Tech Brew Ride Home - (BNS) The Legality Of The AI Stuff

Episode Date: April 5, 2025

Is what happened with the Studio Ghibli ChatGPT image generation stuff even legal? I spoke with AI lawyer Rob Rosenberg, former general counsel at Showtime Networks and founder and principal of Telluride Legal Strategies about it, and what he told me was very interesting. Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to another weekend bonus episode of the TechMeme Ride Home podcast. I'm your host, Brian McCullough. As always, we are going to do something today that is probably long overdue.

We've been talking about it so much the legality of AI around copyright and things like that that I thought today we talked to an expert on these sorts of things. We're going to be talking to Rob Rosenberg, who is the principle of tell you ride legal strategies. Rob, thanks for coming on the show. Thanks for having me. Forgive me for being this blunt about it, but give us your bona fides briefly, if you could, so that we understand your background in this area.

Sure. I'm an attorney. I've been practicing 30-plus years. I am formerly the Executive Vice President General Counsel of Showtime networks, and since leaving Showtime about two years ago, I've been running my legal and consulting practice tell you're on legal strategies.

So let's start really high level first, okay? What is the state of copyright law in this country? Like, I mean that in the sense of, is it settled law, like settled going back to English common law? Or is it something that has been in flux in recent years because of technological advances? Yeah, I mean, I think, you know, So choice B is definitely much closer to reality.

We are, you know, we as a nation and with the technology that is changing constantly on us, I think there's a lot unsettled at the moment. There are certain concepts within copyright law with certain uses, certain exclusive rights to copyright owners that are well settled and well established, and it enables people to, to exploit and merchandise their creative works. They register them for copyright, and they know that they can go out and they can license these works for exploitation. Where I think we are encountering a lot of ambiguity these days is where the courts have not provided direction

on how some of these new platforms and technologies should be handled. So there are people that are proceeding blindly, you know, let's move fast and break things, There are people who own copyrights and saying, hey, but what about copyright law? What about these exclusive rights that I have in my content? And I think there is a real budding of the heads of these two concerns. And I think that, you know, until we get some clarity from the courts, there's going to continue to be ambiguity and lawsuits and this unsettled position within the marketplace. So if I'm hearing you right to kind of put it in Silicon Valley terms, the things that are

unsettled are the distribution, but then also the means of production in essence? I mean, I think, you know, so it's not just distribution, because as we know the position of the AI developers, and I think, you know, that's really where the focus of this conversation is going to fall. They've been using copyrighted works to train their large language models. And they're taking the position that the use that they're making is a fair use, which is an exception to copyright infringement. And that they're saying, look, I'm taking these works, but I'm feeding them into the machine. And the machine, it will become part of my algorithm. But it's not like it is harming the market for people who want to go out and rent Top Gun Maverick or want to go out and rent, you know, the last Jurassic Park film.

because even if my machine, if my large language model is learning off of this copyrighted work, it isn't harming that underlying copyright holder. In contrast, the copyright holders are saying, look, but that's not what the law says. As an owner of this copyright, I have the exclusive right to determine who uses it, who exploits it, how they exploited, et cetera, except for well-known, established exceptions like satire and news reporting and things like that.

So, as sort of alluded to, we're here because of especially the studio Ghibli stuff over the last few weeks after the chat GPT image generation thing came out. What were your first thoughts when you saw this explode? It's interesting to me because I think, You yourself have probably seen. There have been applications on the internet for a long time now

that enable you to feed your photo in, a photo of your daughter, and turn her into a Disney princess, or take your wedding photo and create, you know, Pixar-like characters out of, you know, you and your spouse to send to people. So there have been applications that have enabled this, I think, you know, this is certainly exploded because

it is done so quickly, so seamlessly, I think that this style, the Ghibli style, is basically synonymous with anime, and they really have to find that, and I think that

that's where, you know, certainly they have a humongous fan base worldwide, and I think, you know, fans are seeing this as an opportunity for something novel, something fun. But the question is, you know, who has the rights to do this? And is what OpenAI has done here legal?

Well, is the fundamental question here, from a legal perspective, whether or not they trained their model on actual Miyazaki cells? Is that the bottom line? I think there's two questions here, really. So first is the copyright angle, and the copyright angle is relevant to the training.

Right? So, you know, my guess is because they've scraped the internet, you know, if there are, you know, Miyazaki content, I mean, whole entire films that are out there, then they've scraped that and they've used that to train their models. The question is just, is that a fair use under the law, is it not a fair use under the law? And, you know, the fact that it is taking courts a long time to decide on this, I think, is the part that is really strong. stifled certain advances in development, you know, certainly if the courts can resolve the issue and there's certainty going forward, either for the AI companies where, yes, it's a fair use, go ahead and scrape away and use whatever you want, as long as you're not harming the market for those works, you know, from the original copyright owners. But I think you, you know, the, there's, there's a real challenge of, um, whether, you know, on the copyright holder side, I was going to say is that, um, you know, from, from their perspective, they believe that,

A, they should be credited and B, they should be compensated, you know, and, and I think that's a real, um, issue for them. That if, if, um, it's not like, you know, maybe a couple of years ago, you would say open AI, open AI was a startup company, but they just had a, uh, uh, fundraising round last week where they were valued at, I think, $330 billion, you know, that this is not a company that's counting, you know, every penny that it's got and just trying to keep the lights on. They certainly have money. And I think, you know, copyright owners argue they built that business on the backs of other people's labor and other people's creativity. And they should be sharing that wealth with the copyright holders.

I want to come back to that idea of compensation. in a second. But one of the things that I've seen you say is, like, you know, the question of harm here and why does it matter to a Studio Ghibli? You pointed out that if Studio Ghibli ever wanted, like, down the road to launch its own, like, tool, like, people are using it to make, take pictures of their family and make it Studio Ghibli style. But what if the studio, but what if they wanted to do that down the road, create their own app to do this, this OpenAid Update has essentially taken that business. away from them. Yeah, yeah. I mean, I think it's really, it's an unfortunate set of facts, particularly for studio-like studio Ghibli where it is, you know, it is the definition of an art form. You know, it really created the anime art form. I mean, I think purists will argue it existed before them and it exists, you know, after them, it exists beside them, but I do think that it's It's still a recognizable style within. Correct.

Correct. I think, and that was actually like the, so the other side of the coin, like I talked about copyright issues and where I said that potentially there are claims to be made under the Lanham Act, which is a law in the United States that protects trademark rights. And I think that when people are likely to be confused, that there is. some endorsement or relationship with Studio Ghibli in creating these images through OpenAI. And the fact, you know, like, it's not like people are going in and they're typing, turn this photograph into anime style.

If they had gone in and said that and that's what was going on, but it's the fact that, you know, Open AI has been open and notorious about the fact that they're inviting people to request Ghibli's style translations of their photographs. And the fact that they're using the name of Ghibli in these communications with their users, with the press, with everyone else, that I think there's an argument that they're trading off of the goodwill established by Studio Ghibli.

That's very interesting. I'm going to come back to Napster also later on, but like the equivalent would be, instead of saying play me a song or download me a song a rap song if you say download me this specific track by a tribe called Quest

that's a different definition of unfair use or something I guess yeah I mean I think you know that that one's a little bit muddy I mean what I'd say a clearer example is if you just had a program that said you know turn my photograph into Disney into animated characters

and it created animated characters in some sort of generic animated art form versus you saying Disneyify my photograph. And it creates these photographs in the style of the Walt Disney Company, you know, and with their famous, you know, iconic form of, you know, cartooning and animation. Well, okay. But that's a definable and trademarkable or copyrightable character. Can you copyright or trademark style? It's actually, it is not, a style is not protectable in its own right.

If there is a particular style, like a trade dress, a trade color, that becomes so known and so associated in consumers' minds with a particular brand. with a particular owner, then in those instances, protection is allowed. Again, I've never worked with Studio Ghibli. I don't know whether they've gone out and sought protection for their anime style. But I think the fact that the people are, you know, they're not talking about anime. They're talking about Studio Ghibli style. And, you know, in between, you know, you, me and, you know, the other guy, the only person who can turn your stuff into Studio Ghibli style is Studio Ghibli.

You know, they own this brand. They own this style. What I think, you know, the smarter way Open AI should have handled this is really use the generic, calling it anime and saying you can now create images in anime style. So one of the arguments, and we'll get into real quick here soon, the, the idea of where the law is on this. What about the argument that I've heard made by the AI companies in some of the lawsuits that have already gone to court, although not resolved yet, that what they're doing

is transformative. My understanding is that part of copyright law is if you're transforming the original work, then that's protected. What would you say to that argument that what the LLMs are doing is transformative? Again, I think there is, it's certainly an argument. I don't know whether courts will buy it. I don't know whether it will hold weight. I think, you know, the question is just, you know,

you are creating something that, number one, you wouldn't have been able to create if you didn't have access to and basically scrape the works of copyrighted owners to begin with. And now, you know, again, And Open AI originally was a nonprofit entity. You know, now they're seeking to be a for-profit company.

They're selling monthly subscriptions, you know, starting at, what is it, it's $40 or maybe it's $20 a month or something like that, 20 a month and, you know, end up, you know, like you can, you can subscribe to higher levels of different subscriptions from them. So, you know, they're not doing this out of the kindness of the heart. We want a better mankind. you know, that was the stated goal when they first started, but, you know, they have since changed their goals and their ideals. And I think, you know, they have been clear that I guess, you know, they need to raise more money in order to fund the kind of storage and, you know,

back end that they need in order to keep processing AI requests. And all that's fair and good. But if you're now making money off of this, if you have built, that off of other people's technology. It's like, you know, I can't go out and, you know, create a car and, you know, basically go and, you know, design something that looks exactly like, you know, whatever kind of car, apparently. The cyber truck. The cyber truck. The cyber truck and start selling it, you know, like this is a Rosenberg mobile, you know, like this is, this is mine. And I'm selling it, you would say, no, of course you can't, because you are taking somebody else's ideas, somebody else's inspiration, somebody else's output, and you're creating a business,

using somebody else's creativity, somebody else's property. All right, you're mentioning things like compensation for, if, as a remedy for, like, So they would have to pay royalties essentially if the law bounces against the AI companies. Now, how would you, let me say I'm playing devil's advocate. I'm not arguing on their side. But one of the things that it would occur to me is, is the whole model a unitary thing? Is the algorithm a unitary thing? Like, let's imagine that, okay, it bounces against them and you have to pay royalties per what?

question asked, or how would you break it out if you did into constituent parts? Like, okay, so, you know, I wrote a book, and I'm assuming my book got scraped, and if someone asks a question about the history of the internet, maybe they used my book. But, like, how would you break it into constituent parts? Like, like, again, to use the music analogy, you pay by play or by each sample or whatever. But how would that even be possible at scale if it functions like a search engine where the underlying model and the algorithm is maybe a unitary thing? Do you see what I'm saying? Like, can you, can you, yeah, go ahead.

Yeah, I mean, so, you know, I think that, you know, that is part of the problem is that, you know, even the people who have built these models find them to be opaque to a certain extent. They don't fully understand how they operate. You know, like one of the questions I've been asked is, okay, let's say the court's rule in favor of the copyright owners and certain copyright owners. and certain copyright owners turn around and they say, open AI, I didn't license you to use my work. I want you to take my work out of the algorithm.

You know, could they even do it? You know, I'm not sure the answer is yes, because has it become so much a part of that animal, that large language model? You know, number one, even if, you know, let's say they say that they didn't. Yes, I've removed it. How would we know? You know, you can't go look at a server and say, you know, oh, it's not there anymore. It's not on the server because they've got billions of servers.

You know, there are so many files that are saved. You know, I'm not sure we would know. I'm not sure they could do it. And, you know, I think that that's a real issue. So then you say, okay, if you can't, even if you win these lawsuits, copyright owners, you can't have your works taken out of out of the large language model. what's your recourse? What's your remedy? And I think the remedy for them is compensation. And I think the remedy is recognition, you know, that people are recognizing that you've used these works from

these sources in order to create the system. So, I mean, really, you know, one of the things that I've been talking about a lot is, are you familiar with the idea of statutory licenses in the music business. So in the music business, there are certain uses of song recordings where you don't have to go to the company that owns the copyright and license that work if you wanted to make use of it. What you could do is you can say under the Copyright Act, there is this notion of a statutory license where if I want to use some of these works, you know, to use them in public spaces for in certain broadcasts and telecasts, I can go ahead and do it

as long as I process payment and the formal paperwork under this statutory license regime. And this way, we're making sure that the people who are creating music get compensated for these uses, but at the same time, it provides, you know, there's so many musical works,

it would be nearly impossible for somebody, who is, you know, it owns a venue where they're playing music in public spaces, etc., for them to go out every single time there was a new song that they wanted to play and make available. Instead, they can rely on their statutory license. To me, I think that is the smartest, most logical way to proceed. And I think that there's something in it for both sides. I think that the copyright owners get the recognition and compensation that they deserve.

and the AI developers get peaceful resolution of all of these questions that are out there. Like, you know, there are lawsuits, you know, all over the place dealing with these issues. Imagine if there was certainty for them knowing, okay, I can use these works in training my large language models as long as I pay a statutory license. Is there liability at all for the end user? Like if I made a Miyazaki picture, am I ever at risk for being sued someday down the road or having to pay money or anything like that? I mean, I think it would be very difficult

for them to go after individual users. They did it in the extra case. Yeah, I know. I agree. But I also think, you know, If you went into OpenAI and you said, okay, Open AI, I demand that you produce the names of everybody who created a Ghibli style photograph using your platform. You know, I'm not sure, you know, whether they could demand that of Open AI. I'm not sure, again, you know, whether Open AI could produce that kind of a response, that kind of discovery.

I think, but plus also I think that that's the kind of thing where I think. think it would be a huge lift, it would be a huge expense, where each individual violation would be for very small reward. So that's why I think that, you know, we're probably more likely headed to some sort of a blanket licensing scheme. The time, we're talking about how long this is taking. Does something like this have to go all the way to the Supreme Court before there is like settled precedent or like some of the cases that are in the courts right now, if they get resolved in a year, a year or two, will we have visibility or resolution to move forward? Well, I mean, I think certainly you can get to the Supreme Court very

quickly if, you know, there are demands, you know, external demands, internal demands to to try to resolve an issue with some degree of finality. You know, like I agree with you, there have been lawsuits filed in many different jurisdictions. I think we're likely to see some diversity of those decisions that vary by different jurisdictions. And if that's the case, then it would seem that the case would be right to be heard by the Supreme Court. So, you know, does it need to take years?

You know, maybe it'll take a year. but I do think that it's the kind of thing where, you know, particularly where there's emphasis right now in this country to try to keep progressing, keep innovating, stay competitive with other international players in the AI space. I think that there are a lot of interests who would push this forward to try to get some resolution for one side or the other side, but at least for there to be some answers that people could rely on. A slightly separate question or issue, but related. Where are we on the issue of whether or not something generated by AI is copyrightable? I believe that there has not been any change in that where it is created, by AI, then it is not copyrightable because there needs to be a human component to authorship.

You know, what I think we've seen a little bit of is that where the human input is such a high degree that, you know, I think the way I've heard it described is, is the AI the author, or is the AI the pen and the pencil and the paintbrush? The tool, right? Is it the tool? And, you know, certainly, you know, artists, particularly since we're talking about animation, you know, moved from, you know, doing it with pen and ink and moved to computer. And, you know, is the computer the one that is creating these images? No, it's a tool that's being used by these artists to create these works. So, you know, I think that there, you know, there's also been cases where, you know, the

compilation of images. So the images themselves were not protectable, but the way that somebody arranged them, the way that somebody, you know, I believe that there was a comic book or a graphic novel that was created using different images that were curated through AI. But the, you know, the authorship of the dialogue of the character speaking and the arrangement of all of the images was found to be protectable, separate and apart from the images themselves that were created by AI. Second and the last question, but you kind of made me think, again, coming back to that Napster-era argument, and this is devil's advocate again. But is there an argument for saying this is where technology is going,

and so really the problem is, is that copyright law itself needs to catch up? Well, I think that's exactly what we're seeing. And, you know, like the fact that we have been waiting for quite a long time already for the courts to provide some answers in some direction on this, I think that every day that goes by, AI is used more and more broadly. It's being used by individuals, by businesses, by governments. You know, like we are, you know, we are definitely at this stage where there's really going to be, you know, very little chance of rolling back the time and saying, okay, open AI, if everything that you trained your model on was copyrighted works, you have to throw that out and start over again and obtain licenses from all of these people whose works that you use. It's not going to happen. You know, like it's just technology has advanced to the point and AI has advanced to the point that I think, you know, there are certain options that may have been available a year ago that are not going to be available any longer. And I think the courts are, I think that's part of

what the courts are grappling with is that they are trying to come up with decisions, but literally, you know, every time they land somewhere, the plate shift on them again, you know, they are, you know, they're trying to, you know, with revisions to the Copyright Act, it's like building the plane while flying it. You know, they are, but, but that's nothing new. I think that it's kind of the nature of the beast. And I think the pace of change of technology has been tremendous, certainly for the last 20 years. And somehow we continue to evolve the law to address those new changes.

Final thing, is there anything else that the AI industry and the tech industry generally needs, to be thinking about around this idea of copyright and legality that maybe isn't on our radar that you can see coming down the pike would be sort of like the next big issue? I mean, I think, you know, one thing that's been out there and continues to be out there and it's going to need to be addressed is privacy. You know, I think that there are tremendous issues. You know, there are, you know, the world has gotten smaller, you know, like they're cross-border. uses by other countries, other territories. And I think, you know, that the privacy laws have developed

and they've gotten stricter, not looser, in terms of what companies can do with other people's personal information, name, image likeness, all of those sorts of issues. And I think that regardless of the copyright issues, I think the privacy issues are going to be and continue to be front and center. Things like deep face. Yes, that's 100%. I mean, that is the, you know, the prime example. And I think it's not just, you know, a lot of people talk about it in the context of celebrities or well-known personalities, but I think individuals have the same. And that comes back, sorry, one more thing. And that comes back to the end user issue versus the model exists. But if I'm an end user and I upload a photo of somebody,

then I as the user might be liable because I'm the person that did that that prompted it to do that, right? It may be violated somebody's privacy. Yes, exactly. Rob, thank you so much for talking about all this. Is there anything you'd like to tell us about or plug or anything like that before we let you go? No, I just, you know, obviously, you know, the issues are here. They're not going away. AI is here.

it's not going away, you know, every company is dealing with it. They're grappling with, you know, how do I prevent AI from harming my business? How do I move forward and have it help my business? How do I do it in a way that is, you know, less threatening to my employees? How do I, use it to make my current employees work better, work smarter? I think that, you know, there are a lot of issues. You know, there are some companies that I think a year ago had said, oh, I'm just going to sit on the sidelines and wait it out. But I don't think, I don't think we're at a stage where any company could afford to do that at this point. I think, you know, you're, you know, I kind of joke, you know, they're the companies that are embracing it and they're the companies that are going

out of business, you know, that I think, you know, the question is, which one do you want to be? And certainly, you know, there are people who, you know, I think everybody is literally. learning as we're going. Things are changing constantly that, you know, certainly people can reach out. There are people like me who are studying this every day and they can help navigate your way through it. Well, and people like you are Rob Rosenberg from Tell You Ride Legal Strategies. I'll try to remember to put a link to your website in the show notes. Terrific. Rob, thanks so much for coming on the show and discussing.

Thanks so much for having me. I really appreciate it.