Transcript of EP 188 – Robert Tercek on Intellectual Property in the Time of AI

The following is a rough transcript which has not been revised by The Jim Rutt Show or Robert Tercek. Please check with us before using any quotations from this transcript. Thank you.

Jim: Today’s episode, I’m going to remind people about my mobile game Network Wars. Profoundly deep, it only takes 30 seconds to learn. Takes weeks to master, and just 99 cents with no ads, no in-game purchases, and no time limits. Check it out. Network Wars, two words on the Apple App Store or Google Play, or read more about Now onto our show, today’s guest is Rob Tercek. Rob is an award-winning author and one of the leading authorities on dematerialization and the virtual economy. He’s supervised the launch of new digital services that are used by hundreds of millions of people every day, including the first streaming video on mobile phones, the largest live educational program on the web, and some of the earliest games on a variety of platforms, including PCs, the web and mobile. Welcome, Rob.

Rob: Thanks, Jim. It’s great to be back. It’s nice to see you again, my friend.

Jim: Yeah, we always have such good conversations. This is Rob’s fourth appearance on the Jim Rutt Show. Back in episode 142. Rob and I talked about the Metaverse. That was a really good episode. And episode 139, we talked about education today, mostly post-secondary education as I recall. And back in 133, we did a deep dive into work that Rob has been doing for a long time on digital business strategies. All three of those are good episodes. Check them out. There’ll be a link to them on the episode page at

Rob: Thanks, Jim. Boy, I need to retain your services as a PR coordinator. You’re doing a great job. Thank you.

Jim: Yeah, it’s always helpful to put in little references to stuff.

Rob: That’s true.

Jim: Today we’re going to be talking about intellectual property in the time of AI, especially generative AI like ChatGPT, Midjourney, and there’s many more such things coming in different domains. So where do we want to start, Rob?

Rob: Well, I thought I’d share with you some perspectives from Los Angeles, which is where I’m based. As you may or may not be aware, and not everybody in the country’s aware of this, the Writers Guild is on strike. So the Writers Guild are the people that create the stories that the motion picture business runs on, and that’s everything from movies and theaters to TV shows, to streaming video shows. All of that work is written by union members. The Writers Guild is a guild, is a union, and it’s one of the three big creative guilds. The other two are the Directors Guild and the Screen Actors Guild. And these are the three big guilds for independent workers.

Now, there’s also unions. People don’t realize this when they visit LA, they see Universal Studios or they’re on their way to Disneyland or something. They see the palm trees and the glitz and the fake boobs and the perfect teeth and everything else. And they don’t realize that this is a union town. LA is a massive union town, and in some respects it’s also the biggest freelance economy in the world, the freelance creative economy in the world. And the unions are kind of the backstop to ensure that freelance workers don’t get exploited.

So in addition to the three big guilds I mentioned, there are actually several other performing guilds for musicians and voice artists and so forth. And then there’s also the Teamsters and IATSE, which are the theatrical workers. Those are the grips and gaffers, electricians and other people that would set up your crew. So when you see a motion picture, what you’re looking at is a union enterprise. This is all union labor that’s building those shows and films. And the system works. People are able to have a decent income and buy a house and put their kids through school because the union provides a backstop on their wages.

Otherwise, you’d have kind of dog-eat-dog raised to the bottom on pricing for services. So the union provides a floor. Interestingly, it doesn’t provide a ceiling. The agents, the job of the talent agent is to probe and figure out where the ceiling is. So you have a situation where the top tier writers and other creative talent can earn a great deal of money because their agents will get them extraordinary deals. But then you have a large number of people in those guilds who are working writers or working actors, and they make a decent living. So you have room for both. And I guess the point I’m trying to make is that we live at a time where wage stagnation and wage erosion has been occurring a lot of different businesses. And this is one part of the world where unions seem to work to preserve those wages. However, they’re on strike now.

So we have the Writers Guild has gone on strike. They have a long list of things we can chat about, their list of grievances. They’re striking against the Motion Picture Studios, which are represented by a group called AMPTP, and that’s the Theatrical and Television Producers. Which includes things like the classic movie studios like Walt Disney Company and Paramount and Universal and so forth. But it also includes companies like Netflix, Apple, and Amazon, because of course they’re big streaming platforms. And they too have signed up to union deals so that they can get access to that talent.

Now, what makes this story quite interesting is that the writers, their contract was the first to expire. They couldn’t reach an agreement with the producers, so they decided to go on to strike. And shortly after that, the Director Guild, their contract came up for renegotiation and there was some thought that they might go on strike. And now the Screen Actors Guild, their deal is just about to expire, and they’ve already voted to authorize a strike. It looks like the Directors Guild has reached a settlement. What I would say, just projecting out a few weeks, you’re probably going to see the writers and the actors go on strike, which will shut down the entire motion picture business and everything except for news, sports, and reality TV production. And that’s going to hurt the streamers and the TV networks because you’ll start to see a shortage of shows by September.

People aren’t in the habit of turning on a TV and seeing a blank screen. So that’s going to become a problem for them as they burn through their catalog. Anyway, that’s what’s happening right now. The reason I thought that would be an interesting place to start is this is the first organized labor strike against artificial intelligence. One of the biggest issues for all three of the guilds is the use of AI in the creative process. And I know you love tech and I know you’re interested in creative work. So I thought that would be a fun starting point for us.

Jim: That’s very interesting. And also it’s just remarkably timely. And truthfully, I wasn’t even aware of this writer’s strike when I first started working on this thing. It was before the strike actually happened. But as you know, I’m working on an AI empowered movie script writer. People want to check it out, search Jim Rutt script helper, all one word on Medium, and you can find a detailed essay where I lay out what it does and how it works. And actually since I wrote that, I’ve added a bunch of new features. So my head is totally into this though.

Rob: And actually that article is good. So I’m going to give you a plug here because I’ve been talking to a number of folks about the subject matter of AI and the creative arts, and I’ve used that as a reference point because what you do in that long article is you kind of expose where you run into a problem and how you figured out a workaround. It’s a very practical approach to dealing with AI. I say that because we hear a lot of people saying things like, “AI is going to steal all of our jobs.” Or you’ll hear people say, “And artificial intelligence can never write a screenplay.” These are extreme views, and I think they’re both kind of phony. They’re not really very valid. So I like that.

The other thing though that I think is really relevant here, and I want people to understand what you’ve done, you’re not talking about replacing a writer, you’re talking about helping a writer. And that is directly the issue that is being discussed right now, that’s being debated right now in Los Angeles. Are these tools going to replace us or are they going to give us superpowers? And Jim, what you’ve developed is a way to develop superpowers. If you’re a writer, it’ll make you a better writer.

Jim: It’ll certainly make you a faster writer and maybe it’ll make you a better writer. I think when the things further along it’ll make… Well may be able to make you a better writer too, but it’ll certainly make you a faster writer. And this is where it gets interesting. And I can understand from the perspective of people trying to protect a franchise, if suppose it does, like GPT seems to be doing for programmers, giving you about a three x productivity advantage. Now the world probably could use three x as much software if it were less expensive. But is there room for three x as many feature films and streaming shows? Not probably until we can go text to video without any of the actors, and that then passes the actors off.

So it may be that I believe that for quite a while, maybe for decades, humans in the loop is really going to be the way to get the most value out of these AIs. But it may well mean three X productivity, and particularly in an industry that’s very expensive on the production side, there’s probably not headroom for three x as many feature films and streaming shows. So that might mean one third as many writers on the other side.

Rob: Well, that’s a possibility. So let’s break that down because you included quite a few concepts there. First a near-term, midterm and a long-term scenario, and those are all quite different as you point out. So let’s definitely come back to that. But in the short run, one way to look at the advent of AI in the motion picture business and in the creative arts in general is where do you get a gain? Where do you get a benefit? And one of the points you made is that it could reduce the number of humans that you have to hire. That’s one possibility, but it’s not the only one, and let’s remember that. It could also accelerate the time of production. It could collapse the creative cycle. And actually the tool that you’ve developed kind of does that, right? It helps you move a little faster.

And time is very important because effectively you’re on the clock when you’re making a movie. Time is money. So what I’m starting to see is these creative tools, AI, but that’s not all also production. They’re collapsing the cycle of production down from many months, sometimes a year, shrinking it down to weeks and even days. And that’s a big deal because if you’re running a streaming company right now, the budgets for streaming have been cut. That’s true from Netflix across the board to all the other streamer businesses. They were in kind of like an arms race for the last few years. Everybody was pouring cash. And I mean billions of dollars was being invested in producing new series, new scripted series. The number of scripted series has risen tremendously here. It’s risen, there were at some point more than 500 scripted series in production. Those are the most expensive kind of shows to produce because you have writers and actors and so on.

Reality TV is very cheap to produce. But there were more than 500 shows in production, scripted programs in production thanks to streaming. Now that’s been trimmed back some. So that means we have to be more efficient with our budget. So we could produce shows with fewer writers, perhaps that’s one way to do it. That’s one of the issues in contention in strike. What are the actual number of writers that should be assigned to a particular show? The writers have a very strong argument for why they think a show should be staffed a certain way, obviously the producers would like to reduce that number. But you can also think about this in a couple of other ways. And one of the things I like to think about is how can this improve somebody’s work workflow? Either by giving them a better result or a faster result. And I think that’s a constructive way to look at it. But if you like, I can tell you what the issues and contention are because I think they’re really relevant to this system.

Jim: Yeah, let’s do it. This will be fun. And it’s probably, let’s aim for the ones that are likely be impacted by AI because I’m sure like anything else, they want better health plans-

Rob: Yeah, exactly. So you got it- [inaudible 00:11:25]

Jim: But let’s focus on the things that are relevant.

Rob: So for the last 20 some years, the motion picture industry and broadly Hollywood has been on the receiving end of technology, digital technology. And it’s often a disruptive impact. It’s never, it’s something that people welcome here. It’s always viewed as something that’s kind of scary and disruptive. The motion picture industry has been defensive and I think unimaginative in how they approach these technologies. So when it comes time for union negotiations, the movie studios always say the same thing. They’re like, “It’s too soon. This technology is too early. We don’t know how it’s going to play out. Let’s not put all this union stuff in there because that’s going to encumber us and make it hard for us to navigate in the future.” And every time the unions capitulate. And what happens is then that new distribution mechanism, whether it’s home video, DVRs, some other system for streaming on the internet and so forth, typically the unions cut a bad deal.

And what they’re trying to do is maximize current earnings on the well-known business. So they negotiate for better salary, more writers on a show, working longer and so forth, better pay basically, and they’ll trade off the future. Well, that didn’t work so well the last two times. And the last time they struck a deal was in 2007 just before streaming really caught on in a big way. The thing they kicked down the road into the future was streaming deal and the last round of negotiation. And the studio said, “It’s too early. We don’t know. It’s not a very big thing. Let’s not encumber. And the writers acquiesce to that.” Well, now of course, streaming is the biggest part of the TV business and the fastest growing. So they want to fix that.

So they’ve come back to negotiate. And what they want to do is bump up their royalties that they get, the residual payments that they get on streaming in particular. You can imagine a company like Netflix has had 10 years of benefit of paying writers basically less than the movie studios pay. They want to fix that. They want to fix that differential, so it’s the same. Netflix obviously is very strongly opposed to that idea. And then the number of writers is an issue. So with a TV show, there’s a certain number of writers assigned. And in streaming the number is considerably less. And they call those mini rooms. And the writers are, it’s referred to as a writer’s room. That’s where the writers work. And often there’ll be like 10 writers on a show.

That might seem like a lot to you, but you have to remember they’re creating a whole world and a whole series of stories. And it’s a dynamic process. So the scripts are written, but they’re changing constantly as they’re shooting. And that’s because sometimes the story takes on a life of its own or they want to emphasize one character a little bit more than another. Or sometimes someone gets hurt or injured, they don’t show up, and you have to write them out of the story. So what you need to have as a writer on the set, and this is one of the biggest issues right now in this debate, the streaming companies don’t want the writer on the set. Because they’re like, “What are we paying a writer for to sit around? We’re shooting, the script is done.” And the writers say, “Oh, no, but you need us here for continuity. And if the script changes, someone’s got to write dialogue on the fly.” And they have to make sure that fits with the master plan, the longer story arc.

So it’s an interesting debate, but the biggest issue and the issue I think is most interesting for us today is the advent of artificial intelligence and in particular ChatGPT. Now, if you’ve tried using ChatGPT, you probably have noticed you get an okay result. I don’t think you get a great result. I don’t know of anybody who’s like boasting about the quality of the writing they’re getting. So many people who are professional writers say, “Artificial intelligence will never replace a professional writer.” Now, Jim never is a long time, and I think it’s probably a good idea in this case to not say never. Never say never, because it’s improving at a blistering pace. It’s getting so good. So the writers are quite smart and they put forward a very thoughtful proposition and they said, “Look, here’s what we want to do. We do not want you to train AI on any literary material that was created by writers.” That’s request number one. You can’t use our screenplays. Human written screenplays can’t be used as a source for training AI.

The second thing they want to do is they don’t want the producers to dictate how and when they use AI, the writers want to make that decision themselves. It’s important to know they’re not saying they don’t want to use AI. They’re not Luddites here. They’re not saying, “We don’t want this tool.” The writers are just saying, “We want to make the decision when we use AI.” And then the third and most contentious, and I think most interesting for today issue is screenplays that are started by an AI and then they hire a human writer to come in and fix them. So as you know, there is a standard rate that is paid to a professional writer to write a screenplay from scratch. On the other hand, you can imagine a scenario where a miserably producer is driving in from his estate in Malibu and talking to his cell phone and talks to ChatGPT and says, “Okay, give me a screenplay for a buddy picture featuring these two famous artists,” et cetera, et cetera, et cetera.

It’s a buddy cop movie set in Los Angeles in the summer. And by the time he gets to an office, he’ll have a full screenplay ready. And then he’ll call in some hapless writer and say, “Well, we have the script we’re working with. We’d like you to punch it up. So we’d like to hire you to do a rewrite.” That writer, will get paid a quarter of the amount they get paid to write an original script. So the Writers Guild is like, “We want to close off that loophole because otherwise in the future, all screenplays will be written by ChatGPT like tomorrow. It won’t be a good screenplay, it won’t be a quality screenplay. It’ll have terrible dialogue and boring flat characters. That doesn’t matter. They’ll hire the professional writer to come in and punch it up at a quarter of the price.”

So they see this as a way to devalue the human work. And Jim, this is a really interesting topic because it goes much broader than entertainment and much broader than the entertainment and film business and the writers and the screenplays and so forth. It actually affects all of us. The approach the Silicon Valley companies have taken to artificial intelligence is to train these systems, these large language models on data, what they call data. And when they talk about data, they use that term. It sounds like data is just as natural resource that’s lying fallow on the field somewhere, kind of like a vein of copper that an explorer finds in a mountainside. And now we’re going to go exploit that resource. But let’s not forget that that data actually represents the work of human beings. And very often the data we’re being trained on, the LLMs are being trained on. That data is the work of professionals at the peak of their profession.

These are people who built these hone crafts and built careers on this. Their work is being devalued, and with it, the humanity is being devalued. And that’s actually what the main contentious issue at the heart of the strike is. So we have a strike of a creative union of professional writers, and one of the major issues here is whether or not AI can be used to devalue their human work. I find it fascinating. I’ve been watching this debate and I’m super interested. Obviously I’m a little sympathetic to the writers in this case. And their ask is so modest. What they’re asking for is really relatively humble. The studio executives are paid more than what the writers are asking for.

Jim: And the producers, even more than the executives if they’re any good, right?

Rob: That’s exactly right. And movie studio executives, the people that run the motion picture companies on average are paid more than 30 million dollars. Many of them are paid 50 million dollars a year plus, plus, plus. At the very high end, David Zaslav, who runs Warner Brothers Discovery now, he was paid two years ago, he got a 232 million payout. So these folks are paid quite a lot of money. If you take everything that the writers are asking for, let’s say they magically got everything they’re asking for in this negotiation, the annual cost of that is less than 400 million. The executives who run the movie studios collectively, they’re paid more than that each year.

So you have a really clear statement of what the studios value. And they’re saying, “We want to degrade the value of the writers so that we can pay executives more.” And in my view, I welcome this because frankly, I think that the approach the technology industry has had to many fields that get disrupted is very deeply disrespectful or uncourteous with regard to human labor and human output and humanity in general. And that devaluation is really at the heart of this debate. So it’ll be super interesting to see what happens. I can’t handicap it right now. I can’t tell you which side is going to prevail. They might just settle for money as they did in the past. It’s a certainty that AI will be a part of the process, and it’s not limited to screenplays. Artificial intelligence will be used in every aspect of motion picture production very soon.

Jim: Yeah. Let’s set the historical context here. Certainly it’s been true that technology has been changing the nature of work, and increasing productivity, and reducing the number of people in different fields for a long time. We often forget that as late as 1870, 70% of a Americans were working in growing or distributing or processing food. 1870 was just before the automation of agriculture really took off. By 1922, it was down to about 20%. And today it’s 1.5%, depending on who you count, it’s about 0.7 if you just count the actual farmers themselves. And- [inaudible 00:20:38]

Rob: The same thing happened in manufacturing. So all those displaced farm workers ended up in factories in- [inaudible 00:20:43]

Jim: Yeah, I was going to give you the next number, which is- [inaudible 00:20:45]

Rob: Oh, good. Do it. Yeah, I want to hear it. That’s good.

Jim: … Which is in 1965, General Motors had 500,000 employees. General Motors now has 50,000 employees and makes more cars than it did then. Now, it is true, a fair bit of the work’s been outsourced, but automation has been the biggest thing. And in fact, job loss in manufacturing in the United States is 80%, at least automation, 20% offshoring. But that maybe it’s a little bit more than that now, but it’s principally about automation. And as we were talking about in the pregame, I can’t think of the last buggy whip manufacturer that I’ve known or met. Industries come and they go, so this is the nature of late stage game A capitalism that the players are going for productivity, if productivity could be achieved for a reasonable capital cost. And within the current game, it’s really hard to slow that wave down. So I would say unless we do, as many of us are proposing a radical rethink of our socioeconomic operating system, I’m going to bet at least maybe not in this round, in the long haul on the technology, and the humans are going to have to adapt- [inaudible 00:22:03]

Rob: That’s true. Let’s talk long haul, because the long term here, if you want, we can speculate a little bit like, well, what would a motion picture company in the future look like? Super interesting question.

Jim: Yeah. For instance, it’s amazing number of people I’ve talked to since I’ve throwed my little essay out and my software out, and people are coming to me saying, “Hey, we’re only six months away from not too bad text to video.”

Rob: It’s happening every day now. There’s upgrades coming, and they’re amazing- [inaudible 00:22:29]

Jim: Well the ones I’ve looked at aren’t so amazing yet, but they will be soon. I mean, these are solvable problems-

Rob: The fact that it does it at all is mind-blowing.

Jim: Yeah. And we talk about speed, we talk about dog years being seven years to one. Back in March, after I first got my hands on GPT-4, I said, “I’m going to redefine LLM years.” And that’s a hundred to one. Things are moving so ridiculously fast that if you see as you say something that it works a bit, it’s a miracle that it works a bit. In three years, it’ll be working great.

Rob: I think that’s right. That’s exactly. And that’s the right timeframe. That’s a really reasonable timeframe. We see improvement month by month, but it takes a year before you have the whole system figure out. Because right now you can generate a scene, but you can’t generate a film with text animation.

Jim: And obviously you’re not going to produce Mission Impossible that way even in three years. But you can certainly produce kids shows for instance. I’ve got just about to turn three granddaughter, and I got to tell you, the discernment of three year olds ain’t that great. As long as it’s, the music is snappy, the colors are bright, the people are doing a lot of singing and dancing the three year olds will bite. I would predict that text to video, the first market will be very young kids. Then they’ll move up, and as the quality gets better, it’ll move up to higher and higher age groups that are more discerning and more critical. Now, the next point you mentioned is very interesting, and this opens up a gigantic area. Maybe we can talk about this for a bit. And that is their request that scripts not be used for training. This is of course, one of the really big hot issues. A number of artists that are suing Stable Diffusion. Of course they pick on the weak guy, which is kind of interesting. Rather than suing open AI.

Rob: Well, the sloppy guy in the case of Stable Diffusion, they asked for it. They made some blunders, which they’re trying to backpedal from. But we’ll see. We can talk about those cases because they’re really interesting- [inaudible 00:24:34]

Jim: And this, I find to be one of the most intellectually interesting cases. I’ve been involved in businesses, which copyright was an issue and was also involved when I worked for Thompson in providing some of the top databases to the players in the intellectual property industry. The big databases of trademarks and copyright, et cetera. So I learned a little bit about it, and I’ve been talking to people since. And it strikes me that it’s unlikely that copyright, as we currently know it will apply to training data. Because if you think about it, American copyright specifically is very much about the artifact and not about the idea. Quite specifically excludes the idea from protection. And in fact, even the title is not protected. Most people don’t know that you cannot copyright a title.

The idea that’s a mathematical relationships between the words in a work are co-mingled with billions of other works to produce some other numbers, is so far away from a material instantiation of an artifact. I would find it to be shocking if the case is actually decided on its merits as opposed to backdoor deals, that copyright will provide any recourse at all against model building. Quite interestingly, Japan has already come out really hard on this. They have issued a statement that in Japan model building, and this is amazing, the scope, they said from legal or illegal sources. Illegally obtained stuff, you can still make bottles out of it, and it doesn’t violate the copyright because the nature of these statistical artifacts are not what copyright was intended to produce or to protect.

Rob: So you’re saying that in Japan, they already have a ruling from the copyright office that make fair training possible?

Jim: Yes. Fair training in the most extreme possible fashion. And truthfully, if you read the copyright laws, that is actually probably the right outcome because the copyright laws… I mean, John Perry Barlow, who’s a kind of interesting character, he’s the one that wrote the Declaration of Independence of Cyberspace. He was also a lyricist for the Grateful Dead. He was really quite an interesting fella. He wrote a cool little essay about intellectual property and the digital world. And he said, because copyright was designed back in the days, early days in the days of printing, they ended up protecting the bottle, not the wine.

Rob: Yeah, he’s right. That’s a really good point. It’s the container, not the content inside of it that’s protected.

Jim: Again, unless the fix fix is in, which is not to say that it won’t be probably copyright as we know it won’t apply, at least at this time-

Rob: Well, we’re going to see, so there’s four cases coming, and there’ll be a million more. This is going to get litigated for the next 10 years. So the dust hasn’t settled yet. And let’s point out to everybody, we’re not attorneys. So anybody who’s listening to this, we are just two guys that are interested that are talking about this. Don’t take our advice as legal advice. And by the way, you shouldn’t take advice from anybody on the internet anyway. But let’s talk about- [inaudible 00:27:46]

Jim: Well, I talked to Abe Lincoln the other day, and he told me…

Rob: Yeah, exactly right. So the concept of fair use is at the heart of this, right? And-

Jim: Not really, I think there’s a deeper issue. You could kick this thing out of court before you get the fair use, which is that you’re not using an artifact. All you’re doing is-

Rob: No, no, hang on on, if you don’t mind- [inaudible 00:28:09]

Jim: Let’s hear the argument for fair use.

Rob: Yeah. I mean, no I’m not arguing for… Fair use is well established. So the principle of fair use is simply this. If the US government is going to create these mini monopolies in the form of intellectual property, and for the founding fathers of the US monopoly was an anathema. They did not like it. They kind of held their nose and said, “All right, we’ll carve out patents and trademarks and copyrights.” But they didn’t love it. Nobody who created the United States was in favor of creating monopolies. So they put some constraints around it. And one of the principle ideas, principle concepts, is that there has to be this kind of commons, because that’s what enriches the culture. That’s what people grow from. And people do use other works, previous works, writers and others are inspired by those works. They have to have the freedom to do that.

So the principle of fair use was established to make sure that there’s a very vibrant culture and that people who participate in that culture are not prohibited or even inadvertently accused of copyright infringement. What I would say therefore is copyright cases are going to be very narrowly determined on use, and they’re going to look at how is this being used? This goes to your point. So currently there is no company that has a program for licensing their content for the purpose of training and artificial intelligence, because these systems are pretty new.

By the way that’s changing fast. This is why Elon Musk is always out there saying, “I will license you the Twitter content,” because he’s trying to get carve out some sort of protection. So that he can say, “Hey, we do have a business here of licensing for AI training purposes, and if you’ve done it without paying us, without our consent and without compensation, then you’re infringing.” And actually that’s a pretty smart strategy. So you’re starting to see a lot of online communities put a little ring fence around their content saying, “You don’t have permission to train for free. If you want to license it, you can-”

Jim: Now, let’s stop right there. Let me, stop right there. There’s a very important point. Because it turns out I’m actually advising one of the companies that’s trying to build a licensing collective. And one of the things you find, or at least the position that I’m taking, and I think these people have gotten the same advice from their lawyers, is that you can’t rely on copyright at this point. Instead, what you should try to rely on is license. In the famous open source license, MITA and G, the new such and such, those are not copyrights. Those are shrink wrap licenses, in fact. And their legal status itself is somewhat questionable because you don’t actually explicitly agree to be bound by them. But they have these embedded links in them, and the implication is that you’ve agreed to the licensing terms when you use that content.

It turns out that licensing is not as powerful as statutory copyright for one very important reason. Which is if somebody, let’s say, steals a body of license intellectual property and takes the license off of it and gives it to you, you are not in violation because you did not violate the license. Now that person is even bigger trouble with a license holder, but let’s suppose it happens in the Ukraine or something, and people are kicking out versions of these files with these little licensing links chopped off. Copyright does not have that flaw. Copyright violations, a copyright violation, doesn’t matter. That’s the distinction between contract law and statutory protection. So the attempts to do this in contract law are going to be a lot more difficult than doing them with- [inaudible 00:31:38]

Rob: But you have no choice. You have to do it right? Because in, let’s say it’s a copyright case, someone brings a claim and says, “Hang on, I have this artwork and Stable Diffusion was trained on my artwork, and I can prove that they trained it.” By the way, that’s the first thing they have to do is prove that they had access to your artwork, prove that you own it and prove that they trained on it. But even if you can prove it, all that says training was done. Next thing is I have to show damages. Okay, so that harmed me. How did it harm me? And quantify that. I have to actually say, “I have a business of licensing this artwork for the purposes of whatever the business is.”

Now, like I said a minute ago, three years ago, no one was in the business of licensing their artwork for training AI. It just wasn’t a business. Now everybody’s racing to do that so that they can establish damages in a copyright case. Without that, you’re not going to win a case because you can say, “Here’s probative evidence that the training happened.” That’s actually the case right now that Getty Images is bringing. They can show the Stable Diffusion and others in the LLM that they used that was trained at Getty Images because the watermark shows up. So it’s not really a copy, it’s just proof that it was trained. So there’s probative evidence that training happened, but that in itself does not constitute infringement. Now we have to show the second part, which is that somehow Getty was damaged. Well, Getty’s in the business of licensing images. And if you can use- [inaudible 00:32:56]

Jim: Well, that’s still not enough yet, actually. Because you then would have to say that is the language model itself, even if, again, this is a Japanese said, even if you illegally appropriated the content, is the set of statistical relationships in a language model what copyright protects. And it’s hard to see that it is because it’s not a manifestation. The copyright law is very clear. It has to be an artifact, that’s a specific artifact. Now, I think this is also important, and this is where the artists and writers may have more protection than they realize the old copyright laws still apply.

So for instance, if an LLM kicks out a musical lyric, and as we know, the rules around musical lyrics are much tighter than they are about prose for various reasons and for good reason. If a musical lyric is kicked out that has more than 10 words in a row from something you wrote, that is a copyright infringement. The use of large language models that violate copyright will be an issue, but I think they’re going to have a really tough case. At least in the absence of a license on one side or new statutory provisions on the other of claiming that the mere existence of an LLM or even the deployment of an LLM is a actionable copyright violation, even if it costs some money.

Rob: You make a good point because one thing that’s definitely not happening, it’s not a copying machine. You’ll hear people say this, even the Writers Guild, you’ll see signs where they say, “ChatGPT is a plagiarism machine.” Well, no, it isn’t. It isn’t. That’s a mistake. That’s a mistake and understanding of how these technologies work, and you’re making a very good point. So images, I’m going to talk about images more than words, but the images for things like Stable Diffusion and Midjourney and DALI-2. Images are used in training, but that doesn’t mean it’s making a copy. And actually some people say, “Well, it did make a copy for a second or two while it was studying that. While it was breaking it down and doing a kind of waveform analysis or whatever it does to turn that picture into math.” But that’s not a stable copy.

And that’s what the copyright case would hinge upon is it’s storing a stable copy. There’s a case against Cablevision. Cablevision, you may recall back in the nineties, was trying to run, it was back in the early two thousands, was trying to create a streaming DVR in the cloud. They would buffer versions of shows that were playing and then they could play them back for you on demand. The Cartoon Network sued them among others, and the case was decided finally in favor of Cablevision because they said, “Yeah, they did make a copy, but was not a stable copy.” That case, the defendant prevailed. And I think that’s going to become important in this matter because even if they’re making a copy during training, they’re not making a copy when they’re producing new stuff. So in other words, Stable Diffusion isn’t using images, it’s using pure math. It’s like these gigantic database, that’s what the parameters are, the billions of parameters- [inaudible 00:36:02]

Jim: And the other thing that’s very important is, let’s say I’m more familiar with the text databases than I am the picture ones, even though I’ve played with them some. But both of them use the same deep learning technology. They’re a bunch of numbers, and each number is influenced by every single piece of content that’s put into it. So you can’t map one to one this artifact that was plugged in and some impact on this one number, which by the way, this one number is one of a hundred billion numbers in the case of ChatGPT, 180 billion numbers GPT-3.5, probably a trillion for GPT-4. And your little bit of content had a teeny, teeny, teeny, teeny, teeny impact on a whole bunch of numbers. But to what degree? Nobody knows, nobody can know, at least at the current level of technology. Again, it makes this extraordinarily difficult to rely upon copyright- [inaudible 00:36:51]

Rob: It’s going to be really hard. It’s going to be really hard.

Jim: … Again, we’re not lawyers, but I would bet real money that copyright will not win in this case, and that they’ll have to move to licensing, which provides weaker protection. So IE, you could slap on as a term of use that if somebody like Getty or Adobe will have be able to get this better because they could build it into a… They have a wrapper around their applications click wrap agreement. “I agree not to use this for training.” As opposed to something you just scooped off the web that may have a little file link embedded at the bottom of it- [inaudible 00:37:26]

Rob: Well, let’s talk about a related issue, which is a lot of people ask me, they say, “Hey, have you seen these new Wes Anderson knockoffs?” And they’re really cool if you haven’t seen them. For the folks who are listening, if you’re interested in generative video, the ability to use these large language models, not to generate pictures or text, but actually to generate video sequences-

Jim: What’s the good ones? I haven’t looked at that at all yet. What are some of the tops one?

Rob: There’s two that I would recommend the Wes Anderson… So they basically have generated movie trailers in the style, if Wes Anderson directed Lord of the Rings, or if Wes Anderson directed Star Wars. If you think about a director like Wes Anderson, he’s got a very stylized look. So he use a lot of pastel colors. He always squares the camera off. So it’s a very symmetrical [inaudible 00:38:15] or arrangement. He uses some actors a lot of the time, like Bill Murray and other famous actors that sort of stable cast that he has. So what they’ve done is they’ve created fanciful movie trailers. They’re clearly parodies. So they’re not making a Star Wars movie here. So parody is one of the fair uses in copyright, so that they’re free and clear there. But the issue some people have raised to me, they said, “Hey, wait a minute. Wes Anderson is a director, he is a writer, he is a guy with a career. They’re ripping him off here.”

I’m like, “In what way are they ripping him off? I think this is homage. This is flattery. They’re honoring him.” They’re saying, “No, no, no, they’re generating new content under his name.” Oh, this is a super interesting question, but here’s the interesting part. You can’t copyright a style. A style is an amorphous thing. So certainly artists have styles. Picasso is a style. Andy Warhol’s a style. But if you think about it in New York right now, if you were to go to those street art where they sell art artwork on the streets of New York by Central Park, you’ll see that there’s dozens of artists who are knocking off Andy Warhol.

Jim: Absolutely.

Rob: Why are they not in jail? Why are they not getting shut down? Because you cannot copyright a style. All you can copyright is a work.

Jim: Exactly, an artifact. The bottle, not the wine.

Rob: And you could say, “Well, it’s a series. They’ve stolen from six different works or 10 different work.” You could do that. You could make that argument and say they’re not just infringing one, but then you’d have the burden if you bring that complaint, it’d have the burden to show substantial elements for each of those paintings. So what I’m saying is you can’t broadly copyright a style. Now this Jim, is super interesting because you have to start to think now that we don’t have an intellectual property format to help people who are creators, like creative people. You’re going to start to see knockoff actors, people who sound alike.

There’s already generative things for certain music artists and some have embraced it. So there was a ripoff of Usher. Some other artists have been basically cloned as AIs for singing. And at least one artist now, Grimes has struck a deal with people. She said, “Okay, if you want to use my voice to generate music, it’s a 50/50 deal. You can license my voice and I’ll authorize the work.” And now tens of thousands of people have taken her up on the deal. It’s really quite fascinating. So you have one artist who’s super scaling herself, and there’s a whole group of volunteer workers out there that are like, “Cool, I’ll go design new tracks,’ and DJs and so on.

Jim: Isn’t she Musk’s girlfriend or something like that?

Rob: She is. Of course.

Jim: I didn’t know, I sort of vaguely, I don’t follow the pop culture gossip much, but I did recognize that one from- [inaudible 00:40:59]

Rob: Exactly, no, she managed to make herself relevant in this particular context, which is such a controversial context.

Jim: She’s the right place at the right time to do that, obviously.

Rob: Yeah. Listen, I think it’s really creative and it’s super fun and interesting. This is actually going to be an issue for the Screen Actors Guild, back to the strikes. So what the screen actors, among the issues they are protesting, they’re concerned about the use of AI to replicate their name and likeness, their visage I should say. Their face, their likeness, their voice tonalities and so forth. They’re worried that they’re going to have to compete against robot versions of themselves or generative versions of themselves. So they’re seeking prohibition on that, that they’re seeking that the producers will undertake not to do that without their permission or consent. But already I’ve heard that there are screen actor contracts that are being submitted to actors that have this language in them. So sneaky, the producers are already trying to get this right wherever they can and obtain it.

You can imagine how scary that would be if you’re an actor. There’s already not enough work for actors to begin with. It’s a very risky profession in terms of stability. And you can imagine how difficult it would be to compete with a cheap version of yourself that’s generative. Who’s going to get whacked by this first is not the Actors Guild, it’s the influencers. All these personalities on TikTok and Instagram, they are either going to replicate themselves and turn themselves into AIs or they’re going to find that they’re competing against an army of robots.

Jim: Ooh, that’s clever. I hadn’t thought of that one-

Rob: For sure going to happen, right?

Jim: … Three-year-old TikTok influencers. Right?

Rob: You got it.

Jim: Because though, of course, that’d be interesting because there’s something about TikTok influencers, why does one work and not another one? It’s a subtle thing- [inaudible 00:42:43]

Rob: Yeah, there’s a lot of chemistry involved.

Jim: … And it’s a true free market. There’s no ability to do anything other than compete in the market-

Rob: Just by being stylish and current and relevant. And those people, they can sometimes have a short career, sometimes they have a long career-

Jim: And charismatic, somehow they have a certain kind of charisma.

Rob: And they connect on a way that television does not connect with people. So we have this extraordinary situation now on YouTube and some other platforms where an individual creator can command an audience in the tens of millions. And bear in mind, a very successful cable network will have a few hundred thousand viewers at any given moment. For an individual to command an audience in the millions is really kind of an extraordinary statement about the democratization of media and access to media. But those are the first people that are going to get digitized in my view, because they have no protection, they have no system, there’s no ecosystem to protect them and so forth. And also think about the brands who sponsor them. What they really don’t want is to find out that the influencer they were backing last week was caught with a bong or something, did some crazy thing.

Jim: Exactly. Or without a bong. If it was, say, a marijuana company. Oh my God, the person confessed that they never smoked reefer in their life. Oh, shit.

Rob: Exactly. So human influencers are hard to manage and they’re willful. And they’re whimsical and they sometimes do crazy things. There’s plenty of stories about Jake Paul and other influencers. So the brands would prefer to deal with a robot. They would prefer to deal with a synthetic personality-

Jim: I love this. This is great. This is brilliant, Rob. So young entrepreneur out there, go automate. Go find yourself a big fat, dumb TikTok influencer and do a 10% better job and knock them out of the box. There’s nothing to stop me.

Rob: There are already dozens of companies that are pursuing what they call digital people or digital humans. There’s a company called Fable in San Francisco and here in Los Angeles, John McInnis at McInnis Lab, and there’s a number of other companies- [inaudible 00:44:44]

Jim: A company called, which is doing dialogues. In fact, I’ve actually had a dialogue with Plato. No. So Aristotle, JRR Tolkien and Tony Soprano. And the one with Tony Soprano was surprisingly good, particularly once Tony discovered my father was from Patterson, New Jersey, and we went down quite the Northern Jersey rabbit hole, but it was surprisingly good, right?

Rob: Yeah, it is good. Well, what’s amazing is that the quality’s good today, or it’s like adequate today. Where just a year ago, I remember really clearly just a year ago, it was messing around with Midjourney and the quality was not good. I mean, it was really unpredictable. It was hard to steer it. I didn’t understand prompts very well. The output, sometimes you’d have an image where the person would have 10 fingers on one hand or something. That was just a year ago, Jim. And now the quality of Midjourney five is so extraordinarily good. You can make a photo realistic image or something in the style of an anime artist or something in the style of a real professional artist, and you can control it much better. You can steer it towards an outcome.

Jim: And 5.1’s even better, right?

Rob: Yeah, that’s exactly it. So now let’s just project forward 18 months. The timeline here is not super long till we get to a- [inaudible 00:46:01]

Jim: Hundred years to one year. This is like LLM years, stuff’s going crazy.

Rob: Yeah, that’s exactly right.

Jim: And as my good friend Peter Wang said that ChatGPT, think of it as the equivalent of the Wright Brothers in 1903. It’s just barely off the ground, just you put together with chewing gum and bailing wires.

Rob: Yeah But then 10 years later, World War 1, airplanes played a role- [inaudible 00:46:25]

Jim: And then multiple cut that by at least a factor of 10, a year from now will be the equivalent of Wright Brothers to the Spad XVI or something.

Rob: When do you think you will see the first fully generated film or TV show?

Jim: As I said, I have not looked into the text to video stuff. I wish… I need to do that, I don’t- [inaudible 00:46:47]

Rob: It’s a hard problem, so it’s going to take longer. Generating an image is obviously easier, but there’s already apps where you can input an image and then you just click on a couple of vertices and it’ll generate the in betweens and it will move. So you can make a dog turn around or you can make someone’s head turn and so on. And this is all in the last three months. So this is coming so quickly.

Jim: I need to do a little dive into this. I don’t know anything about it.

Rob: My thinking is, it’s a matter of months. It’s not years, it’s a matter of months.

Jim: Who are the leading technology providers in this space? Do you know?

Rob: Oh, there’s hundreds-

Jim: But who are the top ones? There’s always a top two or three.

Rob: Oh, off the top of my head, I can’t recall the names of any of them. Sorry I should, it’s my head’s full of other stuff right now.

Jim: That’s okay. Let’s, we’ll just move on here. There’s lots of other things to talk about with- [inaudible 00:47:30]

Rob: It’s a lively space. The problem, actually, let me back up once. So ever since Meta’s LLaMA model slipped into the public domain in March, which is really just a couple of months ago, there has been an explosion of open source LLMs. Now there’s new LLMs being introduced every single day. And some are tuned for games and some are tuned for comedy and some are tuned for something else. What you’re going to start to see… We’ve been hearing about large language models, but I think we’re actually going to start to see the emergence smaller and more targeted language models for precision use for certain cases. And this is a kind of empowering story because then you see that there’s a role for creative human to just select the right LLM to achieve a certain effect. Well, that’s like an artist, except their palette won’t be paint colors, their palette will be different language models.

Jim: And they can train and they can fine tune the models too, right?

Rob: That’s right.

Jim: For a couple hundred bucks, you can fine tune LLaMA to load the kind of content you want kind of at the bottom of the stack essentially.

Rob: That’s exactly. So back to the issue of training data, clearly this is the year. This year you’re going to start to see law firms doing their own language models internally, using all the briefs and all the legal matters that have been written within that law firm. So it’s not going to circulate on the internet, but they’ll access to all the work that’s been written. And why would they do that? Well, there’s certainly going to be some writers at every law firm who are quite good, and there’ll be other writers that are kind of average, and there’ll be other attorneys who just aren’t as good. I think this will level up the writers who aren’t as great. So I think you’re going to see a leveling up. People are always worried this is going to disrupt jobs and take away jobs. I don’t really see it that way, at least in the near term, maybe in the longer term. But in the near term, I see this as empowering for people.

And it’s certainly true for ChatGPT. If you just think about who’s really using ChatGPT, it’s people who aren’t great writers and it elevates them to be okay writers. Which honestly, that’s a win for everybody. I don’t see it disrupting professional writers or really talented writers, but even there are use cases for it. Like ChatGPT, if you’re trying to write an argument, one of the best things to do with ChatGPT is say, “Tell me all the arguments against whatever it is you’re trying to argue for.” And you can rehearse your debate. You can kind of get the antithesis like nailed be from ChatGPt, and it’s quite good for that. You’re not having it write anything for you, it’s basically a sparring partner.

Jim: It’s a great analysis tool. I also, here’s a little trick people want to use GPT particularly for is to say, if you wanted to really understand the intricacies of comparison and contrast, there was a reason our English teachers in high school had us write comparison and contrast essays. We usually did bad ones, but GPT writes good ones. For instance, you know how to compare Marx and Ayn Rand and say, “Compare and contrast Karl Marx and Ayn Rand.” You pick the aspects that give the biggest insight into the differences and put it out as a table. Just give that prompt word for word and GPT will produce a nicely formatted table. It will pick 10 dimensions that maximizes the comparison and contrast between the two, and it will fill in all the cells in the table with reasonably intelligent stuff.

Rob: It’s really instructive and it’s actually a kind of quickie way to understand. Candidly, there’s a lot of stuff in college we didn’t get a chance to read. I couldn’t swallow postmodernism because it’s very dense. It’s hard to read, particularly someone like Lacan, I give up and I don’t think I’m the only person in the world who would say that. I confess it freely because I think most people have that problem. So I was playing with ChatGPT and asking it to give me summaries of the postmodernist. And I recommend this to everybody because one of the most fun things you can possibly do, that stuff can be dense and unreadable. With ChatGPT you can say, explain it the way you would explain it to a five-year-old or a fifth grader, and then it will break it right down into plain English and you can get-

Jim: Of course, keep in mind it will hallucinate to some degree. But here’s something that’s important about hallucination-

Rob: We’re talking about poetry, art and simulation, that’s what it’s all about. We’re living with the simulation man.

Jim: Yeah. Though it is true that the more central something is, the less likely it is to hallucinate. Humorously, I use myself as a test case on these large language models. Because I have just enough internet footprint that all of them sort of know that I existed, sort of vaguely who I am. But most of them, the GPT-4 is now pretty accurate, but the lesser ones are terrible. Like Bard, about 90% of what it says is just total fabrication-

Rob: Why is Bard so bad? What the heck is going on? Google has more experience with this than anybody. Do you have any insight into that? I’m so curious.

Jim: I’ve heard that because they’re giving it away free, they’ve chosen not to actually use the big models that they have, and it’s a small, tight model-

Rob: Because their cost per compute is too high.

Jim: Yeah. But they can afford it. But for whatever reason, they have not put their best model out there. Bard is a small model. It’s very, very fast because size and speed are almost linear related. I can tell you from doing my own benchmarking, the GPT-4 is five to six times slower than GPT-3.5, which leads me to believe that the model itself is five or six times bigger. And that fits in quite closely with the numbers I’ve heard. I’ve heard 1.3 trillion parameters for GPT-4, and we know one about 180 billion parameters for a GT 3.5. So to your earlier point, this is hugely important. There’s going to be a very interesting competition with models and around the 80 to a hundred gigabyte parameter range, because those are fast. Those are small enough that even with today’s hardware, they can operate in near real time.

Rob: You can run them on your laptop, you can run them on your phone. I have one on my iPhone.

Jim: Not 80. Not 80.

Rob: No?

Jim: You can run seven. I’ve got some seven and eight and 10 gigabyte giga parameter models loaded up on my PC and they run. They’re slow, but they suck. They are terrible. Even the ones, the lLLaMA ones that are supposedly fine-tuned and are really good, they’re terrible. They’re way worse than Bard. But again, it’s Wright brothers or it’s now 1905. So we’re just slowly on the way up and they’re learning how to do new training. Essentially there’s at least five dimensions of, when you think about an LLM, what goes into it. The corpus, how big it is. The corpus, what’s in it that makes a huge difference. The actual algorithm that did the training, the number of parameters that you output in the model, the difference between an eight gig, a 13 gig, a 60, a hundred, and then you can just see the difference. There’s all kinds of academic studies on that.

And then the last one is after the fact, fine-tuning. Either straight fine-tuning the way you can do with GPT-3 or Laura, which you can do with LLaMA and some other ones. Or the reinforcement learning human feedback, which was used to build GPT-4. Those five pieces are dimensions in a competitive puzzle. Smart people will be thinking about how to play those knobs to produce the optimal benefit at the lowest cost. It may turn out that to say, teach elementary school content, a 60 gig model trained on such and such and this and that, and tune this way and heavily reinforcement learning to keep bad stuff out. Maybe optimal for say K to nine education. Then college if you want education, there may be a very different, bigger data set, a bigger model so it has more nuance in how it thinks about things. Maybe less nanny rails to keep it from saying fuck or something. Turning these knobs is going to become a product management art form. And I think it’s going to be really, really interesting.

Rob: Me too. No, I think we’re already starting to see that happen and it’s an exciting time. So one way to think about that is a proliferation of tools. And for some that might feel like a proliferation of handguns. That the environment is dangerous already, it’s going to get a lot more dangerous because there’ll be a lot more LLMs out there. So who gets kneecapped by artificial intelligence? And I don’t think it’s writers and I don’t actually think it’s creative people. I think creative people will use these as tools. I think it’s big entertainment companies, and I think big entertainment companies have a lot to be fearful of here. And they’re not focusing on at all in the right way. In other words, like the executives who are negotiating against the writers, they’re the ones who are more vulnerable, I think to displacement from AI in the very near term.

Remember when we talk about a copyright case, who brings a copyright case? Who is the entity that makes that complaint? Is it an author? Is it an artist? Very rarely is it going to be an individual artist or author. Happens sometimes, but it’s a very expensive thing to bring a lawsuit for copyright infringement. Typically, who’s bringing that lawsuit is going to be a big media company, either a publishing house or motion picture company. So when we talk about copyright versus AI, it often sounds like, “Oh, these evil technologists, these evil tech corporations, and they’re preying on artists.” That’s how it’s portrayed often.

I think that’s a misunderstanding because actually what it is a battle between two corporate titans. The technology companies that want to focus on automation and thereby productivity growth, making stuff faster, cheaper, better. That’s what they’re always focused on. And over on the other side, you’ve got entertainment and media conglomerates that are globe spanning media empires. And they have a lot of copyrights, a lot of content to protect. And what they want to do is throw the future under the bus in order to preserve their monopolies of the past.

So it’s a battle between past and future. And this is a super interesting dynamic to look at because I think the media companies are going to lose. I think for the reason you just cited, there is just this tidal wave of innovation coming, and it’s here already, but it’s going to continue. It’s just a torrent of innovation in the open source world. It’s not owned by anybody. It’s accessible to anybody. And as a result, I don’t see how the copyright industries are going to be able to maintain these strongholds that they’ve had. Companies like Disney are the ones to very much pay attention to here. And I might point out, if you don’t mind, let me give a little copyright lore.

Jim: Let’s do it.

Rob: The Copyright Act from like 1912 limited the duration of the copyright. Because remember back to our founding- [inaudible 00:58:13]

Jim: 44 years. Right?

Rob: Yeah. They didn’t want to have permanent monopolies, they were trying to avoid that-

Jim: And that the Constitution specifically rules that out. I actually have the language from the US Constitution-

Rob: Permanent-

Jim: … In my notes. But anyway, move on.

Rob: But anyway, so copyright over the years has gradually extended the term of copyright, whereas say for patents. Which is another kind of government granted monopoly over intellectual property patents only last 20 years. That’s it. And then they’re-

Jim: They used to be 17, now they’re twenties, but-

Rob: So they’ve extended a little.

Jim: Yeah.

Rob: But companies like Walt Disney are famous for extending the copyright. And actually it’s always the authors lifetime plus a few years-

Jim: 75 years now.

Rob: … As we approached the time when Mickey Mouse was going to fall into public domain and Donald Duck, which was the lifetime of Walt Disney, plus whatever number of years. Disney has successfully lobbied Congress again and again and again to extend it. Just a few more years, passed the death of Walt Disney. And they managed to do it in 1976. They got life plus 50 years. And then in 1998, they got life plus 70 years. Well, now finally, we’re starting to see that process has come to an end. And for everybody who’s not a big media company, this is actually good news because what we’re really saying is that the copyright monopoly or the copyright mafia, we’re trying to extend protected works. Extend their monopoly into the public domain and take something back from the public.

Now this trend towards extending copyright into almost infinity, that process has come to an end, which means a lot of creative stuff in the 1920s is going to start to fall into the public domain. And that’s a good thing for everybody because that enriches the culture. It gives people stuff to remix and play with and new ideas and so forth-

Jim: Let me put a Ruttism out here at this point. So something I championed for 10 years at least, which is I’d like to see copyright reduced to 10 years, period. And here’s the logic. I know I’ve been involved in publishing businesses. I’ve been a writer, I’ve known lots of authors, almost no actual creator of intellectual property when they’re creating assumes any significant economic tale more than 10 years out. Every once in a while, somebody gets lucky. Somebody writes a perpetual bestseller, 100,000. [inaudible 01:00:30] The same is true of movie business. Most of the revenue comes in the first two or three years. Every once in a while one becomes a classic, it has a nice annuity tail. But nobody creates because of that annuity tale, because it’s so unpredictable.

So therefore, the constitutional provision that the purpose of copyright is to encourage creation does not actually apply for more than about 10 years by logic. Therefore, if we only had 10 years think how cool that would be. When are we most high on music? When we’re 15? So by the time you’re 25, you can use the content from your 15 year old years to remix. You want to do your own video, you want to use Led Zeppelin, the Beatles, go for it. It would be a wonderful gain for humanity to say, “Let us have a commons. Let people make a reasonable return over 10 years, and after that it goes to the public domain.”

Rob: Jim, you’re preaching my language. I love it so much because this is about enriching the commons. That’s exactly it. And until recently, the commons was being preyed upon. It was like the enclosures in England-

Jim: I was truthfully quite shocked that the Supreme Court, that it did go to the Supreme Court, approve the after the fact extension of copyright. That it’s so clearly not the intent of the Constitution because it can’t possibly encourage the creation of something that’s already been created. That’s like- [inaudible 01:01:49]

Rob: And somebody who’s been dead for 60 years.

Jim: Somebody got paid off. I mean, I hate to question our US Supreme Court, but that was so contrary to the clear intent in the Constitution, in black and white that I figure some dirt had to been done on that deal.

Rob: All right, let me shift it back to this discussion of the media companies, because there’s one more thing here that I think you’re going to dig and respond to. And this is another reason why big entertainment and media companies need to be concerned. So we’ve talked about the dynamic where copyright sets up kind of a ring fence, and then effectively it allows those companies to decide who gets to manipulate this material in the future. Who gets to write a new version of Star Wars, like Disney gets to make that decision. Now, it’s not George Lucas anymore, because he sold the intellectual property. It’s now the copyright of Disney. So it’s a way for someone who owns the past to control the future, that’s what the movie studios want. That’s what media businesses like, that’s what they depend upon. But all that involves a human creator. And copyright is only available to human creators, human authors.

People have tried to get works that were created by monkeys copyrighted. And the copyright office ruled, I think wisely, that no copyright is available only to humans, not to monkeys. You can’t do it with an animal. You can’t do it with a machine, and you can’t do it with generative AI. They’ve now given a guidance. In March, the copyright office said any generative works where you type in a text command and then the machine generates the actual work or the output, that cannot be copyrighted because it wasn’t created by a human, it was created by a machine. And many people objected, and this will be tested. We’re going to see some people try to test this. But actually I think it’s a very prudent decision by the copyright office because think about the alternative scenario.

Let’s say that generative work could be copyrighted. Well then you know some kid in a basement in New Jersey would crank out every possible combination of blues riffs because there’s only 12 notes. And then everything in the music industry from that point on would be subject to copyright infringement because this kid would have copyrighted the whole waterfront for every possible combination. I know that sounds a little silly, but it’s-

Jim: Could be done.

Rob: … How things would unfold. It’s the copyright office. They already get inundated with 5 million applications a year for copyright. That’s a lot of copyrighted stuff. So you could imagine that would’ve gone up by a factor of a hundred, that would blow up the copyright system. So they made a good decision. The second reason though, and I think this is a really important, this is why the movie studios and TV companies and media companies need to be concerned.

That means that there’s going to be an ever larger and growing exponentially fast growing body of works that are not copyrightable. So back to your idea of the commons, this generative stuff can’t be copyrighted. So it’s just going to be out there on the internet and there’s going to be an enormous flood of it. I’d argue it’s already happening because the people I know who are seriously into prompt craft, they’re generating something in the order of 20,000 images a month. They’re generating hundreds of images each session. And if you’ve ever spent time trying to learn Midjourney, it’s very easy to generate hundreds and hundreds of images because you kind of go down the rabbit hole till you get the look that you like. So you’re cranking out a lot. Okay, so imagine then this giant pool of un-copyrightable content.

Meanwhile, you’ve got a static pool of copyrighted content that is owned and controlled by media companies. That doesn’t grow as fast. The fastest you can grow that is by hiring more writers. It’ll grow linearly, but it will not exhibit exponential growth the way generative work will. At some point, the free copyrightable content generated by AI is going to exceed the pool of copyrighted content. And new LLMs will be trained on this copyrightable stuff. So we are going to get into remix hell. And there’s some examples of a kind of dementia that occurs when you train it, when you kind of do this circular process of training LLMs on generative content. This is going to be so interesting because-

Jim: Yeah, we don’t know what happens. Because interestingly, some of this many, it’s a little dirty secret. Many of the smaller models are using GPT for to do their reinforcement learning human feedback, and they use GPT as the human. And it doesn’t work all that well as it turns out, but it’s better than none at all. But it’s a shitload cheaper.

Rob: If I were running a media company right now, I would be trying to develop a strategy for that. It’s like, how do we stay relevant when there is an ever expanding pool of free un-copyrightable work out there? There’s an argument to be made to say, you know what? You should relinquish into public domain, some of your works so that you can start to generate fan fiction that keeps your works relevant and keeps them vital and active. We’ll see what happens. I don’t think they’re going to take that advice from me right now, but-

Jim: I guarantee you they won’t. And here’s why, which is, I’ve been in the business of building walled gardens and extracting monopoly rents, and it’s a great fucking business if you can get it. And competing fair and square ain’t nothing like those kind of profit margins. And it’ll shrink both the top line, but it will massively shrink the bottom line. So my prediction is, we used to actually call it this, so you’ll love this, the Errol Flynn defense, which is, you’ll know you’ll lose, remember when Errol Flynn would fight six bad guys coming down a flight of stairs or go, usually going up a flight of stairs, and it’d be six of them. So he’d delay them for 10 minutes. So the heroin could escape by climbing down the trellis or something. But in the end, they would overpower him and he’d have to run off and jump out the window.

The Errol Flynn defense is what these guys are going to have to do. They’re going to have to make it defend these monopoly rent extraction machines gradually running downhill over time, until they eventually clap out. And in reality, the area under the curve will be a lot bigger for them. The bag of profit they will make will be a lot bigger than if they tried to transition to a competitive marketplace. That’s my prediction.

Rob: I think you’re probably right about that. But let’s bear in mind that every time there has been a new technology introduced into the media business, it’s been resisted. The media companies tried to stop the VCR, they tried to stop home video, they tried to stop DVDs, they tried to stop the DVR, they tried to stop streaming. They fought every single one of these innovations arguing that it would destroy their existing businesses, which were all based on scarcity. It’s a Motion- [inaudible 01:08:06]

Jim: Monopoly, rents. Rent extraction. That’s what they’re all about- [inaudible 01:08:09]

Rob: So the movie business works on artificial scarcity. TV business works on… Shows are only available at a certain time in a certain venue, and then they go to another window, another venue for a period of time. But in fact, the history shows us that every time one of these new innovations came along, it expanded the market and expanded the appetite for media consumption. So in fact, their businesses continued to grow and thrive. Even though they fought tooth and nail against these in- [inaudible 01:08:33]

Jim: That is so funny. They fought against the DVD, I remember that one. That for a while was their cash cow.

Rob: That’s exactly right. A few years later, that was generating billions of dollars from movie studios until streaming came along, and now that’s the new thing. So yeah, I think we can be excited about what’s coming. I think for artists, it’s a better story than people think or than what you’ve heard. I’m not a big believer in this apocalyptic doom saying about AI. I’m pretty skeptical about that concept. I do think people will be displaced, that’s inevitable. But that’s actually one of the interesting things. I was doing a little research into how many jobs are destroyed every year, because we always hear that AI is going to steal all the jobs. Do you you know how many jobs are destroyed in a healthy economy? You want to take a guess.

Jim: What do you mean when you say destroyed?

Rob: Okay, so we know that a growing economy is going to destroy jobs through automation and efficiencies of various- [inaudible 01:09:26]

Jim: That’s just companies going tits up, right? That happens-

Rob: No, no, no. That’s where we automated job away. So that you had a machinist doing something one year, and now you have a robot doing that job-

Jim: I can come up with a number on that probably. I’m going to say it’s on the order of, I know about how many of those 13 million over a period of 50 years, I’m going to say about 300,000 a year.

Rob: So of all jobs in the United States economy, it varies. But the range of jobs that are destroyed each year ranges from 4% to 15, 15 strikes me as high. These are estimates-

Jim: That seems impossibly high.

Rob: That seems impossibly high to me too. The 4% number even that struck me as higher than I would’ve guessed-

Jim: That would be four billion a year, I don’t think so.

Rob: But think about that for a second. So 4% of jobs are destroyed each year, and that’s just through automation or through efficient process, or-

Jim: That means in 25 years, all jobs are destroyed.

Rob: That’s the point, Jim. That’s exactly the point, because you keep reading these scary doom and gloom reports that come from universities like Oxford saying 300 job types are up for automation, or AI will displace 40% of the jobs or something. Well, the reality is, in a healthy economy that would occur faster anyway, that would occur in 25 years no matter what.

Jim: I like that. Actually, I don’t… That’s why I think the four percent’s high.

Rob: And the jobs that are gone, we have no memory. We live in a time of instant amnesia. Most people can’t remember stuff 10 years ago. There are names baked into our cultural history. If you think about people’s names, there are people with names like Baker and Butcher. Your last name often was your profession in the past. And there are names like Chandler and Fletcher, and people forget that a Chandler-

Jim: A Fletcher is a person who put a feather on an arrow, is a Fletcher.

Rob: It was an arrow maker. And the Chandler was a person who made candles. And guess what? We don’t have those jobs anymore.

Jim: As we talked about earlier, buggy whip makers.

Rob: Exactly. Well, and it’s not just buggy whips. So if you think about the automobile, there was a whole horse industrial complex of- [inaudible 01:11:26]

Jim: There were 200,000 horses in New York City, and famously in the late 1890s, there was one of the first international conferences on the threat of horse shit on urban streets. And somebody did a calculation that by 1960, the horse would be 10 feet deep in every major city in the world.

Rob: And think of all the humans that worked on horses, because there were people who bred horses and people who groomed them and people who drove them, and people who put shoes on them and fed them and so forth. So there was a horse industrial complex.

Jim: Yeah, it was big. It was a big [inaudible 01:11:57] it was a measurable percentage of the GDP.

Rob: Yeah, exactly. And in 1910, when the earliest automobiles were going around, it didn’t seem like that was going to be a disruptive threat to the horse economy. But by 1920, it most certainly was because there was a big spike in the number of cars sold in that decade. And then it became quite evident that those jobs were going to go away. Now, it happened slowly because even up to World War ii, we still had horses in the economy. Some of the armies in Europe during the beginning of World War II had horses on the field.

Jim: Even the Germans did. Significant amount of the German transport in the first part of the war was by horse.

Rob: That came to an end in the 1950s. So that’s a long interval. That’s a half a century of time. I think when we’re thinking about the disruptive effects of AI on the workforce, it’s useful to think in increments, maybe not 50 years, but certainly 10 years. I just don’t see people getting displaced this year by ChatGPT. Maybe some jobs, maybe like people who write copy for social media marketing kind of in a way, good riddance, honestly. But I don’t think professors are going to get replaced. In the long run though and where it gets kind of fun to speculate about it, think about how AI could help with education.

We talked about education over the previous episodes, and if you think about a talking book or the idea of a young lady’s primer or a young boy’s primer and a book that grew up with you. That could help you understand, it could see if you understood that, could ask you questions to test your understanding for you personally, not for a whole classroom that companion could age with you and maybe be a little bit progressive in the sense that it could keep challenging you to grow. This could be the greatest innovation in education. Because honestly, when we spoke about education, much of it hasn’t changed. The lecture as a format is 500 years old. The book as a format is 500 years old. Surely we can do better now with our network technologies and maybe generative AI is the path forward for that.

Jim: Yeah, I think that’s a huge opportunity. We’ve already talking about it. Both of us are already using it to teach ourselves stuff.

Rob: Yeah, that’s right.

Jim: You want to find about what does Baudrillard think about the stack of simulacra and simulation and what’s the difference between those two? GPT will do a hell of a lot better job than plunking around on Google will do for you. I guarantee you. Again, another friend of mine who’s a physicist, he says, “God, I wish I had all this stuff when I was 11. I could have learned the PhD’s or the physics by the time I was 16.”

Rob: Yeah. I think that’s going to happen. You’re going to start to see prodigies emerge because they’ll teach themselves. Because candidly, if you’re a really talented student right now, if you’re gifted, you’re not getting tested to the max, you’re not getting pushed to the max in the public education system. You’re not because they just don’t have the resource to do it.

Jim: In fact, a lot, especially on the coast, they’re moving away from that. California’s stops teaching accelerated math, for instance. What a ridiculous things that is.

Rob: Wow. They call that a own goal in European football, and I think-

Jim: Right. Oh, well, because everybody can’t do calculation. We should offer- [inaudible 01:15:01]

Rob: It’s like that Kurt Vonnegut book, right?

Jim: Exactly.

Rob: We’re going to cripple everybody to the same level so that we can lower the common denominator. Oh, brother. Well, Jim, it’s been a great pleasure chatting with you. As always. I enjoy these conversations immensely, and I sure hope your audience does. If people are interested in more of my work, they can follow me. My sub stack, I publish a newsletter each week. It is called The Owner’s Guide to the Future, and I also, I’m doing the podcast of my own, The Futurists, and it’s now the number one rated future focused podcast in the world. So I hope people will check that out as well. And thanks for having me, Jim.

Jim: Yeah, I forgot to tout those in the opening part. We will certainly have links to both up on the episode page. And interestingly, I didn’t even know The Futurist was Rob’s podcast, and I got, somebody reached out to me to be a guest on The Futurist, and I looked at it and they said, “It looks okay. Sure, why not?” Had my assistant book it. So you’ll hear good old Jim Rutt on the podcast-

Rob: If you’re a little log rolling, I’ll interview you for The Futurist. Thanks for having me on your show, Jim, it’s always a pleasure to reconnect with you. I enjoy it.

Jim: Yeah, it was wonderful, high energy all over the place conversation.