey guys, this was
- 11:12created by machine and should have copyright
- 11:13protection. Obviously oversimplifying, but that's
- 11:17how we do it in these hour long seminars. So
- 11:20he's like, no, it's created by machine, but we
- 11:23want to give this machine copyright protection.
- 11:25And he actually registered as a work for hire,
- 11:27basically saying the machine was the author and
- 11:29he was the owner. And the Copyright Office again
- 11:31said no, third time said no. Fowler still was
- 11:34unhappy. He brought it to the district court
- 11:36in DC. The district court in DC said the copyright
- 11:39office was correct. This is created by a machine.
- 11:41Machines don't get copyright protection. He appealed
- 11:44it. The appellate court said no. It doesn't get
- 11:47copyright protection. He went to the Supreme
- 11:48Court requesting cert. The Supreme Court said
- 11:51no. We're not going to consider this issue. He
- 11:55said it doesn't. it's very clear, the law is
- 11:58very clear that you need human authorship in
- 11:59order to get copyright protection. You were not
- 12:01granted copyright protection, so there's nothing
- 12:03for us to really consider here. So this case
- 12:06is really important because it's actually the
- 12:08first case, it's an appellate court that says
- 12:11definitively, no human involvement, no copyright
- 12:14protection. So we have this case, and that was
- 12:17back in 2019, I believe that that ruling was
- 12:20held out was... But then I think it's actually
- 12:24been going on. I think the Supreme Court cert
- 12:26request just happened this year. So he's really
- 12:30been fighting for this protection for his machine,
- 12:32and he has not been successful. So the second
- 12:36one, Zari of the Dawn, is another interesting
- 12:38one. So this author who created this, it's a
- 12:42graphic novel. I wish I had bigger pictures,
- 12:44but anyways, this is what we got. You guys can
- 12:46probably get them online. So Zari of the Dawn,
- 12:50the creator of this, registered with the Copyright
- 12:53Office. She did not disclose that she used AI
- 12:56in the generation of any of the images, but you
- 12:59guys cannot fool the Copyright Office. The government
- 13:02is watching, and they found out through social
- 13:05media that she is posting things that disclose
- 13:09she did use MidJourney. She used the platform
- 13:11MidJourney to create these images. So the Copyright
- 13:14Office sent her a letter and said, hey, we're
- 13:15not going to grant you copyright protection unless
- 13:17you explain to us why we should. The Copyright
- 13:20Office is not like, when people are not transparent.
- 13:24They weren't happy with this, but she actually
- 13:26wrote a letter back explaining exactly how it
- 13:28was created. So she stepped up, she explained
- 13:31the images were created by Mid Journey, she created
- 13:34the text, and she also selected the images and
- 13:36created the compilation. So, with this one, the
- 13:41Copyright Office looked at it, they analyzed
- 13:43it, they said, okay, great, you're gonna get
- 13:45copyright protection for the text that you created,
- 13:48for the compilation of the work. but we're not
- 13:51going to grant you protection for these images
- 13:53that were created by Mid Journey. So this is
- 13:55sort of a hybrid, and I think this is where probably
- 13:58a lot of content will land in the future. You
- 14:01know, we're dealing with films right now that
- 14:03where there's AI elements that we believe probably
- 14:06will not be granted copyright protection because
- 14:11of the extent of the AI, but we do believe that
- 14:14the work as a whole will be granted protection.
- 14:17And then obviously, the human created elements
- 14:19will also be granted protection. So this is an
- 14:22interesting one. Again, it's a hybrid, but the
- 14:24copyright office said, OK. They analyzed it carefully,
- 14:27and they gave protection to the elements that
- 14:29did have human involvement. And this last one
- 14:32on your right, a single slice of American cheese,
- 14:34is an image generated entirely through AI. But
- 14:39this was a new platform. It's a platform called
- 14:42Invoke. And the founder of this platform created
- 14:46an image, a different image. He submitted it
- 14:49to the Copyright Office in 2023. The Copyright
- 14:51Office said, no, we're not granting you protection
- 14:53because this was generated by machine. He was
- 14:56not happy with that. He created another image.
- 14:59And what he did this time is he was really transparent
- 15:02with the Copyright Office and he really made
- 15:04a plea for all of the work that he put into this.
- 15:07He said, yes, this was created by this technology
- 15:09that I created in Vogue, but there's so much
- 15:12human involvement here. And he showed how basically
- 15:16this technology, it creates images from basically
- 15:18nothing, but initially it was an image and then
- 15:21it has like a photoshopping quality to it or
- 15:25tool. where the artist goes in and he can color
- 15:30things. I think they changed the hair, he added
- 15:32the third eye, changed the color scheme. I can't
- 15:35even see the piece of American cheese really
- 15:37in this image, but there is a piece of American
- 15:39cheese on the head of the person. And they added
- 15:44the American cheese, they titled it. So the Copyright
- 15:46Office, and he actually submitted a video, which
- 15:48I thought was really interesting, although when
- 15:50I told my husband he did that. he's very involved
- 15:53in AI. He was like, who wants to do that? Nobody
- 15:55wants to do that. I was like, well, eventually
- 15:56you probably won't have to. But it was helpful
- 16:00for the copyright office to see, yes, there's
- 16:03all this human involvement here. Yes, he used
- 16:05a tool, he used an AI machine to generate this
- 16:08image, but there was so much human involvement
- 16:11and that modicum of creativity, of human creativity
- 16:14was certainly satisfied here. So this is a really
- 16:17important case, really important registration
- 16:21because it's a purely AI image that was granted
- 16:23copyright protection by the Copyright Office.
- 16:26So three different cases, very different facts.
- 16:30But I think, as I said before, I do think the
- 16:32Copyright Office was consistent in this idea
- 16:35that there just needs to be human involvement
- 16:37in the creation. And if there's not, you're not
- 16:39getting protection for it. We can go to the next
- 16:42slide. Okay, so this is kind of going to what
- 16:49we just talked about when we're advising our
- 16:52clients about how to register works created with
- 16:55AI. My takeaways are thoroughness and transparency.
- 17:01Just be honest, be transparent and be willing
- 17:04to live with what the consequences are. I mean,
- 17:06trying to not disclose certain things or argue
- 17:11that it was created by a human when it really
- 17:14doesn't serve anyone's interest. I think the
- 17:17Copyright Office is being really thoughtful about
- 17:19these registrations, but they do want the information.
- 17:23So if we're going to be using, if our clients
- 17:25are going to be using this technology, and then
- 17:27they want protection from the Copyright Office,
- 17:29we have to go through the steps to make sure
- 17:31the Copyright Office has all the information
- 17:33they need to make an accurate assessment. So
- 17:36you guys can read what these things are, but
- 17:38this also is available, I think this is available
- 17:40on the Copyright Office, and they've issued a
- 17:42lot of circulars and information. on how to do
- 17:45this and how to be transparent. We can go to
- 17:47the next slide. And this is just more, you know,
- 17:54so in this last one that we grabbed, I think
- 17:56this was from the Copyright Office, the applicants
- 17:57were unsure of how to fill out the application,
- 17:59simply provide a general statement that it contains
- 18:01AI -generated materials. I think a lot of times
- 18:04what I understand and what they say is the Copyright
- 18:06Office will actually contact you to go through
- 18:10it and find out exactly what it is. But, you
- 18:13know, it's again, I like this idea of thoroughness
- 18:16and transparency. And if we think with that and
- 18:18advise our clients with that, I think that really
- 18:22sets a good tone. And also, if we can advise
- 18:25our clients in the beginning for record keeping.
- 18:27And I'll talk about this a little bit later.
- 18:30But when we are record keeping, when they are
- 18:33putting together their content, to make sure
- 18:35that we understand how they're doing that. So
- 18:37this all kind of goes together. All right, we
- 18:41can go to the next one. Okay, so the next issue,
- 18:49so we talked about these outputs by individuals
- 18:53who are creating materials using GenAI tools,
- 18:57either to assist them or to create it entirely.
- 19:00Now kind of, I guess going backwards in my mind,
- 19:03is training LLMs with copyrighted materials a
- 19:06fair use? So I think slash hope, we all have
- 19:10sort of a general understanding at this point
- 19:12that these AI models, and here I say LLMs, which
- 19:16are large language models, basically is a type
- 19:19of AI that learns off text and how to create
- 19:22and emulate human language. So really, this question
- 19:27should be, is training any AI model with copyright
- 19:30materials a fair use? But what we have in the
- 19:32books right now is the use of LLMs, which I think
- 19:35the analysis should be the same, regardless of
- 19:37what type of platform we're looking at. We actually
- 19:42have two cases about this and the rulings are
- 19:45consistent. So we can go to the next slide. I
- 19:50believe right now there's over a hundred cases
- 19:53pending, authors and creators and other advocates
- 19:56of artists advocating on behalf of different
- 19:59groups saying Gen .AI platform should not be
- 20:02able to train their models using our stuff without
- 20:05permission. And really, the issue comes down
- 20:08to fair use. Is it a fair use? Does the ingestion
- 20:10of copyrighted work, such as books, articles,
- 20:12photographs, and art, to train an AI system render
- 20:15the system's creators liable for copyright infringement?
- 20:18Or is that ingested and defensible as a fair
- 20:20use? So we have these two cases that both came
- 20:25down last June. And they're very interesting.
- 20:28I think that they are They're obviously consistent
- 20:32with one another, which we'll talk about, but
- 20:33they do leave a lot of questions open, and it's
- 20:36not, when reading these cases, it's not a blanket,
- 20:40yes, it's a fair use, which again, spoiler alert,
- 20:43again, it was determined to be a fair use in
- 20:45these two cases, but the courts were very clear
- 20:48and very, I think, and they believed it very
- 20:53important to lay out that it doesn't always mean
- 20:55that copying materials is gonna be a fair use.
- 20:57We can go to the next one. So believe it or not,
- 21:02we're using the same four factors, same four
- 21:05fair use factors that we've been using since
- 21:06literally the 1800s. They still work, which is
- 21:10amazing. And the courts, when they looked at
- 21:14these cases to determine whether the use of other
- 21:18people's materials to train these AI models is
- 21:21a fair use, they looked at the same four factors
- 21:23we've been looking at since the 1800s. You can
- 21:27go to the next slide. So the two cases, we'll
- 21:32talk about these a little bit. So the first one
- 21:35is Barts versus Anthropic. In this case, there
- 21:39was a copying of, I believe, millions of books
- 21:42to train these LLM models for Anthropic's Claude
- 21:46AI. A group of authors came forward and they
- 21:51said, hey, you can't use my stuff without permission.
- 21:54You're basically displacing us. You're training
- 21:56this technology to displace human authorship.
- 22:01They initiated this legal action, obviously alleging
- 22:04copyright infringement, saying that they unlawfully
- 22:07copied their books. And Anthropic came back and
- 22:10said, yes, we did unlawfully copy the books,
- 22:13but it's a fair use. The court actually held
- 22:16on summary judgment that the copying of the books
- 22:18to train the LLMs was a fair use. And I think
- 22:23the main factor that the courts Considered and
- 22:28what they looked at was the transformative factor.
- 22:30So the purpose and character of the use and what
- 22:33they determined is that it was exceedingly transformative.
- 22:36There's a lot of hyper hyperbole or excessive
- 22:38words. It was exceedingly transformative because
- 22:42it was a mean to generate and develop a generative
- 22:45AI model. It was not akin to. reading the book
- 22:49for entertainment. And they actually kind of
- 22:52compared it to a reader who internalizes texts
- 22:54in order to write something new. This just happens
- 22:57to be a machine that's doing it. And we're not,
- 23:00they're not replicating it. They're not supplanting
- 23:02the original work. And they believe that the
- 23:07way in which those books were used was very transformative.
- 23:12And I tend to agree with them. And I think a
- 23:14big part of that is the outputs, which I'll get
- 23:17to in just a second. But there's another issue
- 23:20in this case. And I think some people confuse
- 23:22it. But there was an issue with pirated books,
- 23:26so books that were obtained illegally and also
- 23:29books that were purchased through proper means.
- 23:32The court did not draw a distinction between
- 23:35those two in terms of training the LLM model.
- 23:38So the court said, regardless of whether the
- 23:41books were obtained illegally or obtained through
- 23:44proper means, it was still a fair use. It made
- 23:47me look down upon the pirated books, but they
- 23:50said it doesn't impact the fair use analysis.
- 23:52But there was another issue in this case about
- 23:55a library where the Anthropic also created a
- 23:58library. So sort of a permanent digital library
- 24:00of all of these books. And what the court held
- 24:03in that case, they said, you can't steal books
- 24:07and then create a library. for those books and
- 24:10keep them permanently. That we view as an infringement.
- 24:14And so that part, the library element of the
- 24:17pirated books and the lot being stored as a digital
- 24:20library, that was determined to be a copyright
- 24:23infringement. The training of the LLM model,
- 24:27regardless of whether the books were obtained
- 24:29legally or illegally, was deemed to be a fair
- 24:32use. And that's where the Parties went on to
- 24:36settle. So there was this issue of this pirated
- 24:38library. The court said, this is not a fair use.
- 24:42It would have gone further on in the litigation.
- 24:45The court settled, I mean, sorry, the parties
- 24:47settled for an unprecedented $1 .5 billion, which
- 24:50is the biggest copyright settlement that's ever
- 24:52happened, to reimburse or to pay those authors
- 24:58from whom they took the books without permission.
- 25:02And also that settlement did not allow them to
- 25:04keep it. They actually had to destroy all those
- 25:06digital copies. And in order to keep that library,
- 25:10they would have to purchase actual copies legally.
- 25:15So anyway, it's an important distinction. I think
- 25:17a lot of people kind of confuse those two elements,
- 25:19but there are two very distinct elements of that
- 25:22case. One, the training, and then two, the library.
- 25:25So make sure you keep those ideas separate. And
- 25:30as I said, like I said, they used a lot of strong
- 25:33words. I think when they were talking about training
- 25:35the LLM models on copyrighted books, there was
- 25:37one time the court said it was spectacularly
- 25:40transformative. They weren't reproduced or distributed.
- 25:43They were abstracted into statistical representations
- 25:47for generating new and different texts. Basically
- 25:50said that authors can't exclude others from using
- 25:53their works to learn. So they really kind of
- 25:55treated this AI model the same as what a human
- 25:58who ingested a lot of material and information
- 26:01to create something new. And then they also said
- 26:05the digitizing of the purchase books was a fair
- 26:07use. And then as I just discussed, the digitizing
- 26:09and retaining the pirated copy was an infringement.
- 26:13And what they talked about there, just to go
- 26:15back to that for a second, is that they displaced
- 26:17the demand for copy -to -copy purchases. So creating
- 26:21this permanent general library was not a fair
- 26:24use, excusing that piracy. So interestingly,
- 26:28and I think it's not that interesting because
- 26:29the case was about the inputs, but they did not
- 26:32address outputs. And the court was clear to say,
- 26:35and I'll talk about this more with the second
- 26:37case, Kadri, um but the court was clear to say
- 26:40that the outcome may have been different if the
- 26:42outputs were infringing and so again another
- 26:45important element to keep in mind is what do
- 26:47these outputs look like so in the in ken we'll
- 26:51talk about that here with the cadre case so this
- 26:53cadre versus meta very similar case authors brought
- 26:57a claim against meta they were trained they were
- 27:00creating this llama model and the court again
- 27:03ruled again this is the ninth circuit i think
- 27:05one was decided on June 23rd, one on June 25th.
- 27:09So glad they're very consistent holdings, two
- 27:12different judges. But in this meta case, the
- 27:16judge actually focused a lot on the guardrails
- 27:20that were implemented. So he talked about various,
- 27:23talked about a lot of the same, the reasoning
- 27:24was very similar, but he also talked about the
- 27:28guardrails that were implemented in this llama
- 27:30platform and really emphasized was a big reason
- 27:34why it was transformative. So when Meta was creating
- 27:38this Llama platform, they put in place a lot
- 27:41of guardrails to ensure that the outputs were
- 27:43not infringing. So among those were guardrails
- 27:48to make sure that you couldn't copy a certain
- 27:50amount of text, and they used the word 50 tokens.
- 27:52You can't copy more than 50 tokens of text. It
- 27:56also would refuse to do things that would create
- 27:59an infringing output, like write me a book in
- 28:02the style of this. So they spend a lot of time
- 28:05talking about that. And there was also some information
- 28:08about how these not only were these, did they
- 28:11say they put these measures in place, but they
- 28:13actually tested them. So they're experts and
- 28:16even the plaintiff and they tried to do things
- 28:19that Mehta was saying, you can't, you're not
- 28:21going to be able to do that. and they weren't
- 28:24able to. So they weren't able to create these
- 28:27infringing outputs. And also, one of the other
- 28:29guardrails was about memorizing. So the tool
- 28:32couldn't memorize text or a certain amount of
- 28:35text. And again, I think that's really important.
- 28:39The Barts case talked about if the outputs were
- 28:42infringing, this outcome may have been different.
- 28:45Cat, Meta, Medicaid, same thing, where they really
- 28:48put an emphasis on these guardrails that were
- 28:51put in place. So let people know it doesn't just
- 28:55mean these holdings don't mean you can just blatantly
- 28:58copy things and not have any responsibility for
- 29:02it. There may be circumstances in which it's
- 29:05infringing. I can't remember which case it was,
- 29:07but one of them in the court even said something
- 29:09like, we're not really saying that it's OK to
- 29:12do this. We're kind of saying that the plaintiffs
- 29:14didn't come with the right claims, which to me
- 29:16meant we really need to look at whether or not
- 29:20the outputs are infringing, because then that
- 29:22really gets into that market harm, where I think
- 29:26the court could really maybe take an action against
- 29:28that company to say, hey, you're really creating
- 29:31market harm. You're displacing human creation,
- 29:34and this is not acceptable. But so far, they
- 29:36have not yet been able to do that. OK, can we
- 29:41go to the next slide? All right, so looking at
- 29:45these AI considerations, so this is, wait, I
- 29:48just did something weird on my computer. Sorry,
- 29:52one second. Okay, I'm back. You can see my screen
- 29:56now. Okay, so I think that these are good to
- 30:00look at because, again, these are just the four
- 30:02standard fair use factors. We talked about these.
- 30:07I talked about these when I was discussing the
- 30:09facts of the case, but the purpose and character
- 30:12restrictions on the output. We discussed that,
- 30:14especially in the meta case, that there was restrictions
- 30:16on the output, therefore making the the training
- 30:20of the LLM models even more transformative. Is
- 30:24the purpose to generate outputs and summarize
- 30:26copyrighted works? I think we can all, hopefully
- 30:29everybody on this call, on this Zoom knows that
- 30:32that would be a no -no. We're not supposed to
- 30:34generate outputs that summarize copyrighted works.
- 30:37Whether the actor is seeking a profit from the
- 30:39use of the copyrighted materials that train the
- 30:41models. And this is important because it's using
- 30:43those materials, profiting off of those materials
- 30:47rather than profiting off of something new. and
- 30:50then access knowing use of a database that consists
- 30:55of pirated or illegally accessed works should
- 30:57weigh against fair use without being determinative.
- 30:59This might be too strong of a statement. I don't
- 31:01know if it weighs against fair use. I don't think
- 31:03it was, they weren't happy about it, but I think
- 31:09that when we look at the fair use analysis, we
- 31:11have to look at just the fair use analysis of
- 31:13piracy is a different issue. Let me go to the
- 31:15next slide. I think that was the wrong way. Yeah,
- 31:21there we go. So in the nature of the work, I
- 31:23think this factor is probably always going to
- 31:26be in favor of the plaintiff. But as we all know,
- 31:30this factor is of very little importance. So
- 31:32they're usually going to be using creative works.
- 31:36We can go to the next slide. So the amount used,
- 31:41this again, will probably always go in favor
- 31:43of the plaintiffs. But AI developers are ordinarily
- 31:46going to copy the entire works. Again, If the
- 31:49purpose is transformative and there's a need
- 31:52to train on a large volume of works, then copying
- 31:55the entire work is probably going to be OK. In
- 31:58the Bard's case, they said there was millions
- 32:00of books. So even if you're copying the whole
- 32:02thing, you're using millions of books and the
- 32:04outputs are quite different. The use is very
- 32:07transformative. And therefore, this factor isn't
- 32:09going to weigh that heavily for the defendants.
- 32:13Next slide, please. And then the market effects.
- 32:16So this, I believe, is going to be the most important
- 32:19factor, especially given what the court said
- 32:22in the MediCase and in the Anthropic case. We
- 32:25really need to look at the outputs and what those
- 32:27are going to be because the courts may find a
- 32:29way to say, hey, this isn't so transformative
- 32:31if the outcome is a work that's substantially
- 32:34similar to what the inputs are. We can go to
- 32:38the next slide. Okay, so now the next topic.
- 32:43So I think we have these two cases which have
- 32:45been really helpful to really set the kind of
- 32:49the groundwork to say, all right, these works,
- 32:52training LLM models on copyrighted materials
- 32:55is a fair use, so long as you have these guardrails
- 32:58in place. And so long as, which we don't know
- 33:01yet, the outputs are not infringing. We don't
- 33:04know the answer to that question yet. We'll see.
- 33:06There's been no holdings, but there are several
- 33:08cases and three of which I want to talk about
- 33:10today where the studios are going after the AI
- 33:14companies for potentially infringing outputs.
- 33:18So this is going to be another very interesting
- 33:20area to watch and see what happens there. In
- 33:25this first case, Disney versus Mid Journey. So
- 33:27this is Disney and a lot of other studios that
- 33:31have filed a lawsuit against Mid Journey. And
- 33:34it's really about their video technology, their
- 33:38images and video technology. Allegedly, the Mid
- 33:42Journey technology can create works that are
- 33:45substantially similar to Disney and other studios
- 33:48owned IP. And the words that the plaintiffs are
- 33:54using is a virtual vending machine generating
- 33:57endless unauthorized copies of their copyrighted
- 33:59works. And in the complaint, there's actually
- 34:02side -by -side comparisons of some of the Disney
- 34:05and Universal characters alongside the AI characters,
- 34:11which are replicated by Mid Journey, and they're
- 34:13very, very, very similar. The companies are seeking
- 34:17incentive relief, saying mid -journey, you guys
- 34:19can't do this. You're infringing on copyright.
- 34:21Mid -journey's claiming fair use. And they're
- 34:24also claiming that they're kind of separating
- 34:28themselves. So kind of this vicarious or secondary
- 34:30liability. They're saying, we're not liable for
- 34:32what the users are doing. That's their responsibility.
- 34:35What we're doing with training our models is
- 34:38a fair use. So it's going to be interesting to
- 34:41see what happens here. My gut? is that if the
- 34:46technology truly can output materials that are
- 34:50substantially similar to other people's copyrighted
- 34:55images, that the courts will find a way to determine
- 34:58that's not a fair use. Because I think also if
- 35:00you look at the policy and the reasoning behind
- 35:04copyright and to generate interest and incentive
- 35:08to create new works, that this does not do that.
- 35:12So if we have AI platforms that can generate
- 35:16other people's works and there's no liability
- 35:18attached to that, I do think that that will be
- 35:21a point in which the courts are going to step
- 35:23in and say, hey, this isn't cool. You can't do
- 35:27this. This is not serving the purpose of copyright,
- 35:31not serving the purpose of fair use. And we're
- 35:33not going to allow you to continue creating these
- 35:36types of platforms that produce infringing works.
- 35:40And whether that liability lies on the actual
- 35:44tech companies or whether it lies on the users
- 35:46will be interesting to see. Hopefully they'll
- 35:48put some sort of onus on the tech companies to
- 35:51implement some measures and some guardrails to
- 35:53ensure this isn't happening. But again, we'll
- 35:57see. My guess is as good as yours. And then the
- 36:03next case on here, this Disney versus Minimax.
- 36:06This one's very similar, but Minimax is actually
- 36:08a Chinese company. So there's actually issues
- 36:10with jurisdiction that go way beyond what I know
- 36:13and what I understand, but as not a litigator.
- 36:16But that will be interesting to watch as well.
- 36:19It's very similar, but Minimax, I think Minimax
- 36:22actually, I'm gonna read this. It says they market
- 36:25themselves as a Hollywood studio in your pocket.
- 36:27So it allows... the users who have to pay to
- 36:30use the platform to generate AI images and short
- 36:33videos based on text prompts. So we have to watch
- 36:38that and see. And both are very early on in the
- 36:40litigation. I don't know how long we're going
- 36:42to be waiting, but hopefully we'll get some guidance
- 36:45from the courts on those issues soon. And then
- 36:48this third case that I have here is Sony and
- 36:51Warner Chapel versus Anthropic. So these are
- 36:53music cases. There's been a growing number of
- 36:57these cases that publishers are bringing against
- 37:00these AI platforms, alleging copyright infringement
- 37:03for their musical compositions. A lot of it has
- 37:05to do with lyrics. They're basically saying these
- 37:07AI platforms are generating lyrics without permission.
- 37:11There's been, there was a case filed in 2023
- 37:14and then. by a bunch of music publishing groups
- 37:17and another one I think in 2025 and then maybe
- 37:20it was early 2026 and then this case being filed
- 37:23now um Sony Warner Chappell so with this one
- 37:27I guess all of the publishing arms of all the
- 37:30major companies are now litigating against Anthropic
- 37:32and again we don't know what's going to happen
- 37:35here but the complaint is basically I think this
- 37:39one is identifying tens of thousands of compositions
- 37:43allegedly being infringed on through this technology.
- 37:48And then they're also saying they're also arguing
- 37:49with the outputs that they're able to output
- 37:52compositions or lyrics that are substantially
- 37:56similar or the same as what their compositions
- 37:59are. So much different than the first two cases
- 38:04where the outputs were substantially different.
- 38:06Here we're looking at things where the outputs
- 38:07are substantially the same and could create that
- 38:10market harm that I think the court's looking
- 38:12for in order to determine the training of the
- 38:15model is not a fair use. It's so confusing. Hopefully
- 38:20I'm keeping it pretty straight for you guys.
- 38:22So those are those three cases. There's obviously,
- 38:25I think there's really a lot of others that are
- 38:27being filed. There's so much litigation happening.
- 38:30in this area and set it at probably a thousand
- 38:34times, but it will be interesting to see how
- 38:35all of this plays out and what happens here.
- 38:38So let's look briefly at international law. We
- 38:44can go to the next slide. So internationally,
- 38:48I think that most, I mean, I haven't looked at
- 38:51every single country, but I've looked at the
- 38:53EU, I've looked at the UK, I've looked at China.
- 38:56And we're all pretty much on the same page with
- 38:59regard to human authorship, except for the UK,
- 39:01which I'll talk about in just a second. But even
- 39:06in the EU and China, there's no specific laws
- 39:08yet that address copyrightability of AI -generated
- 39:12works. We're all still kind of figuring it out.
- 39:16But most countries, just like in the US, they
- 39:20require human authorship. And there needs to
- 39:23be some sort of, the EU says, human intellectual
- 39:26contribution to be protectable. So we're all
- 39:30figuring it out. We're all pretty much on the
- 39:32same page except for the UK. We can go to the
- 39:34next slide. And the UK is an interesting position
- 39:38because back in 1988, They passed a law for computer
- 39:43generated work saying this, and this is not a
- 39:46typo, that in the UK copyright protection is
- 39:49available for any artistic work which is generated
- 39:51by a computer in circumstances such that there
- 39:54is no human author. So they created this rule
- 39:57back in 88, which obviously applies now. But
- 40:02as I wrote here, parliament is kind of trying
- 40:07to change that or they're at least recognizing
- 40:10that they're in this weird sort of conundrum
- 40:13and their policy is kind of is going against
- 40:15what pretty much the rest of the world is saying
- 40:18with regard to human authorship when they're
- 40:20granting very specific protection to machine
- 40:22created works so another interesting area to
- 40:27see how it develops. But right now the UK and
- 40:29sort of the rest of the world are in conflict
- 40:31because of this 1988 law that was, and I don't
- 40:34know what the purpose, it would actually be interesting
- 40:36to look into, but why they passed this law. There
- 40:39must have been some, there must have been some
- 40:41reason. And now they're dealing with it. So we
- 40:49can go to the next slide, sorry. So in China,
- 40:54China is really interesting because they have
- 40:55a lot of case law going on, a lot of registrations,
- 40:58a lot of activity in this AI world. And so I
- 41:03thought it'd be cool to spend a couple of minutes
- 41:05on what's happening in China. And here, I think
- 41:10that the way this is said, Chinese judicial decisions
- 41:12on the threshold of human authorship required
- 41:14for copyright protection have varied. And so
- 41:17they're struggling with the same thing we're
- 41:18struggling with, which is sort of how much human
- 41:21involvement is necessary to, um, to constitute
- 41:25a work that is entitled to copyright protection.
- 41:29So we can go to the next, they don't, we can
- 41:32go to the next slide. Thank you. We don't, they
- 41:34don't have a yet a rule in place. There's no
- 41:36established sort of national law about what's
- 41:40protectable and what's not. Um, but their practical
- 41:43position basically emphasizes the same things
- 41:46we do, which is the human must have some involvement
- 41:51when they talk about prompting, parameter adjustment,
- 41:53selections, modifications, other things like
- 41:56that to demonstrate the humans involved in the
- 41:58creative choices and how this work was made.
- 42:01So here's some examples from China. So this first
- 42:05case on the left, spring breeze brought tenderness.
- 42:09So this was determined to have enough human involvement.
- 42:13Believe this was created using mid journey. Oh,
- 42:15this was used stable diffusion. But the plaintiff,
- 42:21when they submitted the application, they basically
- 42:25selected from over 150 prompts. They had a lot
- 42:30of human involvement and creativity. So when
- 42:35the court found there was sufficient intellectual
- 42:38achievements reflecting human creativity. So
- 42:43they really showed, and again, the claimant had
- 42:45to demonstrate this. that there was, they called
- 42:50it intellectual labor and personalized human
- 42:54expression. So there was work that a human actually
- 42:58did use AI as a tool, but then use a sufficient
- 43:02amount of human creativity, involvement, control,
- 43:05in order to get copyright protection for the
- 43:07image. So this next one wings, and I guess with
- 43:10shortages, all of these are different courts
- 43:12from around. around China, they're not in one
- 43:15jurisdiction. So the jurisdictions are kind of
- 43:18looking at these differently, but again, I think
- 43:20differently, but consistent. So, but it looks
- 43:24like in the second one, the court there set the
- 43:27threshold a bit higher for demonstrating human
- 43:29authorship. Here, the plaintiff did create a
- 43:33lot of different AI -generated images. Here,
- 43:35they use mid -journey. But what the court said
- 43:39is that the plaintiff failed to provide enough
- 43:41documentation to prove her sufficient human creative
- 43:45inputs in the creation process. So the other
- 43:49thing that I think is important to keep in mind
- 43:53is one thing the plaintiff said here, which they
- 43:55viewed as an admission, is that she couldn't
- 43:58regenerate the exact same AI picture. So even
- 44:01if she put in the same prompts, due to the randomness
- 44:05of mid -journey and how it creates outputs, She
- 44:08couldn't create the same thing. And I think that's
- 44:10important because it shows that the human, here
- 44:13it showed that the human didn't have control
- 44:15over that output, that it really depended on
- 44:18what the machine did, and the machine had that
- 44:20ultimate control of what the output was. So here,
- 44:24no copyright protection. And then in that third
- 44:27work, the companion heart, which I think is really
- 44:30cool, there again, there was, they did get copyright
- 44:35protection here because here, the plaintiff did
- 44:37show, this was using mid journey again, and the
- 44:39plaintiff did show that she used enough, there
- 44:43was enough human involvement to grant, to be
- 44:45granted copyright protection. They talked about
- 44:48there was a lot of details. She used many, many
- 44:51prompts. There was multiple red love balloons.
- 44:54And then I think also she used Photoshop at the
- 44:57end to finish and complete the photo and get
- 45:01to this completed work. So, so really at the
- 45:05end, it was the human who had the control over
- 45:07what the final output looked like and not the
- 45:10machine. And the court found it was protected
- 45:12by copyright law. Okay, let's go to the next
- 45:15slide. I feel like we've covered so much. It's
- 45:20like been a blur, but hopefully there's been
- 45:24some interesting stuff that's come out of it.
- 45:25But I think looking at all of this information,
- 45:29looking at outputs, looking at inputs, looking
- 45:32at people who are really creating images using
- 45:34gen AI, It's all very interesting. I think that
- 45:39the way in which our clients and most people
- 45:43are currently using AI is quite different. These
- 45:46feel like test cases. These feel like outliers
- 45:49in a sense to me. I could be wrong. It could
- 45:52just be because that's my world. But I think
- 45:56people are really using AI in the same ways they've
- 45:59used other tools to help them create content.
- 46:02And again, this is certainly in my world. But
- 46:04here, there's a list of tag tapes, tag tapes,
- 46:08sort footage, assemble rough cuts, rotoscoping,
- 46:12cleaning up ambient noises, multi -language lip
- 46:16syncing, subtitling, video restoration, things
- 46:18that are very technical in nature, maybe things
- 46:21that help enhance the product at the end, clean
- 46:24up, finalize, make it look a little bit prettier.
- 46:27But they're not using AI to replace humans, to
- 46:32at least replace humans in terms of the creative
- 46:35content. And certainly there's a whole nother
- 46:37conversation about the ethics of it. There's
- 46:39a whole nother conversation about it displacing
- 46:41jobs, especially lower level jobs. But really,
- 46:45to me, the uses that people are engaging them
- 46:48on a day -to -day basis are the same types of
- 46:51uses that people have been engaging in new technologies
- 46:53since the beginning of creating content. Really,
- 46:58I see these cases are all very interesting, but
- 47:00then it's really like, well, how is this really
- 47:02impacting our clients and people who are not
- 47:04sort of the cutting edge of AI? And I think it
- 47:08really impacts us in a much more basic and fundamental
- 47:11way. And in some ways that aren't even really
- 47:14that don't change the analysis of what we're
- 47:18doing very much. So you can go to the next slide.
- 47:27So this is, again, just some other practical
- 47:30guidelines about how we're advising our clients
- 47:33when they're using AI tools. The first thing
- 47:35we always tell them is use a model trained on
- 47:38licensed materials. There are those two holdings
- 47:40that have held that training LLM models using
- 47:46materials as a fair use, but we still think it's
- 47:49best practices to use an AI model that's trained
- 47:52on licensed materials. And there are a handful
- 47:54of them out there. I think Adobe is one of them.
- 47:58Getty is another one. There's a handful of AI
- 48:00platforms that our clients can use that are quote
- 48:03unquote safe. They're clean and we advise that
- 48:06they use those. And then the other thing we advise,
- 48:09I'm actually working with a client right now
- 48:11on a film where they're utilizing an extensive
- 48:13amount of AI and they're using it in a creative
- 48:16way where they're... the film, it's a documentary
- 48:19film about an artist from the 60s. And in that
- 48:22they're sort of they're animating a lot of the
- 48:24artwork. They're creating some recreations of
- 48:27what may have happened, what a street corner
- 48:29may have looked like. But they are using AI to
- 48:33a pretty large extent in that film. And they
- 48:36want to make sure they're doing everything right.
- 48:37So we're starting with them at the beginning,
- 48:39making sure that everything really is tracked
- 48:42appropriately. So what platform are they using?
- 48:45What prompts are they using? to generate the
- 48:48outputs. Are they uploading images to the platform?
- 48:52Are those images licensed? You're just really
- 48:55trying to get all of the information we can so
- 48:57that we can assess at the end of the day whether
- 49:02or not it's OK or not, whether or not there's
- 49:03a legal issue there, or whether you've done everything
- 49:06correctly and we don't think there is a legal
- 49:07issue. And then I have these two resources at
- 49:11the end. Follow the APA Gen. AI guidelines. The
- 49:16Archival Producers Alliance has sort of this
- 49:18whole gen .ai platform. They have guidelines
- 49:22on how to use it. They're mostly, I'd say, ethical
- 49:25guidelines, although I think they also comport
- 49:28with all the legal guidelines we have. They also
- 49:30have a toolkit. And then Creators Coalition on
- 49:33AI is a really interesting organization. I'm
- 49:35actually serving as a legal advisor on their
- 49:38legal advisory council. And the thing they're
- 49:41focusing on right now is creating a, it's really
- 49:44a, it's a group of creators who want to have
- 49:47a say in what these AI conversations look like.
- 49:50So they don't want the tech to somebody's just
- 49:53deciding for all of the creators who are being
- 49:56impacted by this technology. They don't want
- 49:59the tech companies to be deciding everything.
- 50:00So we're basically coming forward saying we're
- 50:02a group of creators. Let's create definitions
- 50:05that we can all agree upon. for all of the terms
- 50:08that relate to and affect AI. And then they're
- 50:12also doing these impact surveys and reports.
- 50:15So really talking to people, say, how do you
- 50:18use AI? How is AI impacting your job? Are you
- 50:23using AI in a way where you don't have to hire
- 50:25as many people? Are you using AI in a way where
- 50:27it's helping you do your job more efficiently?
- 50:32Are you? what does all this look like? And so
- 50:35they're trying to get a really broad scope of
- 50:38what it looks, what it actually looks like in
- 50:40real life, people using AI on films and television
- 50:44productions primarily. So that's a cool organization.
- 50:47They call themselves an agnostic AI company where
- 50:50they're just trying to kind of settle more level
- 50:52playing field. And so that everyone's sort of
- 50:55operating from the same playbook. That's that.
- 50:59I know I have a bunch of questions here in the
- 51:01chat, which I will start going through and I'll
- 51:03answer the ones I can. If I don't answer it,
- 51:05it probably means I don't know the answer to
- 51:06it. So, Jack, I think you have the first question
- 51:10here is insurance covering defense. So, so far,
- 51:14we're not litigators. We have been successful
- 51:17in securing E &O coverage for AI -generated works
- 51:21or works that are generated using AI as a tool.
- 51:24We have not had any problem of getting rejected
- 51:26for that. And we've been transparent. So I would
- 51:31imagine insurance would cover it, but I haven't
- 51:35been there yet. So I don't have a definitive
- 51:37answer, but we have been successful in securing
- 51:40insurance, at least at the front end, for these
- 51:43productions that are utilizing AI. And I guess
- 51:46I answered your second question, too. Any coverage
- 51:48issues to date? Not for us, not on Wood. If One
- 51:53Nation grants copyright to AI Works, how do treaties
- 52:00Peter, I do not know the answer to this question.
- 52:03I'm going to be brutally honest. I don't know.
- 52:06You know, this international issue I think is
- 52:07really interesting. I imagine this is my guess,
- 52:11and I'm not really qualified to guess, but that
- 52:13it would just be the same as any other copyright
- 52:16law. You know, that it's probably the country
- 52:17of origin that would, their laws would prevail.
- 52:24I don't know. And I don't know if anybody knows
- 52:25it's so early in this debate. And I don't think
- 52:28these issues have come up yet. OK, Doris, you
- 52:33ask, when the Copyright Office issues a registration
- 52:36for a work, will it identify the AI elements
- 52:39in the work excluded from protection? Yes. So
- 52:42that's the importance of being thorough in your
- 52:46application. So like in that Zarya of the Dawn,
- 52:50example, it specifically excluded the images,
- 52:53which were all of the images in there that were
- 52:55created by Mid Journey. But it gave copyright
- 52:57protection to the compilation of the poll and
- 52:59to the text. So it is important to do that. Again,
- 53:05sometimes you may not know exactly, or it may
- 53:07not be that detailed in your application. They
- 53:10may call you and say, what parts of this content
- 53:14are generated by AI? And to me, I really view
- 53:17this very similarly to using any other third
- 53:20-party material. If a client incorporates archival
- 53:24material that they license and don't own, obviously,
- 53:27they're not going to own the copyright to that.
- 53:28If they put it in music, they don't own the copyright
- 53:31to that. So here, it's just sort of another element
- 53:33of the compilation that you might not have complete
- 53:36ownership over. But it should not, if you're
- 53:39using it in this correct way, it should not impact
- 53:42the copyrightability of the entire work. All
- 53:47right, Christopher, do you expect the same test
- 53:50for copyright ability to apply in the case of
- 53:52Vive coding and AI assisted or created computer
- 53:55code? Can I just say, I don't know. I don't know.
- 54:01I would guess yes. These questions are, they
- 54:05just haven't been tested yet. And I expect that
- 54:08they will. continue to use these same tests,
- 54:12like, especially if you think about the fair
- 54:13use and the four factors and how they've been
- 54:14around since the 1800s, like, I just think that
- 54:17these tests are kind of tried and true and they
- 54:19are inclusive of new technologies. I think that
- 54:25lawmakers have always thought about the fact
- 54:27that there'll be new technologies are going to
- 54:29be things we don't know that aren't in existence
- 54:32now that we need to address. And so crafted things
- 54:34broadly to address that. So my guess is yes,
- 54:38but. I don't know. Jonathan, does the possible
- 54:44gap in ownership that GenAI media can create
- 54:46cause a problem at a specific point like E &O
- 54:50or at delivery or in the reps and warranties
- 54:52in the distribution deal? Can the distributor
- 54:53treat that as a breach and start holding money
- 54:56back? Or is this mainly a problem when it comes
- 54:58to protecting IP for the future? I do think that
- 55:02it can become a problem early on, because now,
- 55:06we're probably all seeing, regardless of what
- 55:08area of law we're practicing in, we're all seeing
- 55:10AI provisions pop up in contracts. And we even
- 55:14have an AI provision in our retainer letter.
- 55:16So it's popping up everywhere. So I think, again,
- 55:19it is really important to read those, make sure
- 55:22we understand them. And this is where the Creators
- 55:25Coalition of AI comes in handy, where they're
- 55:28going to have this glossary of definitions of
- 55:30what do these things actually mean. and address
- 55:34it early. And if it is something, if somebody
- 55:37says, because I have seen provisions where a
- 55:40distributor or a financier has said things like,
- 55:44you're rubbing a warranting that no AI is being
- 55:46used. And it's like, absolutely not. We all use
- 55:49AI every day in our lives. And so to the extent
- 55:52a filmmaker doesn't use AI would be very rare.
- 55:56I would imagine, especially when we're talking
- 55:58about post material, post equipment and that
- 56:01type of stuff, it's all utilizing AI. So I do
- 56:04think that it can be an issue early on, and we
- 56:06should all be aware of it and address it as early
- 56:09as we can. How will 9th Circuit filtration work?
- 56:14Jack, I wish I could ask you what this even means.
- 56:17I'm so embarrassed. I don't know exactly what
- 56:20that means, probably because I'm not a litigator.
- 56:23But maybe you can let me know on the side. And
- 56:27then, Margaret, you asked, aside from the use
- 56:29of AI by creators for standalone works, with
- 56:31respect to works created with some AI elements
- 56:34that are intended for license and exploitation
- 56:36of music, film, TV, what do you think the production
- 56:39companies should consider in using AI -generated
- 56:42assisted materials, whether permission for the
- 56:44project or license from a third party? Somehow
- 56:49they're intended for license or exploitation.
- 56:51I think I've sort of addressed this, Margaret.
- 56:53Hopefully. I mean, I do think that, again, it
- 56:55goes back to my saying of thoroughness and transparency,
- 57:00that, you know, we do have to analyze these things
- 57:04on a case -by -case basis and look at how it
- 57:08was generated, what it was, what the inputs are,
- 57:11what the outputs are, what models we're using,
- 57:13and really view it ballistically. And us as a
- 57:16lawyer is like, for us, representing independent...
- 57:19Content creators. We have to be able to support
- 57:21that use like our goal at this early stage is
- 57:24to say Is to be able to answer questions from
- 57:28insurer or distributor say we'll use AI. Why
- 57:30is it? Okay, we want to have an answer to that
- 57:33question. So we need to do that work on the front
- 57:35end Make sure that we thoroughly understand how
- 57:39this content is being created What AI provision
- 57:44do you have in your inner? In our retainer letter,
- 57:45we have a provision that says we utilize AI.
- 57:47We actually started using Harvey, who's a lot
- 57:50of people's best friend in our office. And it's
- 57:53a legal AI platform that helps with legal research.
- 57:57It does chain of title analysis. It can compare
- 58:00agreements, does all sorts of things. And so
- 58:03we will disclose that to our clients that we
- 58:06do use that technology. And then, how would you
- 58:12advise a video game developer that uses AI in
- 58:16development on copyright registration? I imagine
- 58:18a developer might want to avoid identifying AI
- 58:20elements in their copyright registration completely.
- 58:22I mean, listen, I would not avoid it. I would
- 58:25tell them they have to put it in there. I do
- 58:27think that some people are registering works
- 58:31where they use AI and not disclosing it. And
- 58:33I think if you're using it in a really minimal
- 58:35way, that's probably fine. to the extent you're
- 58:39using it significantly, I do think it needs to
- 58:42be disclosed. And I also think, you know, it
- 58:44really goes back to kind of those of you, we
- 58:47fill out, we help our clients fill out, you know,
- 58:49insurance applications all the time. And to those
- 58:51of you guys who fill out applications like that
- 58:53or help your clients fill out applications, it's
- 58:56so much the attention is in the wording, you
- 58:59know, so I would look at the registration, the
- 59:01Copyright Office, and I would disclose what I
- 59:02have to disclose. I don't think that means you
- 59:05have to over disclose things. I don't think it
- 59:07means you have to put your client in a precarious
- 59:09position when they otherwise wouldn't be. I think
- 59:11it's really important to be honest and to answer
- 59:15those questions accurately. So I would look at
- 59:18the application closely, and I would disclose
- 59:20what it requires to disclose, and I wouldn't
- 59:22try to get too cute about it. I have a great
- 59:27question from Alex. He says today, are you an
- 59:30AI avatar or the real Lisa Callow? I am an AI
- 59:34avatar. Sorry, this is all AI. I'm not a real
- 59:38person. Just kidding. I'm real. OK, my understanding
- 59:43of LLMs is that it produces its responses due
- 59:46in part to the memories it has of the user and
- 59:48based off of what the LLM predicts the user would
- 59:51like to want to hear. Essentially, there is some
- 59:53level of human coordination with the LLM. Is
- 59:55that enough human involvement for an AI output
- 59:58of poetry, for example, to qualify? Oh, that's
- 1:00:00so interesting. I think you're basically saying,
- 1:00:03Michaela, that if you're using an alum, like
- 1:00:06if you have a ChatGBT account, I might be interpreting
- 1:00:08this incorrectly, but if you have a ChatGBT account
- 1:00:11and it knows you and you're using it for a long
- 1:00:15time and then you put in a prompt and it, based
- 1:00:18on what it knows about you, it spits out a poem.
- 1:00:21I would guess not, but it's a very interesting
- 1:00:24question. I feel like that's a great law school
- 1:00:25exam question. Well, you guys, I think we're
- 1:00:28out of time. It's 1 .30. Thank you so much. It's
- 1:00:31been a pleasure talking to myself and hopefully
- 1:00:34there are other people on the other line. I actually
- 1:00:35know there's a few of you guys out there because
- 1:00:37you submitted questions. But this has been great.
- 1:00:40Thank you so much. Take care.