Latest / CLEs You Actually Want to Hear / Hollywood, Highways, and Generative AI Outputs: What’s Next in AI Law
Transcript
- 0:02All right, hello, welcome to Hollywood Highways
- 0:06and Generative AI Outputs, What's Next in AI
- 0:09Law. I am Naderat Imam, president of the Beverly
- 0:13Hills Bar Association and Beverly Hills Bar Foundation.
- 0:17I'm delighted to be here today to help introduce
- 0:19this program, a partnership between BHBA and
- 0:22the Federal Communications Bar Association. At
- 0:25BHBA, we greatly value opportunities like this
- 0:28to collaborate with other bars, as together we
- 0:31can provide timely, engaging and informative
- 0:33programs for our members and the broader legal
- 0:36community. For those unfamiliar with BHBA, we
- 0:39are approaching our 100 -year anniversary with
- 0:42over 7 ,000 members throughout California and
- 0:45beyond. We are home to the largest entertainment
- 0:48law section of any bar association and have more
- 0:51than 40 other active practice area sections.
- 0:55We averaged more than one new CLE per day and
- 0:59recently won the best CLE provider and best online
- 1:03CLE provider with the best of 2025 awards from
- 1:06The Recorder. We welcome your submissions for
- 1:10topics and speakers that interest our members
- 1:12and those of other great organizations we partner
- 1:16with, like FCBA Today. Getting the added Benefit
- 1:23of specialized expertise from bars like FCBA
- 1:26is excellent for our audience and our collective
- 1:29knowledge. So thank you for attending today.
- 1:32We hope to see you again. I thank you to the
- 1:35FCBA for working with us to put this great panel
- 1:38together. I am sending a particular thank you
- 1:41to Jonathan Cohen, who is a member of both BHBA
- 1:45and FCBA. led to the development of today's content.
- 1:49And now I'll turn the floor over to FCBA's president,
- 1:53Matthew Del Niro. Yeah, thank you, Natura. I'm
- 1:57delighted that FCBA is collaborating with the
- 2:00Beverly Hills Bar Association on this very timely
- 2:03topic of what's next in AI law. It's really an
- 2:06honor for the FCBA to be partnering with a bar
- 2:09association that has such an impressive history
- 2:12and leadership and public service throughout
- 2:15Los Angeles County and beyond. Our own bar, the
- 2:18FCBA, dates back not quite 100 years, maybe a
- 2:22little closer to 90, to the 1930s, just after
- 2:25Congress adopted the Communications Act. And
- 2:28since then, our members have really been at the
- 2:30forefront of legal and policy developments in
- 2:32communications, media, and more recently, tech
- 2:35sectors. And we're fortunate to have many committees
- 2:38and among them the Emerging Issues and Technology
- 2:41Committee, which develops compelling programs
- 2:44and events focused on the most cutting edge issues
- 2:46for our bar. And it's thanks to that committee
- 2:49that the FCBA has joined BHBA today for this
- 2:52event. As you might see, I'm speaking to you
- 2:55here from my office in Washington, DC, where
- 2:58probably the majority of our membership reside.
- 3:01But one thing that we are all very proud of is
- 3:03the FCBA's national reach. And that extends,
- 3:06of course, to Southern California, where we have
- 3:09many active members. One of those members is
- 3:12Jonathan Cohen, a partner at Wilkinson Barker
- 3:15Nowher. Jonathan actually divides his time between
- 3:18Los Angeles and Washington, racking up a lot
- 3:20of frequent flyer miles. And he serves as a co
- 3:23-chair both of our Southern California chapter
- 3:26and of our Emerging Issues in Tech Committee.
- 3:29Jonathan's going to be moderating today's panel,
- 3:31which I'm really looking forward to attending.
- 3:34And I will turn the mic over to him now. Thank
- 3:37you, Matt. And hi, everyone. Thanks for joining
- 3:40us. And thank you also to both the BHBA and the
- 3:44FCBA, the Tech Bar, for agreeing to collaborate
- 3:48on this webinar. We're lucky to have three respected
- 3:52academic scholars to teach us about some of the
- 3:55new and novel legal issues that artificial intelligence
- 3:58is raising and will continue to raise. I'll briefly
- 4:01introduce them, but longer bios are linked in
- 4:05the CLE materials that have been distributed.
- 4:08So joining us are... Jess Myers, an assistant
- 4:11professor at the University of Akron School of
- 4:14Law. She's a recognized expert in online intermediary
- 4:18liability law in the US, who writes and speaks
- 4:21on such issues as online speech in Section 230,
- 4:25intellectual property, content moderation, and
- 4:28cyber crime. Her recent research and scholarship
- 4:31has been centered on generative AI. Also with
- 4:35us, is Mark McCarthy, an adjunct professor at
- 4:39Georgetown University in both the graduate school's
- 4:43communication, culture and technology program
- 4:46and in the school's philosophy department. He
- 4:49teaches courses in technology policy, content
- 4:52moderation for social media, privacy, the ethics
- 4:56of speech and the ethical challenges of AI. He
- 4:59also is a non -resident senior fellow at the
- 5:03Institute for Technology Law and Policy at Georgetown
- 5:06Law Center, as well as a non -resident senior
- 5:09fellow in governance study at the Brookings Institution
- 5:13Center for Technology Innovation. Mark is the
- 5:16author of the 2023 book, Regulating Digital Industries,
- 5:21how public oversight can encourage competition,
- 5:24protect privacy, and ensure free speech. Finally,
- 5:29last but not certainly not least is Eugene Volokh.
- 5:32He is the Thomas M. Seidbell Senior Fellow at
- 5:35the Hoover Institution at Stanford University
- 5:39and the Gary T. Schwartz Distinguished Professor
- 5:42of Law Emeritus at UCLA Law School, where for
- 5:46over 30 years he has taught First Amendment and
- 5:48copyright law, criminal and tort law, and firearms
- 5:52regulation policy. Uh, Eugene has written extensively
- 5:56about law and artificial intelligence since 2019.
- 6:01Welcome to you all. So we have a packed agenda
- 6:04and we're gonna get right into it. Um, each of
- 6:06our panelists will give a talk addressing topics
- 6:09from our agenda and after each we'll have a short
- 6:12discussion amongst. There should be time toward
- 6:15the end for audience questions, so please submit
- 6:18your questions using the Q &A button at the bottom
- 6:21of your Zoom screen. We have a lot of attendees,
- 6:24but we'll do our best to get to your questions.
- 6:27First, I'd like to kick things off with an opening
- 6:30question for all of our panelists. Unlike previous
- 6:34technological leaps, AI enables humans to engage
- 6:38with automated systems that can understand content,
- 6:42exhibit apparent reasoning ability, and respond
- 6:45in ways that feel genuinely conversational rather
- 6:48than purely mechanical or transactional. And
- 6:52AI applications are being adopted at a very high
- 6:55rate. So as a general matter, Do you think that
- 6:58the rapid changes occurring because of AI will
- 7:02make the courts in the future inclined to revisit
- 7:05existing precedents while applying different
- 7:08standards leading potentially to surprising new
- 7:11results? Or are we just stuck with the precedents
- 7:13that we have? And if the former, what do you
- 7:16think that means for attorneys litigating on
- 7:19behalf of clients? Eugene, let me go to you first.
- 7:24Well, AI is going to be about as broad as internet,
- 7:30probably more so, right? Used to be people say,
- 7:33oh, I practice internet law. Do you practice
- 7:36internet law? Now we realize everything is internet
- 7:38law, right? So I think the same thing is true
- 7:43with regard to AI. Obviously, they're a huge
- 7:45copyright law. issues raised by AI, heavily litigated
- 7:48right now. There are some libel law issues. There
- 7:52have been four libel law cases in the US that
- 7:54I know of so far. Then there are questions having
- 8:01to do with alleged physical harms that are created
- 8:04by AI applications. There's a lawsuit pending
- 8:08right now. your CIV character AI having to do
- 8:12with liability for suicide that may have been
- 8:16causally linked to a conversation with kind of
- 8:23a chat bot, a character -based chat bot. In that
- 8:26case, conversation by a minor that might make
- 8:29a difference. And then, of course, there are
- 8:31other problems, right? I hope everybody is aware
- 8:34of the dangers of filing briefs with hallucinations.
- 8:38hallucinated citations. The latest database that
- 8:42I've seen has shown over 300 court decisions
- 8:46throughout the world noting this and sometimes
- 8:49sanctioning lawyers for that. About over half
- 8:51of them, I think, are in the US. But that is
- 8:54just a special case of the broader problem. that
- 8:56there's a lot of error having to do with AI.
- 9:00They can infect our work product. They just can
- 9:02affect the facts of a case. I was just reading
- 9:04a court decision this morning, which has to do
- 9:06with a quite traditional cyber squatting case.
- 9:12But the cyber squatters, what they were doing
- 9:15is they were libeling someone by talking about
- 9:18this lawsuit against them. But it was all fake,
- 9:21like all of the all of the purportedly filed
- 9:26documents on their site where AI generated. Again,
- 9:30it's an old thing. Forgery is a very old thing.
- 9:33But AI makes it easier, makes it perhaps more
- 9:36common, changes the frequency. So it is going
- 9:40to be everywhere. It already is in some measure
- 9:42everywhere. It's going to be even more everywhere.
- 9:46Mark, thoughts on this? So yeah, let me say first,
- 9:53thanks to you for having me on this panel and
- 9:57thanks for the book plug as well. I'm appreciative
- 10:00of that and looking forward to chatting with
- 10:03Eugene and Jess on these issues. So back in 2023,
- 10:12the then chair of the Federal Trade Commission,
- 10:14Lena Kahn, I think she summed up the best attitude
- 10:18toward AI and the law when she observed there
- 10:22is no AI exception to the laws on the books.
- 10:27So are companies using AI to establish creditworthiness?
- 10:34Well, then the Consumer Financial Protection
- 10:36Bureau can and should make sure the companies
- 10:39abide by the fair lending laws. Are the companies
- 10:42using AI to screen workers for employment or
- 10:45promotion? Well, then the Equal Employment Opportunity
- 10:49Commission can and should make sure that these
- 10:52companies don't discriminate against protected
- 10:55classes when they use these new technologies.
- 11:00Are companies using AI to make pitches to consumers
- 11:03about the quality and price of consumer goods?
- 11:06Well, then the FTC can and should make sure that
- 11:09the companies are not engaged in unfair or deceptive
- 11:13acts or practices. when they use these algorithms.
- 11:17Now, how to apply existing law might be unclear,
- 11:21as Eugene noted for the case of copyright law
- 11:23and content liability. But there's no AI immunity.
- 11:28In particular, I don't think there's any AI immunity
- 11:31under section 230, since AI systems generate
- 11:36content rather than distributing the content
- 11:39of others. By the way, lawfare. recently published
- 11:43a useful survey of how current consumer protection
- 11:48laws apply to AI, and that might be worth distributing
- 11:51to the group. So do we need new AI laws? Well,
- 11:56it depends on whether we want to impose new obligations
- 12:00on AI companies. For example, discrimination
- 12:04using AI is illegal, but right now only employers
- 12:09are liable. Do we want the vendors of AI employment
- 12:14tools to be liable as well? Well, then we need
- 12:18a new law. Just anything to add? Yeah, first
- 12:25of all, I agree. Thank you so much for having
- 12:27me. This is a great event. And I am with some
- 12:30really truly fantastic people as well. It's kind
- 12:32of a little starstruck here. So on this topic,
- 12:37I first I want to echo Mark and Eugene. I think
- 12:40I agree entirely. And on the Section 230 note,
- 12:42I will say I wrote an article a couple of years
- 12:45ago when the generative AI stuff was booming.
- 12:47And I had said, of course, Section 230 applies
- 12:49to generative AI. This is now the subject of
- 12:52my research, how generative AI specifically specifically
- 12:54how we treat it under the first amendment and
- 12:56how we think about it in terms of immunities
- 12:57and safe harbors. And I would say today, I think
- 13:00my views have probably changed a little bit just
- 13:03in understanding how the technology works. You
- 13:05know, it's probably not section 230, but I do
- 13:07think we need something that's section 230 like
- 13:09if we're gonna keep having this technology. But
- 13:13going back to the original question here of,
- 13:15you know, where are we with the current law and
- 13:17how does it apply to AI? Well, first, you know,
- 13:19I'll point out AI is a discipline. So AI does
- 13:22not refer to any specific kind of technology.
- 13:24And I think that's really, really important for
- 13:26folks who are on this call and are really thinking
- 13:28about the types of technology that they're engaging
- 13:30with. When we talk about AI, there's many different
- 13:32ways to implement artificial intelligence into
- 13:35different technologies. And those different ways
- 13:37matter tremendously. I'm going to cabin all my
- 13:40remarks to generative AI, so the type of technology
- 13:43that allows us to generate new content. You can
- 13:46think of that as like your chat GPTs, for example,
- 13:49your large language model. your multimodal models.
- 13:52And so there's some interesting questions right
- 13:54now in First Amendment law, I know Eugene is
- 13:56probably going to speak about this as well, about
- 13:58how we treat those types of new communications
- 14:01technologies. And Jonathan, you were asking,
- 14:04are we stuck with current precedent or are things
- 14:06going to change? I think the history of what
- 14:08we've seen with every single new iteration of
- 14:11communication technology tells us that even if
- 14:14we think the current First Amendment precedent
- 14:17should apply, in full scope in the way that we
- 14:22see it apply to newspapers, we see it apply to
- 14:25internet services. There's always been a little
- 14:28blip or mutation, speech mutation along the way
- 14:31for the history of regulating communication devices.
- 14:34And so I think what we're gonna see with generative
- 14:36AI is probably another sort of mutation, especially
- 14:39looking at the way in which the courts are treating
- 14:41this technology. Eugene was talking about the
- 14:43character AI case, for example. There is a big
- 14:46discussion in that case about whether generative
- 14:48AI outputs are even considered speech. And I
- 14:50was stunned by that because to me, of course,
- 14:52they're considered speech. Of course, the way
- 14:54that this technology is built by the providers
- 14:57and developers of these tools, they're making
- 14:58very precise editorial decisions that lead us
- 15:01to these outputs. And yet we have sort of this
- 15:04what Joseph Weissenbaum in 1966 called the Eliza
- 15:08effect, this sort of reaction to chatbots where
- 15:12human beings, they project sort of these emotions
- 15:15onto chatbots and they sort of lose their minds
- 15:18when it becomes when we're in the generative
- 15:20context. And so seeing as where the courts are
- 15:23going now and the way the current legal discourse
- 15:25is about generative AI, I certainly don't think
- 15:28we're going to be stuck with the current precedent.
- 15:30I think we are going to see whatever the next
- 15:33adaptation is to this new brand of communication
- 15:36technologies. And I think the Character AI case
- 15:38is going to be the case to watch for that. Fascinating.
- 15:42Thank you. Well, we've used up 15 minutes of
- 15:47our 90 minutes already. So let's launch into
- 15:51our presentations. And Jess, let me ask you first
- 15:54to bring up your slide deck. And we'll sit and
- 16:03hear you educate us. Awesome, and I'm going to
- 16:05try not to take the full 15. We'll see. And thank
- 16:09you for the Beverly Hills Bar Association for
- 16:11running the slides. You can go ahead and go to
- 16:12the next slide. Appreciate this. So I'm going
- 16:16to be mostly focusing on digital replicas and
- 16:19the publicity rights question of generative AI.
- 16:22And so I know many of you who are on this call,
- 16:24you're probably associated in some way with the
- 16:26entertainment industry. There's a lot of fascinating
- 16:28ways in which generative AI tools are being used
- 16:31in entertainment. And of course, with those fascinating
- 16:34uses, if there comes some... potential liabilities
- 16:36and risks. And so that's what I'm trying to call
- 16:38out in these slides today, just to kind of give
- 16:41an FYI about the law developing in this space.
- 16:43That's especially important because, as you all
- 16:45know, we don't have a federal publicity rights
- 16:48statute on the books yet. We don't even have
- 16:50a federal privacy law statute on the books yet.
- 16:52And so what publicity rights has become is sort
- 16:56of this really wonky mix of privacy and First
- 16:59Amendment and IP, all sort of bleeding together
- 17:03in the different state applications. with the
- 17:05publicity rights law. So if we take a step back
- 17:07for a second, we can think about, well, how is
- 17:10AI being used in the entertainment industry?
- 17:12And we've seen things like de -aging, for example.
- 17:16So there's been lots of instances where AI is
- 17:18used, where you have the actual actor is in the
- 17:21movie, but they may have been de -aged. We've
- 17:24seen what I think folks are most familiar with,
- 17:27they're probably recognize the most, this idea
- 17:29of deepfakes, digital replicas, this idea that
- 17:32you can clone another person. And these AI technologies
- 17:36have gotten really, really good at that. I've
- 17:38seen the voice cloning technology. I'm sure you
- 17:40all might remember, you know, the Biden voice
- 17:42calls that were coming out. We've gotten very,
- 17:44very close, if not almost indistinguishable to
- 17:48actual human voices. We've seen folks use this
- 17:52voice cloning technology in the music space,
- 17:54especially where they are making, they're sort
- 17:57of ghostwriting new music and accrediting it
- 18:00to these performers who they weren't even part
- 18:03of it. first place. So there's lots of different
- 18:05ways that AI is being used. Another big one is
- 18:09resurrecting deceased performers and that has
- 18:11a lot of sort of personal ethical and legal implications
- 18:14as well where either you're resurrecting them
- 18:16in sort of a concert setting or you might be
- 18:18bringing them back to life in a movie video game
- 18:21setting as well. The point is is that with all
- 18:24of the traditional issues that come up in publicity
- 18:26rights It's exactly what Eugene was saying. We
- 18:29still have those exact issues, but now AI is
- 18:32making it easier, it's making it better, and
- 18:35it's also making it more affordable and approachable
- 18:38for the general public as well. A lot of these
- 18:41tools... all of us today can download and use
- 18:44them as well to do a lot of the stuff that really
- 18:45was reserved for Hollywood. So I have some examples,
- 18:49some of these celebrity flashpoints. Scarlett
- 18:51Johansson was a big one. She never actually sued
- 18:53OpenAI, but they're, you know, OpenAI had a sky
- 18:57voice that resembled her. The big one right now
- 19:00is the LeBron James pregnant ditty memes. I can't
- 19:02believe I even said that sentence out loud. That
- 19:06one, LeBron James is a state, or sorry, an essay,
- 19:08LeBron James actually sent a cease and desist
- 19:12letter and we're waiting to see how that sort
- 19:14of plays out, but sent that to the providers
- 19:16of the tools that are allowing people in order
- 19:19to make these sort of memes and strange content.
- 19:23We've seen this with Tom Hanks, we've seen this
- 19:24with other celebrities where Viral advertisements
- 19:28are going around that are, you know, they show
- 19:30the celebrities sort of endorsing a product that
- 19:32they have never endorsed before. They never made
- 19:34the video. Talked a little bit about the Heart
- 19:36on the Sleeve music context. And then another
- 19:39sort of recent SAG -AFTRA went after Epic Games
- 19:43for negotiating a license to keep James Earl
- 19:46Jones' voice for the Darth Vader video game series.
- 19:50But they did not, Epic did not go through SAG
- 19:53-AFTRA. So that ended up being an issue as well.
- 19:55with their collective bargaining agreement. The
- 19:58tools that we have available to deal with these
- 20:01types of deep fake issues, it's gonna be your
- 20:04state publicity rights statutes that are already
- 20:06on the books. But then what we're also seeing
- 20:08are states trying to come up with their own AI
- 20:11specific additions to those publicity rights
- 20:14laws. Now, a lot of them have to do with voice
- 20:16because a lot of the state publicity rights laws
- 20:18don't include voice. It's usually your name,
- 20:21image and likeness. So we're seeing some attempts
- 20:23where it's just to bring voice into this already
- 20:26existing state statute or to just address AI
- 20:29on its own. And that's gonna be things like the
- 20:31Elvis Act. California and New York have some
- 20:33new laws on the books we'll talk about as well.
- 20:36The Federal Lanham Act, of course, is also in
- 20:38play. That's gonna be more narrow uses, though,
- 20:39what we're seeing. That's gonna be specifically
- 20:41in the advertising context. And then we also
- 20:43have the Take It Down Act, which was recently
- 20:45enacted in May. I'll talk a little bit about
- 20:47that in a bit. And the No Fakes Act, which is
- 20:49currently proposed, not enacted, and it... As
- 20:52far as I've heard, as of today, it's probably
- 20:54not going anywhere. But I'll put it on the radar
- 20:56anyways. It's kind of a zombie bill. It comes
- 20:58back every year. And then the other thing to
- 21:02sort of throw out there, we saw in the early
- 21:04days of bringing these IP suits against the generative
- 21:07AI companies, they would sort of mix this artist
- 21:11style claim. And they tried to bring that in
- 21:13as a publicity rights claim. And most of those
- 21:16have been dismissed outright as just trying to
- 21:18end run around copyright. So we're seeing courts
- 21:20to fly in the artistic style, that's for situations
- 21:24where you could ask a generative AI tool to create
- 21:28a comic or art in the style of a famous artist's
- 21:32work or a poem in the style of a famous author's
- 21:35work. Those are being rejected as publicity rights
- 21:37claims. And then the last thing we're seeing
- 21:40as we're seeing these take down these notice
- 21:42and removal regimes start to emerge but now in
- 21:45the publicity rights context and that's what
- 21:46we saw with the take it down act which we'll
- 21:48talk more about but this is the idea that not
- 21:51only are the providers of these generative AI
- 21:53tools potentially liable for enabling people
- 21:56to create these create these images or to sort
- 22:00of violate the publicity rights of other individuals
- 22:02but it's also extending to the internet surfaces
- 22:05that then host these deep fakes as well. We can
- 22:08go to the next slide. This is one example case
- 22:12that's pretty recent as of July 2025, Lerman
- 22:15and Sage versus Lovo. Lovo is a voice clone app.
- 22:19I'm just putting this in there just to demonstrate
- 22:22one of the many cases that has to do with these
- 22:25voice cloning apps. Paul and Linnea, Sage, they're
- 22:28both voice actors. They were asked on Fiverr
- 22:31to contribute their voice. They were told it
- 22:34was for some sort of research project. They actually
- 22:36didn't know that it was gonna be used in the
- 22:38context of being part of this voice cloning app.
- 22:40app. They brought a lawsuit against the voice
- 22:44cloning app, so that's the LOVO app, and they
- 22:46brought it under breach of contract. And then
- 22:48also New York's right of publicity statutes,
- 22:51that's section 50 and 51. And they also brought
- 22:54a copyright claim. The courts decided no on the
- 22:58Lanham Act claims, said that that is reserved
- 23:00for, they said first, it's no actionable consumer
- 23:03confusion in that case, and reserved in advertising
- 23:06context. So we see the Lanham Act not being successful.
- 23:09But the publicity rights and the contract claims
- 23:11were allowed to stand. So we are still currently
- 23:14watching this case. I think this is one of very
- 23:18few where the publicity rights actions are still
- 23:21in play. And then most of the federal and trademark
- 23:25theories were also dismissed as well. We can
- 23:28go ahead and go to the next slide. So here's
- 23:32some of the laws slash guidance. currently on
- 23:35the books or pending. The big one to keep in
- 23:38mind is the Take It Down Act. So that was enacted
- 23:40in May of this year. There's two parts to it.
- 23:44The first part actually criminalizes the creation
- 23:48and publication of deep fakes that have to do
- 23:52with non -consensual intimate imagery. Oftentimes
- 23:55this is sometimes referred to as revenge porn.
- 23:59And that was the intent behind the Take It Down
- 24:01Act when it was initially proposed. It was initially
- 24:05proposed to deal with some of the issues that
- 24:07are taking place in schools today where kids
- 24:09are using these quote unquote undressing apps,
- 24:12undressing AI apps, where you can provide a real
- 24:15image of somebody and then the AI can then sort
- 24:18of fill in the rest and those were going around.
- 24:20And again, it's in this sort of revenge pornography
- 24:23umbrella of content. Some of the problems though,
- 24:27there is no carve out for First Amendment parody
- 24:29commentary. There's some vagueness around what
- 24:32is actually considered an intimate image and
- 24:34the reason I have the recent screenshot from
- 24:37the first episode of South Park, I actually wrote
- 24:39an article a couple months ago where I sort of
- 24:43suggested there The Take It Down Act, given how
- 24:46broad it is, could actually apply in political
- 24:49commentary situations like South Park. I spared
- 24:53you all the actual image that I think is potentially
- 24:56at issue here. But in the very first episode
- 24:58of South Park this year, the creators actually
- 25:00used a deep fake of Donald Trump. Nude, they
- 25:04were kind of doing like an emperor's, is nude,
- 25:07has no clothes, where they actually showed genitalia,
- 25:11AI created genitalia. sort of this open question
- 25:14about, well, in that context where the Take It
- 25:17Down Act does criminalize the creation and sharing
- 25:20of intimate imagery of of known people, of known
- 25:26individuals, could the Takedown Act apply in
- 25:29that context? So it's kind of a warning or a
- 25:33heads up to those who are even thinking about
- 25:36political commentary or thinking about doing
- 25:39something like South Park did. There are these
- 25:41sort of criminal penalties that are currently
- 25:43on the books. The other half of Takedown actually
- 25:45refers to the online services, so the social
- 25:47media companies. The social media, this creates
- 25:50a notice and takedown removal scheme where it's
- 25:55extremely broad. The folks who are the subject
- 25:59of deep fakes or intimate imagery that's being
- 26:01shared can contact a social media company. Social
- 26:04media company has 48 hours to remove that content.
- 26:07It's pretty strict. There is no sort of question
- 26:11about whether it's First Amendment protected,
- 26:12fair use, et cetera. It's 48 hours to take it
- 26:15down or they're held liable. So it's currently
- 26:17on the books, and we haven't seen any actions
- 26:19under it yet. The No Fakes Act is currently pending,
- 26:22but I don't think anything is going to happen
- 26:24with it. Knock on wood, watch something happens
- 26:26tomorrow. This is an attempt at creating a federal
- 26:30digital replica right. Remember, we talked about
- 26:32how we don't have a federal publicity rights
- 26:34law. So this is an attempt for us to get a federal
- 26:38publicity rights law that is aimed at AI, and
- 26:42it would also restrict the creation and publication
- 26:47of deep fake content. And it also applies to
- 26:51the internet services as well, creates a similar
- 26:53DMCA style takedown approach. Again, not sure
- 26:56that's going anywhere. I included the FTC impersonation
- 26:58role because there were discussions about whether
- 27:01the FTC was going to extend their existing impersonation
- 27:05role, which only applies to government entities
- 27:07and businesses. There was a discussion earlier
- 27:09this year about whether they would extend it
- 27:10to individuals. They had a hearing, I think,
- 27:12in June, and the decision was not to extend it
- 27:14to individuals. So that is only in play for impersonation
- 27:18of government entities and businesses. And you
- 27:21can go to the next slide. Jonathan, how am I
- 27:23on time? You're good. You may have to forego
- 27:27the sort of follow -up question, but that's OK.
- 27:30This is great. OK. Awesome. Great. This is a
- 27:35little bit older. It's from 2024, the US Copyright
- 27:38Office. They went through a bunch of series for,
- 27:40it was mostly having to do with how copyright
- 27:43or use and infringement applies in the context
- 27:45of generative AI. They also addressed publicity
- 27:48rights in the first part of their report. And
- 27:52so they essentially talked about, we don't have
- 27:54federal publicity rights laws. And there was
- 27:56a suggestion about maybe we need a federal publicity
- 27:59rights law on the book. It talked a little bit
- 28:01about how currently the only thing we have in
- 28:03play are the different state laws, and it discusses
- 28:05some of the limits on those state laws. For example,
- 28:08a lot of state laws have some limitations on
- 28:10post -mortem rights. When somebody is dead, how
- 28:13did their name image likeness rights sort of
- 28:16extend to the state? And so some are kind of
- 28:19the gold standard, like California when it comes
- 28:21to post -mortem rights. And then there's others
- 28:23that may not even contemplate post -mortem rights.
- 28:25And then, of course, other states. It has to
- 28:27be used in a commercial context. And in some
- 28:29states, it can be used in either commercial or
- 28:31other contexts as well. So the Copyright Office
- 28:34was sort of highlighting the limits, but also
- 28:36noting that there are state laws that are in
- 28:38the books to address some of the publicity rights
- 28:40issues. They include existing laws do not provide
- 28:43sufficient legal redress for those unauthorized
- 28:46digital replicas. And they are recommending new
- 28:49federal laws. So the Copyright Office is sort
- 28:51of behind a lot of the discussions with no fakes,
- 28:53for example. The entertainment industry is pro
- 28:57no fakes and I believe the internet industry,
- 29:00the internet services are also behind no fakes
- 29:02as well. And you can go to the next slide. Here's
- 29:07the Elvis Act. Now this has been in play for
- 29:09a while now. It went into effect July 2024. As
- 29:13of my research yesterday, I don't think I've
- 29:14seen any recent actions under the Elvis Act.
- 29:18This is Tennessee's approach to ensuring that
- 29:21voice gets protected in their already existing
- 29:23publicity rights law. This was brought and favored
- 29:27on behalf of many of the music artists in Tennessee.
- 29:30So this was a very big deal in that state. This
- 29:33covers digital and AI generated reproductions.
- 29:36faith and synthetic media. I will note this doesn't
- 29:39get talked about a lot, but there is a part of
- 29:41the bill that also extends third party liability
- 29:43to the providers of generative AI tools that
- 29:46enables people to make voice clones and deepfakes.
- 29:50That one is not talked about as frequently, but
- 29:52just an FYI, again, if you're representing a
- 29:53generative AI company, you too can also be in
- 29:56scope for this. prohibits the exploitation of
- 30:00likeness, but mostly voice without consent. And
- 30:03the Elvis Act is kind of what set the precedent
- 30:05for all of these other states that have been
- 30:08proposing and enacting their own voice and generative
- 30:11AI publicity rights laws as well. So Elvis kind
- 30:16of kicked it off. For remedies and penalties,
- 30:19there is a private right of action. There's also
- 30:22a third party right of action too, which is fascinating.
- 30:25It kind of goes beyond what we see in publicity
- 30:27rights. existing publicity rights statutes, but
- 30:29record labels can also now bring suits on behalf
- 30:32of the individuals that they represent as well.
- 30:36And it's a class A misdemeanor. So the Elvis
- 30:39Act in Tennessee are taking digital replicas
- 30:42specifically having to do with voice very seriously.
- 30:45And again, this is kind of our first approach
- 30:47to seeing an AI specific publicity rights law
- 30:50in the books. Next slide. We also have New York.
- 30:56So this went into effect January 1st, 2025. New
- 30:59York's digital replica law is very similar. I
- 31:02think it mirrors the California law. This specifically
- 31:05has to do with contracting with the use of artists,
- 31:11actors, and production companies making digital
- 31:15replicas of those artists and actors. So for
- 31:19example, in this law, it's going to void any
- 31:21contracts that meet the following of these conditions.
- 31:23and the California law is very similar too. So
- 31:26we're going to, if the contract is allowing for
- 31:29the creation and use of a digital replica of
- 31:31an individual's voice or likeness in place of
- 31:34work the individual would otherwise have performed.
- 31:36This mimics by the way the SAG -AFTRA. collective
- 31:40bargaining agreement around AI in general, then
- 31:43we will potentially void the contract for that
- 31:46as well. The policy reason behind this is they
- 31:49don't want to ban the use of AI in entertainment,
- 31:52but they don't want you take a single face scan
- 31:57or a body scan of an actor, and then you just
- 32:00use that in place of the actor for anything going
- 32:03forward. The New York law is not going to permit
- 32:06that as part of the contract. If the contract
- 32:10also does not include a reasonably specific description
- 32:13of the intended use of the digital replica, that's
- 32:16also going to be a problem. So you can't just
- 32:18articulate in your contract that you are going
- 32:21to make digital replicas or you're going to take
- 32:23a scan of the actor or the performer. You need
- 32:26to specifically say how it's going to be used
- 32:28and the use cases that are going to follow. You
- 32:30can't leave it up to future uses, for example.
- 32:33And then the last thing, if it was negotiated
- 32:35with an individual who was not represented by
- 32:37legal counsel or by a labor union, So this kind
- 32:39of gets at the SAG -AFTRA epic games controversy
- 32:43where Epic just went directly to James Earl Jones'
- 32:47estate and didn't go through SAG -AFTRA. The
- 32:49same thing is happening here with this New York
- 32:51law. If you're going to include AI and digital
- 32:54replicas in your entertainment contracts, this
- 32:57needs to be negotiated with individuals counsel
- 32:59or with a labor union as well. And you can go
- 33:03to the next slide. These are things that are
- 33:06going on in California so California had a whole
- 33:09spate of laws regarding generative AI or regarding
- 33:12AI use and entertainment. Exactly. It pretty
- 33:17much codifies the rules that are already in place
- 33:20with Sagastra, and also it's very similar to
- 33:23the New York bill. So again, if you are contracting
- 33:26with performers in California, not only do you
- 33:29have to deal with the labor union rules around
- 33:31using AI and how you use digital replicas of
- 33:34those actors, but now we also have California
- 33:37AB 2602 on the books, which also requires you,
- 33:41even for non -labor union performers, also requires
- 33:45your contract. to be very clear about how you're
- 33:47using digital replicas and you cannot replace
- 33:49an actual person for future uses as well. AB
- 33:531836 just expands California's already very extensive
- 33:56and existing publicity rights statutes on the
- 33:58book. It prohibits the use of a deceased personality's
- 34:02voice or likeness without consent from their
- 34:04estate or family. This is again trying to get
- 34:06at some of the controversies around resurrecting
- 34:10people and musicians for some of these like future,
- 34:14these like holographic concerts and in films
- 34:18as well. So you will, you now need to have, you
- 34:21know, express consent from the estate in order
- 34:23to do that resurrection with AI. And the last
- 34:27one here that I flagged, and there's a bunch
- 34:28more, by the way, I'm just picking the ones that
- 34:29I think are probably the most on point for entertainment.
- 34:33AV2905, that requires disclosure when AI -generated
- 34:36synthetic voices are used in telemarketing calls.
- 34:39So again, if you're gonna be placing telemarketing
- 34:41calls in California, we have a lot of disclosure
- 34:42laws around the use of bots. I will also just
- 34:46flag, it's not on the slide, but California also,
- 34:48and I believe New York, and I think New York
- 34:50is in litigation with X on this, but there's
- 34:52also some election deepfake related laws as well,
- 34:54so. when you're blurring the line between entertainment
- 34:57and potentially politics, you could run up into
- 35:01some of the election deepfake laws too. Next
- 35:03slide. I went ahead and put this together. I
- 35:08won't read them all out because you all can read
- 35:10and you'll have access to these slides. But just
- 35:12from my own research and the way I've been thinking
- 35:14about things, I kind of put together what I call
- 35:16quote unquote safe use cases. This is not legal
- 35:19advice, by the way, so don't come after me if
- 35:20you get sued. But sort of your safer cases and
- 35:23then some of the more risky cases. And I will
- 35:26say that I think of the most risky side, that's
- 35:29going to be these entertainment contracts. There's
- 35:31a lot of them out there where there have been
- 35:33specific issues where you're either running up
- 35:34against Sagastra, which as many of you know on
- 35:37this call, they are very, very, very brutally
- 35:40serious about the use of AI. It is not something
- 35:42that's favored in the labor unions. So even if
- 35:44you're not seeing a law in the books, if you're
- 35:46dealing with someone who's within a collective
- 35:47bargaining agreement, you'd be very careful about
- 35:49their rules. They're very strict about it. Anything
- 35:52where you're dealing with an unauthorized voice
- 35:54clone, you're dealing with unauthorized replicas
- 35:56of performers, obviously going to be a problem.
- 36:00Even if they do consent, you still might still
- 36:02be in a collective bargaining agreement issue.
- 36:05Commercial advertisements have always been risky
- 36:07before AI, now currently with AI, so you would
- 36:10not want to deal with that. You not want to use
- 36:13somebody who has not consented specifically,
- 36:16especially actors, well -known actors. AI -generated
- 36:20pornographic deepfakes are really the, that's
- 36:23one of the big use cases both federally and in
- 36:25the states is they're really going after pornography.
- 36:27They're going after NCII and specifically CSAM
- 36:31with children, so I would say that's kind of
- 36:33on the riskier side. On the less riskier side,
- 36:37of course, if you've got consent from the estate,
- 36:39you've got consent from your labor unions, then
- 36:43using it to de -age folks is kind of the golden
- 36:47standard right now, better use case. You're not
- 36:50replacing the actor. You're not replacing future
- 36:53actors. Generative VFX for stunts, dubbing, translations,
- 36:57that's all. kind of above bar as well. And then
- 37:01I have parody satire and non -commercial commentary,
- 37:04but I put a big maybe because have fun litigating
- 37:06it. And I think that's probably the end. Yes.
- 37:11I think that's your last slide. Thank you so
- 37:14much, Jess. That was jam -packed. I think we're
- 37:18going to just move directly into Eugene's presentation.
- 37:23So Eugene, you're going to be talking about liability
- 37:25issues relating to AI outputs. So take it away.
- 37:31I'll be happy to. So I am going to do, in a sense,
- 37:35the opposite of what Jess was just doing. I'm
- 37:40going to focus on one major thing and drill down
- 37:44pretty deeply in it. Justice presentation is
- 37:49in many ways much more useful because this thing,
- 37:54well, it's a potentially major topic is something
- 37:55that's only led to about four cases so far, but
- 37:57I think people will find it interesting. And
- 38:00it may be that there will be more of these coming.
- 38:03So all right. All right. I think I should be
- 38:06switching. I am in slideshow, but somehow I'm
- 38:09in presenter format. Is that right? So right
- 38:12now, we see the PowerPoint application itself.
- 38:15We don't necessarily see the slideshow or presenter
- 38:17view. OK. So how about from beginning? How's
- 38:21that? Is this better? Can you see? Yes, that
- 38:22is better. Fabulous. So the name of this is borrowed
- 38:26from an article I wrote two years ago, Large
- 38:28Libel Models. Of course, we know that AI hallucinates.
- 38:33Well, what if hallucinates about particular people?
- 38:36And what if hallucinates things about them that
- 38:38are not just hallucinating, not just false, but
- 38:42potentially damaging to their reputation? Here
- 38:45is one example from a case that was filed, but
- 38:48actually recently settled, obviously. We don't
- 38:50know. on what terms? Starbuck v Meta Platforms.
- 38:54Robbie Starbuck is a prominent conservative political
- 38:57activist. And in his complaint, he says, look,
- 39:02I see this post on Axe where this third party,
- 39:07I don't know him, is saying about me that I am
- 39:14that I was involved in January 6th riot, that
- 39:17I'm linked to the QAnon theory, that I'm anti
- 39:20-vaccine. It's all false. But he got it from
- 39:23Metta, that that is what Metta is saying about
- 39:26me. And in fact, there was evidence that that's
- 39:28what was happening. Eventually, Metta stopped
- 39:30saying that in its text output. But then the
- 39:34complaint says, their voice output began to generate
- 39:40the same things. So again, this is not like garbage
- 39:45in, garbage out, that there's something somewhere
- 39:49online that contained this false information
- 39:52and metas faithfully parroting it. It's gold
- 39:55in, garbage out, the training data, the online
- 39:57data may have been quite accurate, but the algorithm
- 40:01produced false things from it. So if you think
- 40:06about it, if this were a newspaper that published
- 40:08this, it could be a libel claim, right? False,
- 40:12factual assertions. Probably damaging to reputation,
- 40:17participating in a riot as a crime. And then,
- 40:22of course, there'd be questions. What was the
- 40:24mental state? What mental state is required?
- 40:26He's probably a public figure, so this would
- 40:28require so -called actual malice, basically knowing
- 40:31a reckless falsehood. If he were a private figure,
- 40:35maybe negligence would be enough. Well, how does
- 40:36this play out with AI? My answer, I think, is
- 40:39pretty much the same as with anything else. Only
- 40:42pretty much. There are some interesting possible
- 40:44differences. So this case is settled. Let me
- 40:48give you another example. This is from the, as
- 40:50best I can tell, it's still currently pending.
- 40:53I last looked into it about a month or two ago
- 40:55before the Norwegian data protection authorities.
- 40:58They're the ones who apparently in Europe are
- 41:00in charge of hearing these cases. There's this
- 41:03question, who is Arve Hjalmar Holmen? And Chad
- 41:08GPT says he is accused and convicted of murdering
- 41:11his two sons and attempting to murder his third
- 41:13son. I have good news. Everybody's alive and
- 41:16well. Again, this was just something that was
- 41:21just made up about Holman. Totally made up. There
- 41:25are a couple of earlier cases, one of which was
- 41:28dismissed in part because of very peculiar facts,
- 41:30where it was pretty clear that from almost the
- 41:33get -go, the person posing the queries understood
- 41:36this was a hallucination, so no real damage to
- 41:38reputation. Another one was then shunted off
- 41:41into arbitration, where presumably it's going
- 41:44to wither. But there's a fourth one that I don't
- 41:47have slides on in part because it's not quite
- 41:49as visual, but basically there's this company
- 41:52I think called LTL in Minnesota, and they also
- 41:58operate under Wolf River Electric. They install
- 42:02solar panels. So apparently they say if you enter
- 42:05Wolf River Electric, Google with an auto complete
- 42:09to lawsuit Wolf River electric lawsuit and then
- 42:13would report in that AI summary paragraph Generated
- 42:17up on top that they were being sued by the Minnesota
- 42:19AG for for various Violation of consumer protection
- 42:26laws. It's false. They were never sued nor had
- 42:29they been accused of being sued There was apparently
- 42:31a newspaper article which talked about a lawsuit
- 42:33about somebody else and then they were quoted
- 42:37But there was no claim in that article that they
- 42:39were sued. There is this hallucination. Again,
- 42:42by the way, in some respects not terribly new.
- 42:45People misread things, and things get mushed
- 42:50together in their heads, and then they pass them
- 42:51along. Sometimes it happens with reporters. Not
- 42:54uncommon, for example, with there being libel
- 42:57lawsuits where there's a case of mistake in identity,
- 43:00where a reporter thinks somebody is involved
- 43:05in a case where it's really somebody else with
- 43:07a similar name or some connection. But in any
- 43:09event, that case is pending, unlike the other
- 43:12ones. The plaintiff claims they have real information,
- 43:17real evidence, that they lost business. That
- 43:21there were people who canceled contracts with
- 43:22them saying it's because Minnesota AG is suing,
- 43:25or people said, you know, canceled kind of negotiations
- 43:28with them. And they say it's millions of dollars.
- 43:31It's important because, just to foreshadow a
- 43:34bit, as we know, in libel cases, even brought
- 43:36by private figures when it's on a matter of public
- 43:38concern, you can recover proven compensatory
- 43:42damages on a showing of pure negligence. Whereas
- 43:46for presumed or punitive damages, even for private
- 43:50figures, you still have to show knowing the reckless
- 43:55falsehood. So here they're saying we're a private
- 43:57figure, we're just a local small business, and
- 44:00we can show damages. So that'll be a case to
- 44:04watch. So here's one other one. This is the one
- 44:07that went off into arbitration. Basically, there
- 44:10are these allegations about Jeffrey, excuse me,
- 44:12there's this perfectly correct statement about
- 44:14a man named Jeffrey Battle. Then there's another
- 44:16statement about a different man named Jeffrey
- 44:18Battle, slightly differently spelled, but doesn't
- 44:20much matter. In fact, everything is correct except
- 44:23the word however. and probably the implication
- 44:25from the word battle that it's the same battle.
- 44:28But that's plenty to make for a libel claim.
- 44:30Again, fortunately for Microsoft, apparently
- 44:32they had an arbitration agreement that led the
- 44:36case to be sent off there. So how would the law
- 44:38analyze that? I'm going to skip Walters. That's
- 44:41the one that was dropped. So one way of thinking
- 44:43about it, imagine just to simplify things a bit,
- 44:45an analogy. Check BKG, background checker. run
- 44:49by a company called Open Records. Basically,
- 44:51CheckPKG takes a name, reviews court records,
- 44:54provides a list of the criminal and civil cases
- 44:57in which the person has been found liable. And
- 45:00let's say it acknowledges the information may
- 45:01be erroneous, but also touts how good a job it
- 45:03generally does. Someone uses CheckPKG, search
- 45:07for Joe Schmoe, gets a response, criminal conviction,
- 45:12civil liability. But it turns out that That's
- 45:15just not true. That's that's just an error in
- 45:17the algorithm. How do we analyze this? Well,
- 45:20we'd say open records is publishing false and
- 45:22reputation damaging information. They may not
- 45:24be publishing in the sense of placing it on the
- 45:26web, but for libel law purposes, communicating
- 45:29it to one person counts as publication. Doesn't
- 45:33matter that it's through a computer program.
- 45:34I'm pretty sure I mean, we can talk some more
- 45:37about that. But It's responsible for what it
- 45:41does. It's just like with self -driving car.
- 45:45If you get hit by a Waymo, Google can't say,
- 45:48no, no, we didn't do it. The car did it, right?
- 45:51They let the cars loose. They programmed the
- 45:54cars. They're going to be responsible, subject
- 45:56to interesting questions having to do the mens
- 45:58rea. But they can't just sort of shift the blame
- 46:02to an inanimate object. It's not protected under
- 46:05Section 230. I think we all agree on this. And
- 46:08I think Mark captured the point very well that
- 46:12generative AI is generative. It's the one that
- 46:16generates things. Section 230 means you're not
- 46:19liable for things somebody else generates. But
- 46:22here, just to take the open records example,
- 46:25you look at any of the other examples, the output
- 46:28was in fact generated by the defendant. Noting
- 46:30the information may be erroneous. I think we'd
- 46:33say it isn't enough. We'd say, OK, sure, you
- 46:35might note it. But the only reason people are
- 46:37using your software is because they think it
- 46:39may very well be accurate. You're making money
- 46:41from the possibility that it's accurate. A mere
- 46:43disclaimer is not sufficient. We'll get to that
- 46:45in a moment. Now, it knows that its software
- 46:50is outputting false statements. I got the name
- 46:52wrong, about Schmo. And if Schmo. alerts it to
- 46:58that, and then they keep outputting it. Maybe
- 47:00you could say that is knowing a reckless falsehood
- 47:03on the part of the creators of the software,
- 47:06or of the company that's operating the site.
- 47:09And it might be liable in a negligent design
- 47:11theory as well, if Shmo is a private figure.
- 47:15And again, if Shmo can prove damages. So I'm
- 47:18going to skip the section 230 immunity, skip
- 47:20the publication requirement. Happy to get back
- 47:22to that if people have more detailed questions.
- 47:24Let me mention a bit more about the inadequacy
- 47:26of disclaimers. The law has long recognized there's
- 47:30liability for repeating a slanderous charge,
- 47:34even though identifying the source or indicating
- 47:38it is merely a rumor. That's well -settled by
- 47:41the law. Another way of thinking about it is
- 47:43we do sometimes use the disclaimers in traditional
- 47:47defamation. Rumor has it, that's a form of disclaimer.
- 47:50It's not disclaimer colon rumor has it, but still
- 47:52say rumor has it that so -and -so committed this
- 47:56crime is flagging that it may be false, although
- 47:59it may be true, right? We know that some rumors
- 48:01are false, but we also know some rumors are true.
- 48:04Spreading that rumor is sufficient for liability.
- 48:11The republisher is subject to liability. In fact,
- 48:14even though he expressly states he doesn't believe
- 48:16the statement that he repeats to be true. I mean,
- 48:19that's from the restatement. I think that may
- 48:21be going too far. Maybe there is reason to cut
- 48:25back on that, although, again, there is case
- 48:26law that supports that. But at the very least,
- 48:28if all you're saying is, I might be mistaken,
- 48:31but here is this thing that sounds like a factual
- 48:33assertion, liability. And just to quote the GPT,
- 48:40or OpenAI's kind of fact sheet on GPT, or not
- 48:44even fact sheet, it's a long, pretty detailed
- 48:46analysis of GPT that I put out in early 2023.
- 48:50I think they're quite right to say hallucinations
- 48:52can become more dangerous as models become more
- 48:54truthful. As users build trust in the model when
- 48:57it provides truthful information in areas where
- 49:00they have some familiarity. I think that's an
- 49:03important point. And I think it's especially
- 49:05so. I mean, I wrote the article when you had
- 49:07to go to chat GPT and enter a query there. Now
- 49:14Google automatically outputs it at the very top
- 49:17of every search result. That sounds like kind
- 49:20of a... Not quite anti -disclaimer, but that
- 49:23would counteract many a disclaimer. If you're
- 49:26so confident in the general reliability of your
- 49:29output that you automatically put it out there,
- 49:32supposedly for the benefit of your users and
- 49:35therefore of yourself, it becomes much harder
- 49:38to say when you're sued, oh, no, no, no, no,
- 49:40everybody knows this is highly unreliable. If
- 49:43it's so highly unreliable, why do you output
- 49:45it so often? So then, of course, So I'm sorry.
- 49:50So all of this is applying normal libel law and
- 49:53normal First Amendment principles. And the premise
- 49:56is there is First Amendment protection for AI
- 49:59output, but there's an exception for defamation
- 50:03that applies for AI output just as much as for
- 50:06ordinary human output. But some might say, well,
- 50:11maybe AI output shouldn't be constitutionally
- 50:14protected, in which case maybe we should allow
- 50:16more liability for AI than for traditional materials.
- 50:21I don't think so. So this is from an article
- 50:24that Mark Lemley, Peter Henderson, and I co -wrote
- 50:28where we say, generally speaking, AI output is
- 50:30constitutionally protected. It's true. It's not
- 50:33the direct product of a human mind. If all you
- 50:35cared about was self -expression, maybe you might
- 50:37say that's not enough. But it's valuable to readers,
- 50:41contributes to their search for truth and the
- 50:43like. And the law has generally recognized that
- 50:45free speech is protected because of the rights
- 50:47of listeners as well as of the speakers. It lets
- 50:51readers gather information, compare cameras.
- 50:53There's all this caseload that says you have
- 50:55the right to photograph things in public places.
- 50:58And the government can't impose impose restrictions
- 51:04and impose liability for that, unless it fits
- 51:06within some exception. Why? Because it allows
- 51:10you to gather information for your own purposes,
- 51:14but also to then further retransmit it. That's
- 51:16true of AI output. Then let's readers create
- 51:19their own speech. Just to take the example of
- 51:21Garcia, where again, there may be a special complication
- 51:24because this is a minor, interesting question.
- 51:27We can talk about that later. The whole point
- 51:31of these chatbots is precisely that by chatbot,
- 51:34I'm sorry, I mean specifically not just something
- 51:36like chat GPT, but something where there's a
- 51:39character that the person interacts with and
- 51:41has a conversation and engaging conversation
- 51:46with is precisely so readers can create their
- 51:48own speech. In this situation, they create speech
- 51:52to them by the AI, but based on their input.
- 51:57And sometimes, of course, people use it to create
- 51:59other speech. Sometimes. in ways we academics
- 52:02don't much like, such as when our students use
- 52:04AI to create speech that they really should be
- 52:06creating on their own if they want to grade for
- 52:09it. But more broadly, there are all of these
- 52:11perfectly legitimate reasons to use it, basically,
- 52:19if you want to come up with a post or an op -ed
- 52:24or whatever. Presumably, you'd edit it some,
- 52:26as well as prompting, but it is valuable to you.
- 52:30Of course, it's also shaped by AI companies'
- 52:33own design and training decisions, sometimes
- 52:35quite in overtly ideological ways. So I think
- 52:39there may be some rights of speakers there, but
- 52:41there's certainly rights of users. Here's one
- 52:43way of thinking about it. If AI output is just
- 52:45unprotected, period, then presumably Congress
- 52:48could pass a law saying AI cannot output anything
- 52:51that is anti -government, or anti -police, or
- 52:55racist, or anti -Israel, or anti -Palestinian,
- 52:59or what have you. I mean, after all, if there's
- 53:01no First Amendment protection, then even viewpoint
- 53:03-based restrictions would be permissible. I think
- 53:05our reaction is, that can't be right. Why not?
- 53:07Because that would allow the government to really
- 53:09influence public debate. It would interfere with
- 53:11listeners' rights to hear all sorts of views.
- 53:14I think that that shows that it's constitutionally
- 53:17protected. Now, I suppose you could say, well,
- 53:19it's protected, but less protected or differently
- 53:21protected. I just don't think that's right. I
- 53:23think, for all these reasons, it should get full
- 53:25protection, but still subject to exceptions,
- 53:28such as for libel. There's more to say. One thing
- 53:32I do want to mention is if you have a knowing
- 53:36or reckless falsehood standard, for example,
- 53:38for statements by public figures, I do think
- 53:41it leads to a kind of notice and blocking type
- 53:43requirement, that once the company is aware that
- 53:48false statements are being said about someone,
- 53:50it may need to take reasonable steps to try to
- 53:52stop the distribution of that, like having some
- 53:54post processor that sort of removes things that
- 53:58it knows to be false statements that are being
- 54:00generated. For negligence, I think the best analogy,
- 54:04it's only an analogy, but I think it's a good
- 54:05one, is to design defect law for products. Again,
- 54:10like with a self -driving car. It's not strict
- 54:13liability. It's a liability for unreasonable
- 54:16design, but that's very similar to traditional
- 54:19liable negligence. So that's the way I think
- 54:22things will play out. Let me just mention also
- 54:26this might play out not just for libel but for
- 54:29false light, disclosure of private facts, similar
- 54:31questions come up with regard to things like
- 54:34false statements likely to lead to injury, maybe
- 54:37accurate statements likely to facilitate crime
- 54:39because they reveal certain information about
- 54:42people. So in principle, all of these things
- 54:45could apply to the other torts. So far, the cases
- 54:47we've seen involve defamation. So with that,
- 54:50I think I turn things over to Mark. Well, actually,
- 54:54let me, if I may, just ask you a question here.
- 54:58As I'm sure you know and probably many in our
- 55:00audience know, at least one Supreme Court justice,
- 55:04Justice Thomas, has called for the court to revisit
- 55:08the actual malice standard from New York Times
- 55:10v. Sullivan. He says this at practically every
- 55:16opportunity. If the court were to lower the bar
- 55:22for libel suits brought by public figures, how
- 55:29might that play out in the AI context and what
- 55:32do you think it would mean for practitioners,
- 55:35attorneys representing clients in that area?
- 55:38Sure. So by the way, it's not just Justice Thomas.
- 55:41Justice Gorsuch has also expressed skepticism,
- 55:44not always, but also expressed skepticism about
- 55:47New York Times v. Sullivan. Of course, back in
- 55:50the day, famously, the major adversary of New
- 55:53York Times v. Sullivan and other libel decisions
- 55:55was Justice White, the centrist justice. Then
- 55:59Professor Kagan, back in the day, also expressed
- 56:03some skepticism about New York Times v. Sullivan,
- 56:06although I don't know that that means that Justice
- 56:09Kagan would want to reverse it. So if New York
- 56:12Times v. Sullivan is reversed, then probably
- 56:15what would happen is we probably wouldn't get
- 56:16back to strict liability. We'd probably have
- 56:19a negligence -type regime, kind of like the one
- 56:21for private figures, except... it would also
- 56:25apply to public figures. So a government official
- 56:27or a prominent person like Starbuck, for example,
- 56:31he was the one of those four plaintiffs that
- 56:33was relatively prominent, or maybe also Walters,
- 56:36the original one, actually the judge did say
- 56:37he was a public figure because he was an activist,
- 56:40would also be able to recover on a negligence
- 56:42basis. So long, at least, as they can show compensatory
- 56:46damages. Of course, if New York Times v. Sullivan
- 56:49gets rejected, then other cases might too, like
- 56:52Gertz v. Robert Welch. That might suggest that
- 56:55presumed and punitive damages might also be available
- 56:58on a negligent basis. That would of course be,
- 57:01the whole point would be, that that would be
- 57:03true with regard to newspapers and such. I think
- 57:05it would also be true with regard to AI. To be
- 57:09sure, I should just stress on the one hand, I'm
- 57:10glad you were listening to this talk. I do think
- 57:13it's interesting and important, but we're talking
- 57:15four cases. We're talking four cases so far over
- 57:18basically three years worth of. of advanced LLMs,
- 57:24let's say, being publicly available. Maybe it'll
- 57:27increase, but the one thing that I can say for
- 57:30sure is whatever you might think about the possible
- 57:34danger that relaxing Sullivan and Gertz and such
- 57:37would create for newspapers, maybe kind of cause
- 57:44them to really very sharply shy away from things
- 57:46for fear of bankruptcy, I doubt that these are
- 57:50going to be existential threats to AI companies.
- 57:55Not to the big ones, maybe smaller ones, but
- 57:57smaller ones would also presumably be generating
- 58:00a lot less stuff and therefore be less likely
- 58:02to get sued, at least for defamation. So I mean,
- 58:06we can have a debate about whether Justice Thomas
- 58:10and Gorsuch were right or whether Justice Brennan
- 58:12was right back in New York Times v. Sullivan
- 58:15and how it should play out. I do think that as
- 58:17a practical matter, that is probably a lot more
- 58:20of an issue for newspapers and other news outlets
- 58:28than it is for AI companies. Any of our other
- 58:33panelists care to weigh in on this? Then we will
- 58:39move to Mark. Mark, you're going to talk a little
- 58:42bit about the regulatory environment. Note that
- 58:48Mark recently wrote a piece for Brookings on
- 58:53the standards that should apply to liability
- 58:56for self -driving cars. Feel free to read that
- 59:00in if you like also as. out here in Los Angeles,
- 59:04you see Waymo's all the time. And I understand
- 59:07they're coming to DC too. Well, they're prowling
- 59:10around DC, but they're still in mapping mode.
- 59:14And so they're not wandering around without a
- 59:18driver, not yet. But let me start with some remarks
- 59:22about the AI liability for self -driving cars,
- 59:26because as you say, they're... they're wandering
- 59:28around Los Angeles and you may be wondering,
- 59:30what happens if something goes wrong? So one
- 59:35approach is figuring out the standard for liability
- 59:39is their traditional product liability approach
- 59:42under a negligence standard. Under that approach,
- 59:46the plaintiff has to show that there was a design
- 59:48or manufacturing flaw in the self -driving car
- 59:52and this flaw led to the accident that caused
- 59:55the injury. or the property damage. But absence
- 59:58such as showing, the victims would not be compensated.
- 1:00:03There's another standard that I call the reasonable
- 1:00:05human driver standard. And under that standard,
- 1:00:09the car manufacturer would be held liable for
- 1:00:12damages whenever its computer driver fails to
- 1:00:16avoid an accident that a reasonable human driver
- 1:00:20would have avoided. And a reasonable human driver
- 1:00:24is one that's competent, unimpaired, and attentive.
- 1:00:30The victims would have to demonstrate that the
- 1:00:34computer driver's behavior would have been unreasonable
- 1:00:38if engaged in by a human driver, but they would
- 1:00:41be compensated if they could make this showing.
- 1:00:45And this is a liability standard that I think
- 1:00:47judges and juries have some domain expertise
- 1:00:50to administer. Now, in my view, policymakers
- 1:00:54should adopt both approaches. They should maintain
- 1:00:57the traditional product liability standard, but
- 1:01:00supplement it with the liability based on a reasonable
- 1:01:05human driver standard. Now there's one objection
- 1:01:08to that, which is that the reasonable human driver
- 1:01:12standard does not require self -driving cars
- 1:01:15to be better than human drivers. And that's,
- 1:01:19of course, true. Now, of course, the traditional
- 1:01:23product liability approach provides some incentive
- 1:01:26for all of the self -driving car companies to
- 1:01:29move up to the de facto industry standard. And
- 1:01:33also keep in mind that human drivers are pretty
- 1:01:36safe. Human drivers are involved in a fatal accident
- 1:01:40only about once in every 100 million miles driven.
- 1:01:45Now, even though that's pretty good, I think
- 1:01:47policymakers will rightly demand more from self
- 1:01:50-driving cars. And if they want this increased
- 1:01:53safety, I think that might be more effectively
- 1:01:57done through a regulatory requirement. For instance,
- 1:02:01if policymakers expect that computer drivers
- 1:02:04can and should react to a suddenly appearing
- 1:02:07pedestrian faster than a human driver would,
- 1:02:11well, regulators can design a test and make a
- 1:02:13specific faster than human response time a performance.
- 1:02:19requirement for self -driving cars. So in short,
- 1:02:22I think a liability system has to be supplemented
- 1:02:25with a regulatory system to provide full protection
- 1:02:29for people and to drive improvements in the technology.
- 1:02:35I also want to say a couple of words about chatbot
- 1:02:37liability, because I think both of our other
- 1:02:40panelists have touched on that. general framework
- 1:02:45is to address the issue, not assuming that chatbots
- 1:02:49are people because they're not. They don't have
- 1:02:51consciousness and they don't have independent
- 1:02:53agency. So you can't hold them responsible for
- 1:02:57illegal conduct. But the companies that provide
- 1:03:00them should be responsible. I think, for example,
- 1:03:04Metta should be responsible if Metta AI is chatbot.
- 1:03:08provides advice, guidance, or recommendations
- 1:03:11that would create liability if that were provided
- 1:03:14by a human. And I think this intuitive way of
- 1:03:18thinking about chatbot liability would accommodate
- 1:03:22speech issues in roughly the way that Eugene
- 1:03:25was mentioning. If a human speaker would have
- 1:03:28a free speech defense from liability, so would
- 1:03:32a chatbot. But as Eugene pointed out in his presentation,
- 1:03:36people don't generally have a free speech. defense
- 1:03:39against defamation claims. I guess neither would
- 1:03:44AI companies. I'm in large agreement with Eugene
- 1:03:47on that stuff. Let me turn now to some of the
- 1:03:51national regulatory issues. I'm going to do a
- 1:03:54quick summary of the administration's AI action
- 1:03:58plan. One of the first things that the Trump
- 1:04:04administration did, back in January was to repeal
- 1:04:09Biden's executive order on artificial intelligence
- 1:04:12and replace it with in January, in July, with
- 1:04:16an action plan, which has several parts, innovation,
- 1:04:22infrastructure, international diplomacy and security.
- 1:04:25So let me summarize some of those parts. One
- 1:04:30point you would expect from any Republican administration
- 1:04:33They call for agency action to dismantle unnecessary
- 1:04:37regulatory barriers. But it also embraces an
- 1:04:41anti -woke principle. It says, our AI systems
- 1:04:45must be free from ideological bias. And this
- 1:04:49seems fine as a statement of principle. But in
- 1:04:52practice, the AI plan seeks to revise the NIST
- 1:04:56AI risk management framework to eliminate all
- 1:05:00references to information, diversity, equity,
- 1:05:05and inclusion, and climate change. So it embodies
- 1:05:09a political agenda, not a praiseworthy commitment
- 1:05:12to neutrality. Interestingly, it welcomes open
- 1:05:17source models. It says that these models allow
- 1:05:20startups to innovate without being dependent
- 1:05:23on the big AI companies. And these open source
- 1:05:26models can become global standards and so have
- 1:05:29a geopolitical value. Gone completely are the
- 1:05:33worries that open models would create national
- 1:05:36security risks. In the area of infrastructure,
- 1:05:41the plan calls for special exemptions from permitting
- 1:05:44requirements for data centers. And it says that
- 1:05:48agencies with significant land portfolios have
- 1:05:51to identify sites suited to co -location of data
- 1:05:56centers and power plants. And DOE has already
- 1:06:00selected four sites. for AI infrastructure development.
- 1:06:05It calls for grid improvements, but this is largely
- 1:06:07a statement of goals rather than concrete actions.
- 1:06:11It calls for aid and grants to the semiconductor
- 1:06:14manufacturers through a revamped chips program
- 1:06:17that's been folded into a new program in the
- 1:06:21Commerce Department. In the international area,
- 1:06:25it calls for global US tech dominance. It says
- 1:06:28that the goal should be to export the full US
- 1:06:32AI technology stack, hardware, models, software,
- 1:06:37applications, and standards, to all countries
- 1:06:40willing to join America's AI alliance. It does
- 1:06:44call for engagement with international bodies,
- 1:06:47but really to make sure that they don't impose
- 1:06:50burdensome codes of conduct on US companies.
- 1:06:54And it calls for more export controls, largely
- 1:06:57to ensure that our allies do not sell chips.
- 1:07:01to China. In the area of security, there's a
- 1:07:04section calling for a rebranded AI safety institute,
- 1:07:09which is now called the Center for AI Standards
- 1:07:12and Innovation, to evaluate national security
- 1:07:16risks in partnership with AI frontier labs. But
- 1:07:21there's no suggestion that AI companies have
- 1:07:24to submit their models to the government for
- 1:07:26review or describe to the government what they
- 1:07:29have done to mitigate national security risks
- 1:07:32as there was the case under President Biden.
- 1:07:37In biosecurity, there's a continuity with Biden's
- 1:07:40approach. He wanted to control access to DNA
- 1:07:44synthesis tools rather than putting the entire
- 1:07:47burden of protecting against biosecurity risks
- 1:07:50on the AI models. And President Trump's AI action
- 1:07:55plan duplicates that. So let me turn to some
- 1:07:59state regulations of AI. The big beautiful bill
- 1:08:04that passed a couple of months ago sought to
- 1:08:08preempt state AI regulation for 10 years, but
- 1:08:10it was soundly rejected. So the state legislatures
- 1:08:14are a vector for AI regulation. And I'm going
- 1:08:17to touch on two major laws. The first one is
- 1:08:20Colorado's AI Consumer Protection for Artificial
- 1:08:24Intelligence Act. It was It was signed by the
- 1:08:29governor in 2024, May 2024. And it applies to
- 1:08:34AI systems that are used in certain key applications
- 1:08:38in education, employment, financial services,
- 1:08:43government services, health care, housing, insurance,
- 1:08:47and legal service. And it puts some duties on
- 1:08:51the deployers of AI systems, developers as well.
- 1:08:56Let's start with the developers. They're the
- 1:08:58people who develop the AI systems. And they have
- 1:09:02to disclose to the businesses that use its systems,
- 1:09:07that is to the deployers, any known or foreseeable
- 1:09:11risks of algorithmic discrimination. And they
- 1:09:14have to describe the measures they've taken to
- 1:09:17mitigate these risks. And these risks also have
- 1:09:20to be disclosed to the attorney general. Now
- 1:09:24the deployer, that is the businesses who use
- 1:09:27these AI systems, they have to have a reasonable
- 1:09:31risk management policy and program in place,
- 1:09:35specifying what they do to identify and mitigate
- 1:09:39risks of algorithmic discrimination. The deployer
- 1:09:43also has to complete an annual impact assessment,
- 1:09:47which includes an analysis and a review of discriminatory
- 1:09:51risks and the steps. that they've taken to mitigate
- 1:09:55these risks. The deployers also have to make
- 1:09:58certain disclosures to consumers and provide
- 1:10:01them with certain rights. They must disclose
- 1:10:04to consumers when they are interacting with an
- 1:10:06AI system, except of course when it's obvious.
- 1:10:10And they must notify consumers that an AI system
- 1:10:13is used to make a decision about them before
- 1:10:16the decision is made. And they have to provide
- 1:10:20a notice of adverse decision. to consumers, which,
- 1:10:23notice, has to include the data that was used
- 1:10:26to make the decision and the source of the data,
- 1:10:29an opportunity to correct the data, and how to
- 1:10:32appeal an adverse decision, even to a human reviewer,
- 1:10:37if that's technically feasible. In addition,
- 1:10:41the employers have to disclose to the attorney
- 1:10:44general that an AI system that it is using has
- 1:10:48caused discrimination. and they have to do this
- 1:10:51within 90 days of discovering this. And they
- 1:10:54also have to share the annual impact assessment
- 1:10:56with the AGE. The AGE enforces it. The effective
- 1:11:00date was supposed to be February 1st, 2026, but
- 1:11:06a new law was passed in August that delays the
- 1:11:09effective date for five months to June 30th,
- 1:11:122026. That's so as to allow more time to consider
- 1:11:16revisions. There was an attempt to revise the
- 1:11:19bill. in August, and the negotiations collapsed.
- 1:11:24And a delay to revise was urged by the governor
- 1:11:27and the AG. It's an enormously complicated and
- 1:11:30controversial bill, and it will almost certainly
- 1:11:33be revised before it's allowed to go into effect.
- 1:11:37The second state bill I want to talk about is
- 1:11:39from Illinois. It's the legislation that regulates
- 1:11:43the use of AI in mental health and therapy services.
- 1:11:47It was passed just this last August, August 1st
- 1:11:51of 2025. And the key provision has to do with
- 1:11:55therapy services. It says that a company may
- 1:11:59not provide therapy services, including through
- 1:12:03these of artificial intelligence, unless the
- 1:12:06therapy services are conducted by an individual
- 1:12:10who is a licensed professional. And even a licensed
- 1:12:14professional may not allow an AI system to independently
- 1:12:19provide therapy or make therapeutic decisions
- 1:12:22or directly interact with clients in a therapeutic
- 1:12:26capacity or generate treatment plans without
- 1:12:30human review or be used to detect the client's
- 1:12:34emotions or mental states. Professionals can
- 1:12:38still use the AI for administrative support and
- 1:12:42supplemental services like notating, note taking.
- 1:12:46But the problem is not just that therapy companies
- 1:12:49using AI to provide services. That really isn't
- 1:12:53the focus of the problem. People, especially
- 1:12:56young people, treat general purpose chatbots
- 1:12:59as their therapist. So does the law apply to
- 1:13:04providers of chatbot services like OpenAI when
- 1:13:08a user uses the chatbot as a therapist? I think
- 1:13:12it depends on whether OpenAI is providing therapy
- 1:13:16services through its chatbot when the user's
- 1:13:20prompted to answer questions about their mental
- 1:13:23health. And that is not yet clear, at least not
- 1:13:27to me. And if it does apply to OpenAI, how could
- 1:13:31the company restrict these of its chatbots to
- 1:13:34comply with the law? Let me note in closing that
- 1:13:39Nevada and Utah have similar laws. And it is
- 1:13:43widely expected that other states will follow
- 1:13:46suit and imitate the law that's been passed in
- 1:13:50Illinois. At the federal level, by the way, the
- 1:13:53FTC is getting into the act. They recently announced
- 1:13:57a 6B study targeting the treatment of children
- 1:14:00by AI chatbots. They're going to be sending some
- 1:14:03questionnaires to the major AI companies asking
- 1:14:06what they're doing in that regard. It doesn't
- 1:14:09mean that they'll take action, but it's an initial
- 1:14:12step that could lead to some action on the airport.
- 1:14:16There won't be any national legislation in this
- 1:14:18area, I don't think, but I do think these kind
- 1:14:22of laws will spread at the state level. Thank
- 1:14:27you, Mark. Mark, can I ask you a little bit more
- 1:14:32about your book, Regulating Digital Industries?
- 1:14:36The underlying premise of the book is that new
- 1:14:42laws are needed to promote competition, protect
- 1:14:47privacy, ensure free expression in the digital
- 1:14:52age. Some of your colleagues at Brookings, most
- 1:14:58notably former FCC Chair, Tom Wheeler has also
- 1:15:04advocated for the creation of a new digital regulator.
- 1:15:08That obviously would require an act of Congress.
- 1:15:11And as you just said, it seems unlikely for them
- 1:15:15to move in that direction in the foreseeable
- 1:15:17future. But what are the basic arguments that
- 1:15:22support your views on this? Well, read the book
- 1:15:28there. Thank you very much for the plug the second
- 1:15:34time. I mean, let's take, for example, competition
- 1:15:39law, given the recent decision in the area of
- 1:15:45Google's search. It's pretty clear that there
- 1:15:50are some fundamental limits on what antitrust
- 1:15:53can do in the fast -moving world of search, of
- 1:15:57internet, and now of artificial intelligence.
- 1:16:00It seems pretty clear to me that what is needed
- 1:16:03is some sort of regulatory oversight over the
- 1:16:06industry itself. And the competition policy,
- 1:16:12as it's currently structured, is really a matter
- 1:16:15of bringing a law enforcement case, getting a
- 1:16:20conviction, and then having a judge set a punishment.
- 1:16:22And that's just not a recipe for long -term oversight
- 1:16:26of the industry. If you want to promote competition
- 1:16:28in that industry, you have to have behavioral
- 1:16:31requirements overseen by a regulator. Even if
- 1:16:35you were able to split the different parts of
- 1:16:38the company into parts, they would have to be
- 1:16:42a regulator to regulate the interaction between
- 1:16:45those separate parts and to make sure that they
- 1:16:47don't reintegrate themselves. So at some point,
- 1:16:50you need to have an ongoing structure to oversee
- 1:16:54the industry. And competition policy is just
- 1:16:57not set up that way right now. So let me treat
- 1:17:01that as the answer to the general question of
- 1:17:03why you would need a digital regulator. Do either
- 1:17:09of the two of you, Eugene or Jess, have a different
- 1:17:12view on a challenge mark on that? No, I don't.
- 1:17:23But I do want to mention something about the
- 1:17:25about the AI of the Waymo accidents. I don't
- 1:17:29know if people actually all of you folks are
- 1:17:32in L .A. or at least maybe there are people from
- 1:17:34all over the country. But there are Bar Association.
- 1:17:37You cannot possibly have missed the ride share
- 1:17:41amigo, right? Like driving down from Beverly
- 1:17:45Hills down La Cienega to the airport you see
- 1:17:49a one stretch of two blocks without literally
- 1:17:51five ride share amigo bulletin boards Bulletin
- 1:17:56boards there. I'm not sure if you've seen though,
- 1:17:59but the ride share your AI car accident amigo
- 1:18:04is also I just had chat GPT generated while I
- 1:18:09was listening to Mark. But this is the future.
- 1:18:13In fact, I will license this hard work for you
- 1:18:18folks if you want to get into a new business.
- 1:18:23That's great. Well, we did have. Only one question
- 1:18:29so far submitted in the in the Q &A, so I just
- 1:18:32want to reiterate to our audience that that's
- 1:18:37available to you. We have a question that I think
- 1:18:42was meant for Jess about publicity rights. It
- 1:18:46says, I thought publicity rights pertained to
- 1:18:49prohibiting advertising in commercial uses only,
- 1:18:53such as posters, figurines, endorsements, et
- 1:18:55cetera, not creative uses such as generated characters
- 1:19:00within a movie or TV show. Is that correct? Doctrinally
- 1:19:05speaking, yes. And I will state just as a FYI,
- 1:19:09some states they'll even go beyond the commercial
- 1:19:12use aspect and they'll say, you know, in other,
- 1:19:15they'll say in non -gala. protect non -commercial
- 1:19:17uses as well. So it just depends on the state
- 1:19:19that you're in. But no, for the most part, doctrinally
- 1:19:22speaking, creative uses. We do have First Amendment
- 1:19:25protections for commentary, satire, parody, and
- 1:19:30some of this creative expression. There's the
- 1:19:31famous Rogers Grimaldi test that we apply as
- 1:19:34well when we're deciding whether we're in the
- 1:19:37commercial and sort of like a commercial use
- 1:19:39versus a protected First Amendment creative speech
- 1:19:42use. But it all has to be litigated at the end
- 1:19:45of the day. of the day. So at least in the entertainment
- 1:19:48context, we got lots of money and lots of really
- 1:19:51great lawyers out here and on this call. But
- 1:19:54for folks who are maybe smaller creators or smaller
- 1:19:58performers and artists, you can really, once
- 1:20:02you start getting close to that line, could mean
- 1:20:05very expensive litigation down the road. Yes,
- 1:20:09Mark. I've got a question for Jess as well. It's
- 1:20:14back on the Take it Down Act, which you did a
- 1:20:17good summary of. And the notice and take down
- 1:20:21process that they set up is very similar to the
- 1:20:24one that's set up for copyright under the Digital
- 1:20:28Millennium Copyright Act. But that act has a
- 1:20:31number of sort of anti -abuse protections in
- 1:20:34it. I mean, there's a counter -notice process
- 1:20:37where you can say, wait a minute, this is not
- 1:20:38infringing. and you can enter a whole process
- 1:20:42that ultimately will bring the whole thing to
- 1:20:43court. There's nothing like that in the Take
- 1:20:46It Down Act. Do you think if there's buyer's
- 1:20:49remorse about this act, which I think we both
- 1:20:51believe there will be, that would be an improvement
- 1:20:54that would make the thing a little bit more workable?
- 1:20:59look I'll be real with you even under the DMCA
- 1:21:02there's that they even have that I think it's
- 1:21:05what 512 f where even for frivolous claims when
- 1:21:08you bring that in that could be you could potentially
- 1:21:12be sanctioned for that and courts have just failed
- 1:21:15time and time and time again refused to apply
- 1:21:18512f even in situations where it is so clearly
- 1:21:21obviously a frivolous takedown. So I would imagine
- 1:21:23that even if you had sort of that protection
- 1:21:26or guardrail in a publicity rights context, the
- 1:21:29take it down context, I'm not actually sure how
- 1:21:31useful it would be. And I'll also point out too,
- 1:21:34it's structured like the DMCA, but I think many
- 1:21:38will agree the DMCA itself has its own problems.
- 1:21:41The take it down act, It doesn't learn any of
- 1:21:45the lessons that we've learned over the decades
- 1:21:47of dealing with DMCA. It doesn't require the
- 1:21:51person who brings a takedown to state that they
- 1:21:54are, they even have a right to bring the takedown.
- 1:21:56Whereas in DMCA context, you have to be an agent
- 1:21:59or you have to be the rights holder. There isn't
- 1:22:01that, as you were saying, there isn't sort of
- 1:22:03a guardrail in place for if it is a lawful takedown
- 1:22:06or not, if we're talking about protected speech
- 1:22:08or not. Whereas in the DMCA context, there's
- 1:22:10a little bit of wiggle room for fair use. There
- 1:22:13there's a essentially what I've been calling
- 1:22:15a stay down and monitoring provision where it
- 1:22:18says that you not only have to remove the one
- 1:22:21the one that the content that's been noticed
- 1:22:25or at issue, but you also have to remove any
- 1:22:27identical copies. And the person who's sending
- 1:22:30the takedown notice doesn't even have to say
- 1:22:32where those identical copies are or which ones
- 1:22:34are at what time. So I would actually say it
- 1:22:37tries to mimic the DMCA, but it doesn't even
- 1:22:39do that well. It doesn't learn any of the lessons
- 1:22:42that we've learned over the past several years
- 1:22:43from the DMCA is abuse. Let me just ask another
- 1:22:52question to the panel. Given the breakneck speed
- 1:22:57at which AI models and applications are being
- 1:23:03developed, it may not be overstating it to say
- 1:23:08that AI is being used in nearly every aspect
- 1:23:13of society. Is there anything that from your
- 1:23:18particular points of view is being overlooked
- 1:23:21about AI law, or alternatively, is there anything
- 1:23:26that keeps you up at night when you think about
- 1:23:29the direction that AI is going? That's for anyone
- 1:23:33and all. I'll throw out there. I mean, the article
- 1:23:42that I'm writing about, so I'm taking the perspective
- 1:23:45that I actually think generative AI, and I'm
- 1:23:48staying in the realm of generative AI. I think
- 1:23:50it's a publishing technology, and I think it's
- 1:23:52very clear. Yes, we have the listeners case,
- 1:23:55and I think Eugene did a fabulous job really
- 1:23:57drawing that out. But I also think what's really
- 1:23:59being overlooked is the fact that the folks who
- 1:24:01are designing, developing, providing these tools,
- 1:24:04they are operating as new modern publishers in
- 1:24:06the same way that we saw with the internet. And
- 1:24:09what's keeping me up at night is as we are getting
- 1:24:12into this sort of machine, human distinction,
- 1:24:15AI is only getting better. And our technologies
- 1:24:18are only improving as ways to be able to communicate
- 1:24:21with each other, to publish information to each
- 1:24:24other. And so this isn't just going to be conversations
- 1:24:27about AI. A lot of the conversations that are
- 1:24:28taking place about generative AI are starting
- 1:24:31to fall back into these discussions that we are
- 1:24:34having about internet, social media algorithms,
- 1:24:37for example. We've seen this whole discussion
- 1:24:40emerge around like treating generative AI like
- 1:24:42products and treating, and just the same way
- 1:24:45we're talking about social media algorithms,
- 1:24:46like we need to apply products liability. And
- 1:24:48again, we're talking about communications tools.
- 1:24:50There's a reason why we don't apply products
- 1:24:53liability theories to newspapers, for example.
- 1:24:55And so I'm sort of worried about us losing the
- 1:24:57plot here as our technology gets better and better
- 1:25:00and as it allows us to be able to communicate
- 1:25:02in more and different ways. And we're not just
- 1:25:04going to screw things up for AI. We're going
- 1:25:05to screw things up generally in the... communication
- 1:25:08space. So if I could just mention a little bit
- 1:25:13about kind of my perspective on liability for
- 1:25:17for AI companies and such. I just and I agree
- 1:25:26that there is should be broad protection for
- 1:25:31for AI companies as speakers, both on their own
- 1:25:34behalf and on behalf of listeners. This having
- 1:25:39been said, there are exceptions to the First
- 1:25:41Amendment. So one example is defamation. Another
- 1:25:44example, we don't know the exact scope of it.
- 1:25:47It is now being litigated before the court in
- 1:25:50a case, I believe, called Charles v. Salazar.
- 1:25:53But there may be some exception for professional
- 1:25:56client speech. That case involves, if I recall
- 1:26:00correctly, state law that restricted conversion
- 1:26:04therapy, an attempt to basically change people's
- 1:26:09sexual orientation when they seek it out. But
- 1:26:12various legislatures have said, well, this is
- 1:26:15a bad idea, it's not going to be effective, so
- 1:26:18therefore we're going to prohibit psychotherapists
- 1:26:21from engaging in it. And now it's before the
- 1:26:24court as to whether psychotherapists have a First
- 1:26:26Amendment right to engage in it. Maybe the court
- 1:26:28will say yes, but it might also say, well, when
- 1:26:30we're talking about professional client speech,
- 1:26:33you know, maybe viewpoint -based restrictions
- 1:26:34like this one are impermissible, but maybe some
- 1:26:37other restrictions, if court might say, for example,
- 1:26:40you know, This restriction is impermissible.
- 1:26:42But of course, if you can show that this psychotherapist
- 1:26:45was actually engaged in malpractice by doing
- 1:26:48something that reasonable psychotherapists would
- 1:26:51conclude is going to do more harm than good,
- 1:26:53well, then, for historical reasons, that's a
- 1:26:55perfectly legitimate kind of claim. So let's
- 1:26:58say the court says, yes, there is essentially
- 1:27:00this exception for negligent professional client
- 1:27:03speech. You can speak negligently to the public
- 1:27:06at large. You can't sue a newspaper for having
- 1:27:08an incompetent legal advice column or psychiatric
- 1:27:12advice column, but you can sue a professional.
- 1:27:15Then one question will be, can you sue an AI
- 1:27:18company on a negligence type theory? And product's
- 1:27:23liability, at least when you get to design defect,
- 1:27:25is essentially a form of negligence theory. It's
- 1:27:29not exactly, but it oversimplifies it in some
- 1:27:32measure. But basically, it's all about the argument
- 1:27:35being that this is an unreasonable design. You
- 1:27:38designed it to be essentially carelessly in a
- 1:27:45way that might breach professional duties or
- 1:27:50might cause physical harm and such. So you can
- 1:27:52imagine a case like Garcia v. Character AI, which
- 1:27:56says, look, this is something that is protected
- 1:28:02by the First Amendment. That is to say, communications
- 1:28:04with people, including ones that discuss suicide.
- 1:28:06But especially when the company can easily tell
- 1:28:11whether it's a minor involved and can easily
- 1:28:14tell whether the speech is potentially Causes
- 1:28:16a danger of suicide that part may be very hard
- 1:28:19by the way I'm skeptical about it, but if that's
- 1:28:23so you could imagine a court saying look There
- 1:28:26would be negligence liability because speech
- 1:28:29to a particular minor where you are aware, you
- 1:28:33as the company are aware that the person is a
- 1:28:35minor that might be harmful to the minor might
- 1:28:37be unprotected. So that I think is going to be
- 1:28:41where there's going to be a lot of debate about
- 1:28:45negligence and design defect type responsibility.
- 1:28:50if you conclude, as Jess and I agree, one should,
- 1:28:54that the companies are presumptively protected
- 1:28:56by the First Amendment. Very quickly, Mark, because
- 1:29:00we have two more questions from audience members.
- 1:29:04A quick comment. I think the public is now thinking
- 1:29:08that the AI companies are very much like the
- 1:29:11social media companies, and they have to be reined
- 1:29:13in in some fashion. But I think the public was
- 1:29:17disappointed when they found out how little social
- 1:29:20media companies were actually liable for under
- 1:29:22the law. It wasn't just Section 230. It's even
- 1:29:25without the Section 230 protections, information,
- 1:29:29racism, all that other stuff that they went out
- 1:29:31of their way to protect against was not illegal
- 1:29:34speech. Same is true for AI companies. The public
- 1:29:38is going to be very surprised when they discover.
- 1:29:40that applying the law to the AI companies doesn't
- 1:29:43get them the results that they think they're
- 1:29:46going to get. Okay, very quickly. We're near
- 1:29:49our time. I assume it's okay with the three of
- 1:29:51you if we go a couple minutes over. We have a
- 1:29:56question about the New York Digital Replica Law.
- 1:29:59So Jess, how is the term in place of work the
- 1:30:04individual would otherwise have performed in
- 1:30:06person? How is that defined in the law, if you
- 1:30:10know? And does its scope raise concerns about
- 1:30:13being overly broad and vague? I'm not. I apologize,
- 1:30:19I'm not specifically sure with regards to New
- 1:30:21York. I will say, I know they're trying to model
- 1:30:25it off of the SAG -AFTRA CBA and the way that
- 1:30:29SAG -AFTRA goes about it. Sorry, let me pull,
- 1:30:33I'm pulling that back up for you. Okay, in term,
- 1:30:35place of work, the individual would otherwise
- 1:30:36have performed first. The way that SAG -AFTRA
- 1:30:39talks about it is, it needs to be very specific
- 1:30:44use cases. So if you've articulated that you
- 1:30:47are using AI to de -age someone, you will have
- 1:30:50to, in order to do that de -aging process, you're
- 1:30:52gonna have to take a scan of their face, essentially.
- 1:30:55What they're trying to prevent against is now
- 1:30:57that you've licensed it for this use case of
- 1:30:59de -aging, you can't then go and use it for another
- 1:31:01separate use case where you don't even now, if
- 1:31:04you wanna say like, look, the, we, We can just
- 1:31:07AI the same performer. We can just use that in
- 1:31:11an advertisement for the trailer of this movie
- 1:31:13now, or if we're going to use it in some other
- 1:31:15context. Each one of those use cases needs to
- 1:31:18be sort of negotiated individually, especially
- 1:31:20if you are now opting to not use an actor, an
- 1:31:24actor as like a live human actor in place of
- 1:31:26where you would use it for AI. So I imagine that's
- 1:31:29probably where this is going to go in New York.
- 1:31:31And are there concerns around whether it's too
- 1:31:34vague or broad? I think so. And I think we're
- 1:31:36already starting to see that play out, at least
- 1:31:38in the collective bargaining agreements taking
- 1:31:40place with SAG -AFTRA and some other labor unions.
- 1:31:43But I think the New York one just recently went
- 1:31:46on to the books. So we're going to have to see
- 1:31:48how that ends up playing out. And I'll admit,
- 1:31:50I don't negotiate these contracts. I'm kind of
- 1:31:52like moonlighting right now in the entertainment
- 1:31:54space a little bit. So I would love to hear,
- 1:31:57honestly, from experts who do these contracts
- 1:31:59and who deal with entertainment folks as well,
- 1:32:02performers, their concerns too. OK, and then
- 1:32:06just one one other question has been submitted
- 1:32:09regarding algorithms and can algorithms be subpoenaed
- 1:32:16in trial? Interesting question. Anyone have have
- 1:32:24a thought on that? Well, an algorithm can't be
- 1:32:27subpoenaed, just like. Well, not to appear, but
- 1:32:30can the algorithm be required to be placed in
- 1:32:34evidence? I would think that that would be very
- 1:32:37common. Now, if the lawsuit is your product is
- 1:32:42defective and it caused harm the following way,
- 1:32:45of course you need to understand the algorithm.
- 1:32:47If the claim is the algorithm violates anti -discrimination
- 1:32:54law in various ways, of course you need to do
- 1:32:56that. Now, how you're actually going to implement
- 1:32:59that is complicated. First of all, of course,
- 1:33:03even if you It may be that it's too voluminous
- 1:33:08to be able to use it in a meaningful way. It
- 1:33:13may be that it is a trade secret, in fact, perhaps
- 1:33:15the trade secret of the company, and there would
- 1:33:18be restrictions on it. Maybe that as a practical
- 1:33:20matter, what you would do is rather than subpoena
- 1:33:22an algorithm, you would maybe appoint some expert,
- 1:33:27maybe a neutral expert who would then who would
- 1:33:30then look at it and answer certain questions
- 1:33:34or what have you. Sometimes you may just subpoena
- 1:33:37the person most knowledgeable about it in the
- 1:33:39organization. But a lot of these lawsuits are
- 1:33:43at their heart going to be all about figuring
- 1:33:45out what does the algorithm actually do? And
- 1:33:48there would have to be some sort of discovery
- 1:33:50of that in one way or another. Yeah, I think
- 1:33:53it's important to recognize the the limits of
- 1:33:57that because trade secrets is a very strong barrier
- 1:34:00to getting at the details of these kind of, there's
- 1:34:03one case in Michigan, the Loomis case, it was
- 1:34:07Wisconsin, the Loomis case where they looked
- 1:34:11at the recidivism score that was provided and
- 1:34:15the plaintiff said, I want to look at it. And
- 1:34:17they said, no, sorry, can't do it. How can I
- 1:34:19tell whether it's accurate then? They said, sorry,
- 1:34:21secret. Yeah, so it's a really tough barrier
- 1:34:24to get over. So I lied and there was one other
- 1:34:27question that I neglected. You hallucinated.
- 1:34:30I didn't notice. I hallucinated. There's a lot
- 1:34:33of that going on these days. Are social media
- 1:34:36influencers considered to be public figures?
- 1:34:41I guess the question is what is the standard
- 1:34:43for what is a public figure? So I don't know
- 1:34:48of cases that specifically focus on that. There
- 1:34:50may very well be some. Historically, this issue
- 1:34:54has arisen with regard to media figures of various
- 1:34:58kinds. And the answer is there are formal rules.
- 1:35:04One way you can be a public figure is if you
- 1:35:06have achieved pervasive fame or notoriety. Of
- 1:35:10course, that's a judgment call and depends how
- 1:35:13pervasive and sometimes the question is, is it
- 1:35:15absolutely pervasive throughout all of society
- 1:35:17or just among the particular community involved
- 1:35:20there? And I think the answer is different state
- 1:35:23courts look at that differently. However, the
- 1:35:28more common way that people become public figures
- 1:35:30is through being limited purpose public figures.
- 1:35:33They voluntarily inject themselves into some
- 1:35:35debate and they become public figures for purposes
- 1:35:38of commentary germane to that debate. So if,
- 1:35:42for example, somebody gets involved in a political
- 1:35:46campaign, let's say some ballot measure, ballot
- 1:35:48measure having to do with, I don't know, gender
- 1:35:50identity, then if somebody says, oh, this person
- 1:35:54has said these offensive things or this person
- 1:35:57has made these false statements in the past that
- 1:36:00show that he's untrustworthy as to this, then
- 1:36:03in that case, that would be, he'd be a public
- 1:36:06figure for those purposes. But on the other hand,
- 1:36:08if the allegation is this person was, I don't
- 1:36:11know, was engaged in some, just some garden variety
- 1:36:17extramarital affair that might not be relevant
- 1:36:22to that subject. So influencers, Often what they
- 1:36:28do is they get themselves involved in public
- 1:36:30debates. Sometimes, if they're kind of influencers
- 1:36:34having to do with, I don't know, with entertainment,
- 1:36:37they may get themselves involved in not political
- 1:36:39debates, but entertainment -related debates.
- 1:36:41But at that point, they may very well be limited
- 1:36:43purpose public figures. But again, I think this
- 1:36:45is a situation where we just apply existing law.
- 1:36:48Sometimes it's kind of vague, but the law has
- 1:36:50made its peace with that. And of course, we also
- 1:36:53recognize that the rules are different in different
- 1:36:55jurisdictions. All right, well, I think we're
- 1:36:58going to have to leave it there. Thank you all,
- 1:37:01Jess, Mark, Eugene, for taking the time to be
- 1:37:04with us. And thank you to our audience who have
- 1:37:07hung in with us this 97 minutes. So appreciate
- 1:37:12everyone's participation, and have a good rest
- 1:37:15of your day. Thanks very much for having me.
- 1:37:18Thank you.