Latest / CLEs You Actually Want to Hear / Thomson Reuters v. Ross Intelligence: AI/Fair Use Implications Going Forward with Jeremy S. Goldman
Transcript
- 0:00Welcome to our program, Thompson Reuters versus Ross Intelligence, AI and Fair Use Implications
- 0:12going forward.
- 0:13I'm Randy Bandman, chair of the Entertainment Law Section of the Beverly Hills Bar Association,
- 0:19and I'm delighted to introduce today's speaker, Jeremy Goldman.
- 0:23Jeremy is a partner in the law firm of Frankfurt, Kernet, Klein and Sales PCs Litigation Group
- 0:29and co-chair of the Emerging Technology Group, focusing on intellectual property, entertainment,
- 0:36media, privacy, technology and commercial law.
- 0:40Jeremy is an accomplished media and entertainment litigator, focusing on copyright, trademark,
- 0:46privacy, intermediary liability, right of publicity, first amendment and business disputes.
- 0:53Jeremy draws on his sophisticated understanding of law, technology and industry, along with
- 0:59creative problem solving and litigation skills to help clients navigate complex legal challenges
- 1:05in and outside the courtroom.
- 1:08Beyond his litigation practice, Jeremy advises startups and established companies on intellectual
- 1:13property, commercial licensing and privacy matters, especially when transactions and
- 1:18projects intersect with emerging technologies like artificial intelligence, blockchain and
- 1:24new media.
- 1:26Before we begin our program today, I wanted to point out that you, our audience, could
- 1:31submit questions in the Q&A box in the Zoom channel, and Jeremy will do his best to address
- 1:36as many as he can before he concludes.
- 1:39Thank you, Jeremy.
- 1:43You are welcome.
- 1:44Thanks, Randy.
- 1:45Hi, everyone.
- 1:46Thanks for joining today.
- 1:47So excited to present today on Thompson Reuters versus Ross Intelligence.
- 1:52We're going to be covering some context and background involving AI and fair use and the
- 1:58big AI copyright-related litigations that are going on.
- 2:02I know there's been other programs that focus more on the landscape generally and the litigations
- 2:06generally.
- 2:07We are going to get into more of a deep dive of the fair use issues, and in particular,
- 2:11the first decision that has, the first meaty decision I should say that's come out in the
- 2:17AI copyright litigation landscape involving fair use and that's what we're going to do
- 2:22today.
- 2:23So happy to jump in here.
- 2:24I'm going to share some slides.
- 2:27All right.
- 2:28Randy, we're good.
- 2:29You can see them?
- 2:30Yes.
- 2:31Okay.
- 2:32Thank you.
- 2:33Thank you.
- 2:34Okay.
- 2:35So how are we going to get there?
- 2:36How are we going to get to Thompson Reuters versus Ross?
- 2:38Because to jump right in would lose the forest for the trees here.
- 2:46So introduction is done.
- 2:48I've completed it.
- 2:50We're going to talk about generative AI generally, and I'm going to go through some key concepts
- 2:55talking about what generative AI is.
- 2:59And I'm going to assume that most folks here have some knowledge already of chat, GPT,
- 3:05and the like, but I'm going to try to give a little more nuance and definitions for folks
- 3:09to come out with a bit more education and knowledge around these concepts.
- 3:12We'll talk about some of the models, the training.
- 3:14And I'm going to do a, like I said, an overview of the litigation landscape as it stands right
- 3:20now.
- 3:21It's rapidly changing.
- 3:22We'll talk about there's dozens of cases.
- 3:24So of course we won't be able to cover them all or even any of them in any detail, but
- 3:29I want to give the landscape because it's important to be case and issue at hand.
- 3:34We'll do a sort of a case study of probably the highest profile AI copyright case that's
- 3:40going on right now, which is New York times versus open AI.
- 3:44And we're not going to talk about any fair use rulings in that case because there have
- 3:49been none, but I think understanding the facts and the background of that case and the nature
- 3:53of the fair use question that is likely ultimately resolved unless the case resolves earlier.
- 4:01I want to set that up.
- 4:02And then we'll pause and we'll take a step back on the legal front, talk about fair
- 4:07use generally.
- 4:08I'm really not going to make this again a one-on-one course where we're going to plod
- 4:13through the four factors of fair use so that we'll cover it.
- 4:17Instead what I'm going to do in that section is go through some of the key cases that I
- 4:22believe are going to figure into the court's oral analysis as they try to wade through
- 4:29the, what I see as sort of the biggest copyright and existential question facing sort of technology
- 4:37today, and that is whether the use of copyright protective materials to train AI is fair use.
- 4:43With that foundational knowledge of fair use, we will then jump into the recent decision,
- 4:50recent meaning exactly one month ago, really, February 11, 2025, coming out of the district
- 4:56of Delaware with the Thomson Reuters versus Ross Intelligence case, which is why you're
- 5:01all here today.
- 5:02I'm going to go through that case in some level of detail and learn how the court tackled
- 5:07fair use in that case.
- 5:09And then I have some slides that really are more for sort of in-house counsel talking
- 5:12about the business risks and practical considerations that come out of these various AI cases and
- 5:19with a focus on the recent Thomson Reuters decision.
- 5:22We'll see if we have time to get to that.
- 5:26And you know, it's interesting.
- 5:27I think I see that there's no real questions coming in yet.
- 5:30Just links to program materials.
- 5:32Okay.
- 5:33Folks can try to put things in Q&A.
- 5:35If you have questions, I'll either get to them at the end, or if I'm savvy enough, I'll
- 5:39try to grab them as we go along.
- 5:41So thank you.
- 5:43Starting with generative artificial intelligence and really artificial intelligence in general.
- 5:49Let's just talk about some key concepts here.
- 5:52Generative AI is artificial intelligence, and I'm not going to even attempt to define
- 5:57what that is.
- 5:58It's all over the place.
- 6:00But what makes generative AI interesting is that it allows you to produce new content,
- 6:05you know, text, images, videos, et cetera.
- 6:09And the way that it does that is by identifying patterns in very large-scale data sets.
- 6:15And in terms of sort of historical context, what's important to understand is that artificial
- 6:20intelligence, of course, has been around for a very long time.
- 6:26But in the past sort of three years, there's been significant breakthroughs along the lines
- 6:31of the Industrial Revolution.
- 6:33If you're going to, you know, find a metaphor and analogy with certain technologies involving
- 6:38transformer architecture and giant neural networks.
- 6:43And so with the speed and capacity of computers and some of the ways that they figured out
- 6:48how to dump massive amounts of data and use processors that can do parallel processing
- 6:55at the same time.
- 6:56So you're not just giving a computer and spitting out an instruction at once.
- 6:59Rather, a computer can, in parallel, deal with multiple instructions and ingest material
- 7:06at a very rapid pace.
- 7:08What the developers figured out was that by dumping in vast amounts of data, they can
- 7:16create these large language models, which are artificial intelligence models that are
- 7:23trained on these huge text corpuses or corpore, I think is the word, that instead of just
- 7:31predicting behavior, what they're able to do is take in this large amount of text or
- 7:37other data and it's able to form, it's able to convert that into a numerical, alphanumeric
- 7:45string and then find relationships between patterns and relationships between one word
- 7:50and the next, it's called a waiting process, and it can predict the next word that's going
- 7:56to appear based on the prior one.
- 7:58And it doesn't do that using words, it does that in the language of computers where it
- 8:03converts, like I said, these text strings that are natural language and it converts
- 8:08them into these alphanumeric strings that are very complex and can find relationships
- 8:12between one word and the next and find patterns in the way that people speak in language.
- 8:19That sort of concept has been expanded and it's no longer just about text, but they can
- 8:23also do it with images, video, sound, and the like, and there's been an explosion of
- 8:32tools and we'll look at some of those.
- 8:34Again, just going through some of the terminology, prompts are the inputs that users or the system
- 8:40sometimes can, you put into one of these AI models to guide the model's response and
- 8:47outputs of course are the text images or other content that is created and shown to the user
- 8:53or generated from the AI.
- 9:01How are these models trained?
- 9:03Well, as I mentioned, they're trained using massive data sets with billions or sometimes
- 9:10trillions of data points that are ingested into these powerful machines to learn the
- 9:16patterns.
- 9:17Where do they get these massive data sets?
- 9:20Again, I think a lot of this is known, it's been in the news and it's part of these cases.
- 9:25Well, the tech, it comes from various sources, but it's come from websites and the way that
- 9:29they do that is by either hiring a third party or themselves basically spidering indexing
- 9:36the entirety of the world by web.
- 9:38They've ingested books, vast amounts of social media posts, massive code repositories of open
- 9:45source code that live on places like GitHub, around the web.
- 9:52In addition to text, there's also AI, as I mentioned, that's trained on images and video
- 9:58and audio sources.
- 10:02What's implicit in this is that the text and the materials that are used to train this
- 10:07AI, it requires to do these large language models, like I said, a massive amount of data.
- 10:14That data includes, for most models, includes materials that have been licensed, and we'll
- 10:21talk a little bit about that, and material that is unlicensed and may be copyright protected
- 10:28and unlicensed.
- 10:36Just to get a sense of the industry and the big players in the space.
- 10:43You can also, like, you're knowledgeable about the artificial intelligence markets and industries.
- 10:47I think it's useful to break it apart between the foundational models, the applications that
- 10:57are released by those foundational models and the companies behind them, and then the
- 11:04plethora of, I call them vertical AI solutions that, for the most part, are leveraging and
- 11:12powered by the foundational models.
- 11:15In terms of the foundational models, there's really only a handful of them that are in
- 11:20play at that level of being the giant platforms.
- 11:25They include OpenAI, which has its model GPT.
- 11:30Google has Palm or Palm II.
- 11:32Anthropic, with Clawed, Meta, has its model called Lama.
- 11:37These are mainly acronyms.
- 11:42There was a lot of news in the last couple months around DeepSeq, which is a Chinese
- 11:48AI model and company, which sort of blew not the socks off a lot of the AI industry by
- 11:55coming out with a model called R1 that prepared it to be just as powerful as OpenAI, but got
- 12:01our OpenAI and GPT and the other ones that I listed, but allegedly without spending the
- 12:06amount of money and power that the others did.
- 12:08There's some skepticism around that, and we don't need to get into the details of that
- 12:13right now.
- 12:14Those are the big players.
- 12:16Those players, in addition to the models, which are underlying it, and you can't use
- 12:21a model, you need software, just much like you have an operating system and you need
- 12:27applications to sit on top of an operating system.
- 12:31Here you have models.
- 12:32Oh, GROC3, thank you.
- 12:33Yes, Gem and I.
- 12:34GROC3, I didn't list them all.
- 12:37Okay, so there are others, thank you, Ralph.
- 12:40But there are other models.
- 12:43And the foundation of models, they released their own applications.
- 12:49Of course, one of the more popular ones, ChatGPT, which sits on top of the GPT model.
- 12:54Bard is Google's version, which sits on top of its Google one, and so on and so forth.
- 13:00And then in addition to those applications sitting on top of the foundation of models,
- 13:06you have, I mean, I think there's hundreds, thousands of applications that are coming
- 13:11out more every day that are vertical because they're often tailored to a specific industry,
- 13:20legal being a very big one, and there's, I put Harvey.ai, it's just one that I'm familiar
- 13:25with.
- 13:26Bloomberg GPT has one for financial data, Google is coming out with a medical LLM, and
- 13:32I have friends that are doing startups around this and a significant, just a huge explosion
- 13:43of tools.
- 13:44And we'll look at one of them today.
- 13:46The case that we're talking about today is one of those vertical AI solutions that doesn't,
- 13:51is not a foundational model, but is leveraging other AI, and actually, you know, is one of
- 13:56the, it's been around for a while, but we will get to that.
- 14:04So with this massive explosion in foundational models, in applications, and in verticals that
- 14:13are trained on, often trained on unlicensed datasets, it is not surprising that there
- 14:21has been a huge surge of litigation.
- 14:25And so let's talk about, sort of first, we'll do a bit of a timeline of the key, what I
- 14:31see is sort of the key moments in the AI copyright related litigations, and then we'll look at
- 14:38it and we'll zoom out a little bit.
- 14:40So sort of very interestingly, and I guess it sort of makes sense, the first, really
- 14:46the first big one to get filed is the first one that had some sort of real fair use ruling,
- 14:52which is Thompson Reuters versus Ross Intelligence.
- 14:56That was filed out of all places in the district of Delaware, that way back in 2020, I mean,
- 15:02that's going back five years ago, and before folks were, before ChatGPT had become a household
- 15:09sort of product at name, and that case, you know, went on to the courts and we'll get
- 15:13to that.
- 15:14So there's some poetry there.
- 15:16The next sort of, the one that really launched in my opinion, sort of the modern era of the
- 15:22AI copyright cases was a class action lawsuit.
- 15:25All the other ones are, all these cases are still pending, I should say.
- 15:29Doe versus GitHub, which was in, oh, you want a PowerPoint?
- 15:34Okay, we'll talk about that later, which was alleging an allegation by a class action
- 15:43lawsuit really filed by developers, you know, coders, and in particular coders who released
- 15:50their software under open source licenses and posted them on the web for the taking,
- 15:55but subject to these licenses, they filed a lawsuit as a class against Microsoft and
- 16:00open AI alleging that they violated their copyright, they violated the terms of the open source
- 16:06licenses and they violated the terms of the DNCA, in particular, the provisions that relate
- 16:11to copyright management information, which we'll touch on, but it's not the focus today.
- 16:16And the way, the reason they said they did that is because they ingested all this code
- 16:19to create these applications, Codex and Co-Pilot, which are applications that do predictive
- 16:24coding and, you know, again, rely on these large language models that were created by
- 16:29open AI at Microsoft.
- 16:31And so that case was in 2022.
- 16:35That case didn't get a ton of attention.
- 16:38It was really when ChatGPT became big and then, and some of these other applications, people
- 16:45started seeing the generative AI create these incredible images when you saw Getty come on
- 16:52the scene and then there's a class action that is related to it, also against the same
- 16:56company called Stability AI, which is one of the foundational models with tools connected
- 17:03to it.
- 17:04And they actually filed this lawsuit in the United Kingdom.
- 17:06There's a class action that's pending here at the leave in California.
- 17:11And in that case, the allegation is that Stability AI ingested, you know, maybe the entirety
- 17:16of Getty's images to create its image generator, alleging that they're creating this competitive
- 17:22product by ingesting, you know, tens of thousands of their copyright protected images.
- 17:28There's trademark claims.
- 17:29There's copyright management information claims and the boot, but at the core of it is a copyright
- 17:34type of claim.
- 17:36And the next sort of big hit here is the one that I mentioned that I'm sure most folks
- 17:43have heard of, which is when the New York Times pulled the trigger and filed a lawsuit
- 17:47against OpenAI, you know, alleging that OpenAI and Microsoft, which is, you know, is one
- 17:55of OpenAI's biggest investors and alleged that OpenAI and Microsoft had trained their
- 18:01models using copyright protected articles and materials from the New York Times.
- 18:08And we will do a bit of a deeper dive into that case, really as a case study, but also
- 18:13again, because it's sort of the hottest, most high profile case.
- 18:21Just to get into the nature of these cases, and I'm just trying to get, if it wasn't clear,
- 18:26just trying to give everyone a sense to get their arms around these litigations through
- 18:31different angles.
- 18:32And so now what I'm trying to look at is really distilling the nature of these cases more
- 18:36generally, you know, who's suing who for what.
- 18:40And here, you know, who's suing who, the answer is, you know, the media company, really it's
- 18:45media companies, book authors, artists, music publishers, software developers, you know,
- 18:50copyright holders, creators and the companies behind them.
- 18:55And they are suing major AI providers and technology platforms.
- 19:00You know, what's interesting here, I'll just do a quick note, who's not being sued so far,
- 19:05really users of these platforms, and I should say end users of these platforms, I'm not
- 19:10aware of any real litigation that's been filed against folks that are using generative AI
- 19:15tools or any of these tools, alleging that their sort of end use of the tools is violating
- 19:20copyright.
- 19:22I'm saying end user, because in theory, a company like Ross Intelligence might be leveraging
- 19:28and using, you know, open AI or other tools, actually don't know if it's open AI, so don't
- 19:33hold me to that.
- 19:34But they, I believe that they're powered by other third party tools, but no end users
- 19:38have been sued yet.
- 19:39They're really the major AI providers, the big application providers.
- 19:44And the allegations are, you know, typically in the same penumbra of claims, which is that
- 19:50the AI developers, one that they scraped the internet, so there's a whole slew of claims
- 19:56involving web scraping that we're not going to get into today, it's a whole other area
- 20:00of law, but you know, connected.
- 20:02But the one claim is you took all this stuff and scraped it without permission and might
- 20:06have violated our rights just with the scraping.
- 20:09And then you, after scraping it, you then use that to train the AI and these tools for
- 20:16commercial purposes.
- 20:18And what kind of causes of action are they alleging?
- 20:21The sort of the main bucket of causes of action are, you know, direct and indirect copyright
- 20:27infringement, the direct claim being by, you know, copying our materials for the training
- 20:32and putting it in.
- 20:33It's a copy, but also there are allegations of indirect copyright infringement, both with
- 20:38respect to the web scrapers that are doing your alleged dirty work.
- 20:44And you have users, the allegation is the end users that are using our products and tools
- 20:49to create images and outputs that violate copyright and you are giving them the tools
- 20:55to do it.
- 20:56And so that's the indirect copyright claim.
- 20:58And also, as I mentioned, claims under the digital millennium copyright act, which are
- 21:04not claims for copyright infringement, but are claims that the AI tools in your, in the
- 21:09course of your scraping, in the course of your structuring the data to be interpreted
- 21:16and ingested by these AI tools, you have removed our copyright management information.
- 21:22You have removed the name of our publication, the name of the author, the terms and conditions
- 21:27that get attached to it.
- 21:28The metadata information about the copyright, the allegation is, were removed by these AI
- 21:35tools.
- 21:36And the allegation is you did it intentionally and you did it for the purpose of concealing
- 21:42the infringement down the line so that when folks down the line are going to infringe
- 21:48on our rights, there's no way for them to know that there's sort of infringement because
- 21:52there's no copyright management information that is attached to it anymore.
- 21:57And courts have weighed in on some of those CMI claims and the district, it's just district
- 22:03court level decisions.
- 22:04They've actually been split on it with two decisions recently upholding the claim, although
- 22:09it didn't require sort of a high level of proof in order to make out the CMI or DMCA
- 22:15claims.
- 22:16And then one court in the southern district, not allowing that claim to go forward.
- 22:19So there's a bit of a split, a difference of opinion on that issue and indeed on the
- 22:24fair use issue, as we'll say.
- 22:27And then sort of the other claims include breach of contract claims, which really arise
- 22:33out of the web scraping more than anything where you have terms of service that are
- 22:39attached to the content that's online and the allegation is that you violated those
- 22:44license agreements or those terms in doing so.
- 22:49So I have a question here as I'm trying to follow along is, are ingesting copying scraping
- 22:54effectively synonyms in AI cases?
- 22:56Great question.
- 22:57No, I would say they're not synonyms.
- 23:01I know I'm using them a little bit similarly.
- 23:04I mean, scraping, it's sort of the best way to think about that is copying can occur at
- 23:09any level where you can have copying during many different steps of the process that ultimately
- 23:16lead to a large language model converting text into weights and points, which are not
- 23:24text really, not text as we know it.
- 23:29That can happen at any level.
- 23:31Ingesting is what I'm describing as the process of taking that text into the models.
- 23:36And then scraping is really the, it sounds devious, but it's the, that is a particular
- 23:42type of copying where you go on the web and you have typically automated tools that copy
- 23:48the entirety of a web page.
- 23:51And so scraping is really specific to copying material that's posted on the internet.
- 23:56So they're all sort of different, similar types of things.
- 24:03So we're going to talk about constraining AI learning to specific set of data actually.
- 24:06So I'm going to get to that.
- 24:07All right.
- 24:08So just another way of looking at the landscape here, credit to Wired for this slide.
- 24:18And they're doing a nice job of tracking these cases.
- 24:21This is just the visual of seeing the amount of cases that have been filed over the course
- 24:24of the last, you know, basically five years, but mainly the last two or three years.
- 24:32And you say you can get a sense of who the plaintiffs are, who the defendants are in
- 24:35these cases, and depending all over the country.
- 24:38All right.
- 24:41As I hit almost the 30 minute mark, I'm going to, I'm going to try to stop myself from answering
- 24:49questions, not because they're not good questions, but because I want to make sure I cover the
- 24:51slides for everyone.
- 24:52I mean, well, I'm happy to stay on for a while after if there's time, if folks are
- 24:57interested.
- 24:58Okay.
- 24:59New York Times versus OpenAI.
- 25:02This was filed in December of 2023.
- 25:05The docket is like, you know, 6,000 entries long, so I can't possibly dig into it in any
- 25:11level of detail.
- 25:13So I'm just going to give some sort of high points and give folks an idea of what the
- 25:16case is about and where it stands right now.
- 25:20New York Times, they tend, they're actually, they're actually quite good writers over there
- 25:24at the New York Times.
- 25:25They wrote a very nice complaint that's not weighing in on the merits.
- 25:27I should say I am not going to, I'm hoping not to give anyone any sense of my opinion
- 25:31on the merits of any of these cases.
- 25:33I'm here strictly as hopefully an educator, but you know, the Times has for 170 years,
- 25:39more than 170 years, the Times has given the world deeply reported expert independent journalism.
- 25:44And the allegation is that the unlawful use of the Times's work to create AI products
- 25:49that compete with the New York Times and that competition is going to be key, as we'll say,
- 25:54they're claiming that it threatens their ability to provide the service.
- 25:58And you know, the allegation is that, as I just talked about, that they relied on these
- 26:02large language models that were built by copying and using millions of the Times's copyrighted
- 26:08news articles, their investigations, et cetera, that this was infringement.
- 26:14And the allegation is, we'll just look at this because it's quite intriguing, that the
- 26:21allegation is that open AI and Microsoft's AI tools, really they focused on on chat,
- 26:26and BT and Microsoft and the allegation was that these tools have the ability, so it wasn't
- 26:34just the ingestion, but they're saying that they have the ability to produce outputs that
- 26:39produce content verbatim or nearly verbatim, summarize it and mimic not just the facts
- 26:46and the ideas that are included within the New York Times's articles and the way that
- 26:50we speak, but also they allege the expressive style and they have attached an appendix
- 26:55and examples in the complaint itself where they produced large swaths of text that they
- 27:02claim were directly from the New York Times and here you can see sort of a red line comparison
- 27:10that they put in the complaint.
- 27:12And you know, you look at that and man, it does it does look a little alarming.
- 27:17The New York, the open AI shot back, however, and says the allegations in the Times's complaint
- 27:25do not meet its famously rigorous journalistic standards.
- 27:30The truth which will come out in the course of this case is that the Times paid someone
- 27:33to hack open AI's products.
- 27:35It took took them tens of thousands of attempts to generate the highly anomalous results that
- 27:41make up exhibit J to the complaint.
- 27:44Again, super clear.
- 27:46I am not opining on whether they are right or wrong and what happened, but the allegation
- 27:51here by or the response to the New York Times allegation of verbatim copying is essentially
- 27:58that New York Times beat the chat GPT with a stick until it came out and said the things
- 28:05that they wanted to say.
- 28:06And open AI is a sort of saying, you know, when you have, they're kind of relying on
- 28:11the old and we're not really going to get into this case.
- 28:13So I guess I can mention it now.
- 28:15When you create a tool that has substantial non infringing uses and you are using important
- 28:19infringing use, then you can't blame us.
- 28:23And so again, I'm not focusing on this component of it, but it is interesting to me that the
- 28:28New York Times is at least pinning some of its copyright claims on this allegation of
- 28:34verbatim outputs that are created by the software.
- 28:42Causes of action line up with the other slide, but since we're doing the case example, we
- 28:46could see how it fits to these facts claims for direct and various liability for copyright
- 28:52infringement for, you know, both storing processing, reproducing the training sets that include
- 28:59the works and then disseminating outputs that are copyright that are derivatives, they're
- 29:03claiming or copies or derivatives.
- 29:05They're also claiming contributory copyright infringement.
- 29:10As I mentioned before, based upon the direct infringement by end users of the product,
- 29:16and they're also going after Microsoft for saying that they are providing the infrastructure
- 29:23to host and operate and commercialize these various GPT models.
- 29:28They are claiming DMCH section 1202 violations.
- 29:32Those are the CMI claims that we talked about again for, you know, taking an article like
- 29:37the one we just showed, taking this on the left side, a GPT4 model, and what is missing
- 29:43here is any byline, is any attribution to the New York Times, and they're saying that
- 29:47you have stripped that out, and now people might take that text and use it, and that's
- 29:52going to perpetuate further infringement.
- 29:56They threw in an unfair competition claim by misappropriation and a claim for trademark
- 29:59dilution.
- 30:00I don't know what either of those are, being a copyright lawyer, just kidding, I do know
- 30:05what those are, but we're not going to talk about them today.
- 30:09So, just again, just in terms of the big milestones and giving us a little bit of a status update
- 30:19where we are in this sort of mega case, and again, this is not the only one.
- 30:23There are very important lawsuits, all the ones that we showed on that screen by Wired.
- 30:27They all, you know, there's major authors that are suing major platforms.
- 30:31Just one, that's an example that I think it's good to focus on.
- 30:36Complaint was filed back in December of 2023.
- 30:39OpenAI and Microsoft moved to dismiss, I'm just focusing on OpenAI for right now.
- 30:44They moved to dismiss in February of 2024, so about a year ago.
- 30:49They did not move to dismiss the entirety of the complaint, and as we'll see, they did
- 30:55not move to dismiss on fair use grounds.
- 30:59They did move to dismiss the direct copyright infringement claims to the extent that those
- 31:04claims are time barred.
- 31:07They are moving to dismiss the contributory claim based on the end user infringement on
- 31:13the ground that it is only alleging a generalized knowledge that end users may infringe, and
- 31:20there was no specific knowledge of any particular infringement, which OpenAI is arguing is necessary
- 31:26to claim contributory liability for copyright infringement.
- 31:31They're claiming that the DMCA claim fails as a matter of law, and we've, you know, like
- 31:36I said, there's been a split of authority or authority on that, even in the Southern
- 31:41District, judges feel differently about whether those claims should survive or not survive,
- 31:46and that the misappropriation claim is preempted by the Copyright Act.
- 31:51We're not going to really get into copyright preemption today, though I would love to.
- 31:54It's one of my favorite topics to talk about if BHP wants to have me back.
- 31:58I could do a whole thing on copyright preemption, which I'm mentioning here not just because
- 32:03it came up in that case, but because copyright preemption becomes a very important issue
- 32:08in these AI copyright cases.
- 32:11Why?
- 32:12Because of the nature of the various claims that are being alleged, including with respect
- 32:17to the web scraping claims based on things like breach of contract or trespass the chattels
- 32:23plan, ultimately, there may be an argument that all of those are ultimately in the nature
- 32:27of you copied my stuff claims.
- 32:30And there is an argument that the Copyright Act is supposed to cover all of the you copied
- 32:35my stuff claims, regardless of the style, how the, you know, cause of action is styled.
- 32:40So they're making an argument like that with respect to misappropriation claim.
- 32:44After the motion to dismiss the original complaint, the New York Times filed a first
- 32:49amended complaint really just to add some additional articles and copyright protective
- 32:54materials.
- 32:55I believe I did not deep dive into this.
- 32:57What I do know is that there was no sort of second motion to dismiss.
- 33:02I think the parties are all sort of like resting on the original motion saying that nothing
- 33:06really changed with the amended complaint.
- 33:10The other big beat that's important to know is sort of in the fall of this past fall, the
- 33:15court consolidated the case by the New York Times by cases that were filed by other journalists,
- 33:23including the Daily News, not including specifically the Daily News and the Center for Investigative
- 33:29Reporting.
- 33:30They combined those cases in the New York Times case.
- 33:32And so now open AI and Microsoft are facing multiple plaintiffs in those cases.
- 33:38And there was a motion to dismiss hearing in January.
- 33:43And now I think folks are waiting for the court's ruling.
- 33:47I wasn't able to attend that hearing.
- 33:49Maybe some folks on this webinar were able to attend that hearing.
- 33:52What I do understand from the reporting on that hearing is that there were substantive
- 33:56discussions getting into fair use and that big question.
- 34:02What's interesting though is that this case is going to be going on for a long time.
- 34:07Just to queue up with the fair use question is in that case, and I see this case again
- 34:11as like maybe the Bikuhuna or one of the Bikuhunas in terms of the fair use question.
- 34:18Open AI is arguing that the use of the material to train these LLMs is fair use because it
- 34:24serves a new or transformative purpose.
- 34:27The New York Times tries to preempt that argument right away in its complaint.
- 34:33There is nothing transformative about using the Times's content without payment to create
- 34:37products that substitute for the Times and steal audiences away from it.
- 34:42They also claim that because the outputs of the models compete with and closely mimic
- 34:46the inputs to train them, copying those works for that purpose is not fair use.
- 34:52And as I mentioned, Open AI did not move to dismiss based on fair use.
- 34:56They're likely saving that for summary judgment after there's more facts than they are deep
- 35:02into discovery and all of the battles that come with a massive high stakes litigation
- 35:08right now.
- 35:09Okay, so we've gone for, I have 25 minutes left to talk about fair use precedent and
- 35:19do a little bit of a primer on fair use and then we'll jump into the recent decision
- 35:24from last month.
- 35:27Like I said, this is not a 101 course.
- 35:29If people wanted to do copyright 101, we would cover these in detail.
- 35:33Most folks here that are coming to this kind of course are probably going to be familiar
- 35:37with the four fair use factors that are set forth in section 107 of the Copyright Act.
- 35:42These are non-exclusive factors, but every court goes through them and they at least
- 35:46prefer to drive decisions in these cases.
- 35:50Typically the first and the fourth factors get the most weight and with the second factor
- 35:55really sort of where on the spectrum is the copyrighted work in terms of like a really
- 35:59fictional, creative work versus a non-fictional work which gets less and something that's
- 36:03more functional and reflecting ideas, gets less protection.
- 36:06The third one, you know, the amount of substantiality of the work, those two tend to get less weight
- 36:12and so it is with the cases that we're going to talk about.
- 36:17No one needs me to cover these any longer, right?
- 36:19Okay, let's move on.
- 36:21Much more fun.
- 36:22This is Jeremy's grab bag of fair use decisions that I believe are going to play an important
- 36:31role as courts start deciding the big questions around fair use and so I thought it was worth
- 36:39pausing and sort of reviewing fair use jurisprudence from the position of the most influential
- 36:46courts that have weighed in on this.
- 36:49So starting with the most landmark case on fair use jurisprudence versus Campbell versus
- 36:57Acuff Rose, this was a case where the hip hop group Two Life Crew did a song that was
- 37:06used guitar licks and used some of the lyrics from Lloyd Orbison's Pretty Woman and the
- 37:12court, the Supreme Court in that case held that this was a parody which in itself was
- 37:18a finding.
- 37:19This was a parody and that in order to do a parody you needed to use the original and
- 37:23they said that the parody was transformative and it was a transformative use because it
- 37:30created a new work that shed new light and had new commentary on the original and in order
- 37:35to do that you need to conjure up the original and use it.
- 37:38So you had a need to use the original in order for it to be trans use, fair use and they
- 37:42court really liked that they made something new.
- 37:44They made a new creative work in the process and the term transformative use, that was
- 37:51a term that the Supreme Court of the United States borrowed from Judge Pierre Laval who
- 37:57is on the second circuit and had written a fair use article, I think for the Harvard
- 38:01Law Review a few years earlier and that became and has been ever since the prevailing terminology
- 38:09that courts use which is whether the use is transformative that is some folks on this
- 38:14call may say well that transformative use case has swallowed the entirety of fair use
- 38:20although courts tend to discuss it only in relation to the first factor, the purpose
- 38:25of the use but we'll get into that in a little more detail.
- 38:30So parodies are fair use and they're transformative, Supreme Court says.
- 38:33Second case I want to cover is Sony vs. Connectix, there's another case involving Sega.
- 38:38These are cases involving reverse engineering and the Sony case you had Sony which had
- 38:45its PlayStation video game console, they made the PlayStation console and Connectix made
- 38:52a emulator, they made for the Macintosh I think for Windows also, they made a piece
- 38:59of software that allowed a Windows machine or a Mac machine to be able to play PlayStation
- 39:05games on your computer, not something that Sony did, not something that anyone did and
- 39:09in order to create that emulator, Connectix had to take the Sony PlayStation and reverse
- 39:15engineer it and with it the BIOS, the underlying sort of operating system of the console which
- 39:20was copyright protected and they had to copy it and study it in order to create this emulator,
- 39:27the emulator itself didn't use any code, didn't use the BIOS, didn't use any Sony materials
- 39:32and the ninth circuit in that case said this was fair use, they said that this is transformative
- 39:38when you take, when you do reverse engineering to get to the functional elements and understand
- 39:44it but they really liked that what you were doing was creating a non-infringing product
- 39:49which was the emulator and that it expanded the market and allowed consumers to sort of
- 39:53have new ways to play these games and so in that case the court held that was transformative
- 39:58and even though it was undisputed that Connectix had made these intermediate copies of the
- 40:03BIOS, the court was willing to overlook it in under the banner, under the flag of fair
- 40:09use.
- 40:10Next up, Authors Guild versus Google and there's a corollary case, Authors Guild versus Hathi
- 40:15Trust disclaimer, I represented the Authors Guild in those cases but again I'm not here
- 40:20in capacity as advocate, I'm here hopefully as educator.
- 40:24The Authors Guild sued Google after Google in partnership with dozens of massive research
- 40:32university libraries around the country, Michigan, Wisconsin, California and they digitized
- 40:40using scanning beds, they scanned and then used optical character recognition and digitized
- 40:47millions and millions of copyright protected books, they did that in order to make them
- 40:53searchable on Google Books which is books.google.com which is part of their search engine and
- 41:01is according to Google, this is further to their core mission which was to make the entire
- 41:05world searchable and they said it's not enough to make the web searchable, we want to make
- 41:08all books searchable as well so Google had the Google Books application and the university
- 41:14libraries had a consortium called Hathi Trust where they used the same sort of search index
- 41:21that was created in order to make those books searchable so you can go onto a prompt, put
- 41:26in some keywords and find where those keywords hit inside of books and on web pages and Google
- 41:33in the Google Books program they would display a snippet, just a few words before, a few
- 41:39words after and the text that you were looking for in the book and the idea was so that you
- 41:46can then go and get the book and get the original source, authors around the country, authors
- 41:51around the world came together and alleged that this copying and scanning and digitizing
- 41:58of millions of books by a commercial juggernaut like Google in order to help them advertise
- 42:05was copyright infringement.
- 42:07The court when adhered to the second circuit on appeal held that both in the Google case
- 42:15and in the Hathi Trust case held that this use even though you had digitized and scanned
- 42:21millions upon millions of copyright protected books when you did that for the purpose of
- 42:27making them searchable that is a different use than the original, the court ticked through
- 42:31the factors and I'm not going to get into super detail but it was also interesting to
- 42:37me here that one of the reasons the second circuit found the use transformative and
- 42:43ultimately found that a compelling defense for fair use was because it augments public
- 42:47knowledge.
- 42:48There were also very other important factors, the court found that there was lack of a sort
- 42:53of a fledgling or existing licensing market for books for the purpose of making them searchable,
- 42:59of course lots of books get scanned and OCR to be distributed as e-books but the court
- 43:05said that there wasn't enough evidence of a market for books for this specific purpose
- 43:10and the court held that it was fair use.
- 43:14For folks that are sort of following this, to me that case shares some hallmarks of the
- 43:20cases involving the generative AI where you're scanning and ingesting millions of books and
- 43:26you can imagine that open AI and the other platforms will be looking to cases like this
- 43:31to hold that they are on all fours although as we'll talk about there are arguments on
- 43:37the other side and reasons for distinguishing those cases.
- 43:42Next up is Google versus Oracle and this one involved APIs which are little pieces of software
- 43:49that allow computers to talk to each other and in that case the holding was that the
- 43:59API code, first there was a challenge around the copyright ability of the API code, the
- 44:02court said let's assume it's copyrightable.
- 44:05The court found that Google's use of this API code that belonged to Oracle was transformative
- 44:10and why, the reason they found it was transformative was because this code was necessary for other
- 44:17developers to be able to create new computer programs.
- 44:22There's a theme here, it's like the courts and fair use and copyright in general really
- 44:27like when if you're going to use something and it's going to cause other people to be
- 44:33able to create new stuff they really like that and by the way you might have noticed
- 44:38I said when they use it so that people can create other new stuff query how that might
- 44:44change when it's a machine that might create new stuff versus a person.
- 44:47I don't know, interesting.
- 44:49Finally let's just look at Warhol versus Goldsmith just came out a couple years ago,
- 44:54I'm sure that case was covered at Nazium at the Beverly Hills bars it was across all
- 44:59sort of bar groups that touch on copyright and entertainment.
- 45:04That was a case where one of Warhol sued, sorry Goldsmith sued a photographer sued the
- 45:12Warhol estate for licensing a Warhol work for the cover of a magazine.
- 45:19Goldsmith the photographer of a picture of Prince said I licensed this for a commercial
- 45:25purpose to get a license fee so it could be cover of a magazine.
- 45:29Warhol took that photograph and he did his Warhol thing to it of Prince and he put his
- 45:34sort of Warhol touch on it and then the Warhol estate licensed it to be the cover of a magazine
- 45:41and the Supreme Court in that case said just making the doing the Warhol thing to the photograph
- 45:48is not enough when the original use and purpose is really the same as the final use and purpose.
- 45:56So looking at the original license for a magazine and then the secondary license for a magazine
- 46:02and the court said just making these aesthetic changes even though it's really cool not enough.
- 46:08What I also want to draw from that case was the Supreme Court really focused on justification.
- 46:16What is the need for using that work and in that case the court said you didn't have enough
- 46:21justification for using that work you could have gotten a license to use it.
- 46:27We'll pause with the CLE code.
- 46:28I think I have to do I'm assuming yes.
- 46:32Let's do NURL N-E-U-R-A-L.
- 46:41Now getting into the main event Thompson Reuters vs. Ross Intelligence.
- 46:49So the facts of this case well the beginning of the story I would assume that 99 or 100%
- 46:55of you already know which is that Thompson Reuters owns a little platform called Westlaw
- 47:00which is used for legal research and on Westlaw Westlaw has headnotes and a key number system
- 47:07that they build around the public domain judicial opinions.
- 47:12The allegation is that Ross and Ross really in conjunction with the third party doesn't
- 47:17really matter that they used these headnotes and Westlaw's key number system to train its
- 47:24AI based legal research tool.
- 47:29Now importantly and distinguishing this from the generative AI cases this is not a generative
- 47:35AI case this is not a case where folks put in prompts and out pops generative AI rather
- 47:42the allegation in this case and the nature of the training and actually to go to one
- 47:47of the questions that came in was that this wasn't a case where Thompson sorry where Ross
- 47:56Intelligence was creating an LLM from scratch rather this was a case where they were where
- 48:01Ross Intelligence was using some version it was alleged of Westlaw's headnotes in order
- 48:08to understand as I did way that I interpret this is how actual people actually speak and
- 48:14ask questions versus how lawyers and judges talk in their opinions right so if you're
- 48:18trying to find a ruling and you want to get to the decision and not just get like some
- 48:25treatise interpretation of it the idea was that you want to be able to put in a prompt
- 48:29a legal question and then the AI tool would be able to search the judicial opinions and
- 48:34then come back and show you where in the opinion that this to be able to match the plain language
- 48:41human being that puts in the question versus the opinion well in order to do that the allegation
- 48:47is that Ross Intelligence used the Westlaw headnotes to teach the AI to be able to understand
- 48:52how people ask questions and how people talk so on February 11 so I say on February 11
- 49:07because that's what happened last month but when I'm also focusing on years after previously
- 49:11indicating that Ross might qualify for fair use I did a blog post on this you know several
- 49:16months ago this court weighed in on summary judgment and held that fair use was a question
- 49:23for the jury and sort of looked at it and said well you know really what we should look
- 49:27at is whether the Westlaw headnotes were being used for their expressive purpose or were
- 49:32they being used for some other purpose in order to train it and just to understand how
- 49:38language works and so that was going to go to the jury. Everyone was everyone in the
- 49:45case that the parties were getting ready for trial after the court had denied summary judgment
- 49:50and were getting ready for trial and then the court pumped the brakes and said wait
- 49:55a minute I'm not sure that I feel really comfortable with what I did on summary judgment kind of
- 50:00had everyone go back and do some weird briefing I didn't exactly follow all the beats of it
- 50:04but out comes this opinion on February 11 after the courts looked at it again and you
- 50:11know the judge begins the opinion by saying and then put this in my slides but judge says
- 50:17judge oh by the way circuit judge Bevis sitting by designation from the third circuit sitting
- 50:23in the district of Delaware says a smart man knows when he is right a wise man knows when
- 50:28he is wrong wisdom does not always find me so I try to embrace it when it does even if
- 50:33it comes late as it did here great humility by this judge to reverse himself on the fair
- 50:39use ruling and going from saying this is a that we're going to let Ross intelligence
- 50:44present fair use to the jury comes back and says these materials are not fair use as a
- 50:49matter of law now this constitutes really the first major fair use ruling applying the
- 50:56fair use doctrine to AI training and it could of course influence the other AI copyright
- 51:02cases including the ones we talked about though the court was very careful to say I'm limiting
- 51:07my case to its facts and saying that that you know this is not a generative AI case he said
- 51:16that specifically how did the court go through the fair use analysis in this case well first
- 51:23starting with factor one the purpose and character of the use court this factor looks at whether
- 51:29the use is commercial or non-commercial first finding that it was definitely commercial
- 51:34indeed the court really focused on the fact that Ross intelligence's product is directly
- 51:39competitive with Westlaw and then it got into the transformative test and looked at the
- 51:48cases like Sony and Sega where intermediate copying had been held to be transformative
- 51:53indeed in the case in the last ruling by this judge the court did a deep dive into those
- 52:00intermediate copying cases to rule that it was a fair question whether this was fair use
- 52:06this time around the judge had a much more skeptical view of those cases and ultimately
- 52:11held that those cases were in and out the court distinguished those cases I don't think I've
- 52:18got into that exactly why yes the court distinguished those cases and said one those were about
- 52:24computer code and code is by nature functional and operates language in in in books and in
- 52:37in literary works is not the same as computer code by nature and so that's number one and
- 52:42then relatedly whereas in those reverse engineering cases it was necessary to copy the API code
- 52:52to copy the bios in order to understand it and in order to create a new and interoperable
- 52:59product court said here there's no there's no need in the same way to copy those elements
- 53:07in order to achieve your purpose of creating this legal research tool the court also relied
- 53:16heavily and very interestingly on the Warhol versus Goldsmith case in looking at the ultimate
- 53:25use here and comparing you head notes are used I mean I'm summarizing here but hand
- 53:30notes are used these hand notes are being used to create a competing product the original
- 53:35use of the hand notes is to create a legal research tool that is used by lawyers primarily
- 53:40and the ultimate use is the legal research tool that is being used by lawyers and you're
- 53:44going to cut into market share which of course flows into the last factor of market effect
- 53:52the second one the nature of the work you know the court said you know Westlaw hand
- 53:57notes are barely creative indeed the court really had to struggle a little bit with even
- 54:02determining that the Westlaw hand notes for copyrightable in the first instance given
- 54:05that they derived from opinions and I think the court recognized that you know some of
- 54:09them are copyrightable some of them maybe not so much and that they're mostly factual
- 54:14so the court did hold that that second factor favors Ross and the third factor as well also
- 54:19holding that that factor favored Ross interestingly you know the court said you know the fact
- 54:27that you copied thousands of them in training yes that that's true but the court also focused
- 54:33on the fact that the Westlaw head notes were not output to the users so what's interesting
- 54:42about that is that the courts focus on the output to me to me what's interesting the
- 54:47court saying that the output of the Westlaw head notes never being shown to the user
- 54:53by focusing on that primarily with respect to factor three the court in my opinion may
- 55:02might have deemphasized that factor a little bit as I mentioned before factors two and
- 55:06three tend to get a little bit of short shrift from the court in terms of the weighting of
- 55:11the of the for fair use factors and the and Ross intelligence was relying heavily on the
- 55:17fact that the output wasn't shown much like in the Google books case of saying we're only
- 55:21showing a tiny snippet out of the entire and out of the entirety the court didn't buy it
- 55:26and the court said here fine it favors you but it's only the third factor finally you
- 55:30know like I mentioned market effect the court helped one you're competing with them and
- 55:34potentially taking market share away from Westlaw and also the court noted that materials
- 55:41are being licensed that there's a an existing and fledgling and growing market for AI for
- 55:50copyright protected materials to be licensed to AI companies for training purposes and
- 55:56you undermine a potential licensing opportunity and the fact that Thompson Reuters wasn't
- 56:01doing it yet was of no sort of moment to the court they said well they could and they chose
- 56:06not to and it did you know Thompson Reuters has its own AI legal research tool that I
- 56:11started using recently pretty good so just sort of key takeaways from this case one and
- 56:21things I mentioned one is you know the court really emphasized that it's this is not a
- 56:24generative AI case and it addressed you know competing legal research tool I think that
- 56:31isn't really important to focus on this was one of these vertical AI applications it is
- 56:35not a general purpose AI tool it's also not a foundational LLM or a general purpose you
- 56:41know tool this was specifically for legal research and the training materials were from
- 56:46a particular vertical so you have vertical or vertical I think maybe that's important
- 56:50also it was interesting I noted before significant reliance on Warhol when that Warhol decision
- 56:56came out there were many fair use advocates that were concerned that that court that case
- 57:00would be interpreted to limit fair use I think this might be construed as an example of that
- 57:05where the court was looking really at the ultimate uses and what the need was for the
- 57:09original and finding that where they where the secondary use competes with or shows a
- 57:16similar purpose of the original not transformative.
- 57:20From an industry perspective at least right now with this one decision it really may start
- 57:25seeing a separation of or a signal to certain AI models and technology that are being trained
- 57:31on specific materials that compete and thinking oh you know what this may not ultimately be
- 57:36held to be fair use and that it may be better to go and get a license at least for those
- 57:40types of applications and sort of models that are leveraging gen of AI models and then finally
- 57:46like this is very early it's still very early days in the jurisprudence this is really the
- 57:51first ruling from the district of Delaware district court rolling we don't have any word
- 57:56from sort of the courts that you know tend to form the big cases here the second circuit
- 58:03night circuit second second third circuit could be the one to do it I'm just saying that
- 58:08none of those circuit courts have weighed in they could reverse this courts and the
- 58:12other jurisdictions could adopt different views of their use and so it is by far not
- 58:17the final word you know to me these cases are you know ripe the big question ripe for
- 58:23Supreme Court review although I would argue that maybe some of the circuit courts are
- 58:27just as well positioned as the Supreme Court to weigh in on this stuff who knows so with
- 58:33that I'm going to stop there and not talk about the other things because we only have
- 58:37one minute left I think there's a fair number of questions to address alright so you know
- 58:49a question of it asked about whether it's possible to constrain AI learning to specific
- 58:54sets of data well I don't I don't know I don't know the answer I mean I think that
- 59:00I think the answer there is like a technical question about whether you can you know only
- 59:04have certain amounts of data I think the key thing to understand is that with respect to
- 59:09the large language models from the perspective of open AI and its brethren that are creating
- 59:15the foundational models the argument that they are making sorry one of the key arguments
- 59:20that they are making in favor of fair use is you know we can't just go and license like
- 59:25Westlaw head notes or even and maybe even an entire library in order to train our AI
- 59:32rather in order for us to make this technology work in order for it to understand language
- 59:37all languages in order for it to understand all of human culture we need everything we
- 59:43need it's kind of like a little bit of an all or nothing kind of thing and so I think
- 59:48one of the arguments very interestingly is you know usually with fair use it's like you're
- 59:52trying to limit the amount that you take I think the argument here is we really needed
- 59:56it all in order for this to provide the functionality that we're providing question is there a chance
- 1:00:05that fan created work could be protected by case law future statutes I think I think you
- 1:00:11know you know I think that it's the same as all I think it's very hard to argue that it's
- 1:00:16different than any other sort of copyright protected work in terms of the protection
- 1:00:19I think that you know people are going to do them pretty much with whole cloth and in
- 1:00:23a fair use analysis sort of the type of the work once you pass the copyright threshold
- 1:00:27I think you end up in like that factor to what's the nature of the work and I don't
- 1:00:32know if you mean like fan fiction which is a separate idea and I don't I don't I so yeah
- 1:00:37I don't think that you're gonna get everything at the separate treatment person and Mark
- 1:00:42is just talking with with Jay who's of course very capable of answering all these questions
- 1:00:48so and Mark's just talking to Rosemary Mark says worth discussing the concept of torturing
- 1:01:00AI models to output any content you want including legal content this is why mid-journey used
- 1:01:06to include its terms that they will go after any end user who knowingly does it yeah I
- 1:01:09mean it's true I think this goes to this sort of one of the big sort of battles I think
- 1:01:15is just like who should be held liable to the extent that there's copyright infringing
- 1:01:20content that comes out of AI and either just like it's shown to a user which is me like
- 1:01:25a little bit of like a tree in the forest I think the bigger deal is like when these
- 1:01:29AI tools generate content that is infringing and then they use that for the commercial
- 1:01:36purpose or how you know somebody takes it and publishes a book or something that's a
- 1:01:39movie and then somebody says no my materials used to ingest that who get who's liable for
- 1:01:44that is it the user does it matter kind of what kind of prompts they put into the machine
- 1:01:50in order to generate that or does the you know does the AI model have a potential secondary
- 1:01:55liability there I think those are some of the really big questions that are being asked
- 1:02:00did Ross have no legal advice bad legal advice I have I can't answer that question for sure
- 1:02:07Jay I mean so why couldn't Ross just use the public domain to judicial decisions why did
- 1:02:11it need to use the head notes with their albeit minimum copyright protection well great question
- 1:02:16I mean the point there is I will I'll say what the judge really picked up on that and I think
- 1:02:23he agreed with your very leading question which is you know authors get paid to create
- 1:02:29stuff and you know Westlaw Thompson Reuters paid legal writers and the like to create
- 1:02:37Westlaw head notes and they got paid for that and that was expensive for Thompson Reuters
- 1:02:41to do and if Ross wanted to you know create could have created its own or it could have
- 1:02:48used public domain well let me let me just take a pause Jay I think the key distinction
- 1:02:53is that the judicial opinions themselves weren't necessarily enough to train the AI because
- 1:03:01what they were trying to do to train the AI was create sort of bite sized questions or
- 1:03:06the equivalent of bite sized questions which essentially are like head notes right you're
- 1:03:10taking you know these long judicial opinions and you're breaking it down into its constituent
- 1:03:15sort of issues and parts and so really it's not like could they use the judicial opinions
- 1:03:20they're the answers yes but in order for them to train the AI to achieve what it wanted
- 1:03:24to achieve there really needed to be some distillation of that and maybe some human
- 1:03:30authorship there and the court said you could have created your own you didn't need to use
- 1:03:34Westlaw as you could have created your own and paid your own authors to do that job well
- 1:03:40I like this question which is could I provide a link to my blog sure I will I will do that
- 1:03:46I mean I can stop sharing and yeah I'm gonna do that because here's my blog I mean this
- 1:03:55is my own that's my own personal blog that I just typed in I don't know if that worked
- 1:04:04and what else are we are we good we do we're at 1105 I see oh Jay is Jay is Jay is hitting
- 1:04:11back surely AI can summarize long judicial opinions I agree with the court maybe I you
- 1:04:17know it's interesting Jay it's a good question like in in 2020 I don't know that that's the
- 1:04:23case I think when they started doing what they were doing there were wasn't sort of off the
- 1:04:29market chat EPT available to do that and I don't know how good chat EPT would go and
- 1:04:34I do I don't know how well chat EPT would be able to create head notes and I think Thompson
- 1:04:38Reuters would argue that it's they do a much better job of it through their human authorship
- 1:04:43and layering and I hope I hope maybe that's true I don't know all right well thanks everyone
- 1:04:48hope you hope you enjoyed it and feel free to reach out if you have questions or anything
- 1:04:52um all right well um I'd like to thank you Jeremy on behalf of the entertainment law
- 1:04:59section of the Verlachal's bar association for your insightful and timely program today
- 1:05:05about AI and fair use thank everyone for joining us and Jeremy's presentation and also Jeremy
- 1:05:13I'd like to accept definitely on behalf of the Verlachal's bar association your offer
- 1:05:18to come back to continue with additional CLE programming I know we would all appreciate
- 1:05:23that and we appreciate your time so thank you you're welcome a lot of folks are asking
- 1:05:27for slides should I I will send them are you able to distribute those out to folks that
- 1:05:32attended is that a thing yes why don't you send us the materials and we will send to
- 1:05:37the attendees all right thank you for that thanks everyone thank you you may be eligible
- 1:05:43for CLE credit in your state visit bhba.org slash podcasts for more information