How AI is making copyright issues more complicated | Devcom panel


AI and games has become the hottest topic at game conferences like Devcom, the game developer conference that precedes the giant Gamescom expo every year in Cologne.

At the event, I moderated a panel that explored the notion of the limits of intellectual property — exploring concepts with legal experts on questions like whether it’s OK to claim fair use in copying an image that is ubiquitous in the public domain.

We explored current copyright and trademark protection law when it comes to normal copycat issues and how the law is expected to be applied to AI creations, which a U.S. court has already ruled cannot be protected if no human is involved. These are the kind of issues likely being discussed by the SAG-AFTRA union negotiators who are looking for AI protections for actors from the video game companies.

Meanwhile, game and AI companies are racing ahead to develop better AI technology that can be used to automate tasks and gain efficiencies. The legality and ethics behind these developments are likely to trail behind, but it’s just as important to figure out these issues before we start having regrets.

Join us for GamesBeat Next!

GamesBeat Next is connecting the next generation of video game leaders. And you can join us, coming up October 28th and 29th in San Francisco! Take advantage of our buy one, get one free pass offer. Sale ends this Friday, August 16th. Join us by registering here.

Our panelists included Pieter Koornhof, COO of 24-Bit Games; Nav Sunner, CEO of Navatron; Michal Pekala, advocate, a partner in the technology department of Rymarz Zdort Maruta and the head of the Video Games & Entertainment practice; and Andrea Dufaure, counsel at A&O Shearman.

Devcom AI panelists (left to right): Dean Takahashi, Andrea Dufaure, Pieter Koornhof, Nav Sunner and Michal Pekala.

Here’s an edited transcript of our interview. Disclosure: Devcom paid my way to Cologne where I moderated two sessions at the event.

Pieter Koornhof: We mostly work with publishers trying to port their games to a variety of platforms, but I’m a recovering lawyer and recovering academic. Used to represent some video game clients. Used to write on aspects of IP in video games, looking at modding and infringement and AI, things like that.

Nav Sunner: I’m from a company called Navatron. I do business, legal, and investment stuff. I’ve been in the industry about 26 years working for various game companies. I also helped to run a studio at one point. Most of what I think about and do is video games.

Michael Pekala: I’m a video game lawyer based in Warsaw. I’ve been doing this for more than a decade now. I deal with legal stuff related to making video games from all angles. Developers, publishers, service providers and so on. I mainly deal with IP stuff revolving around video games, including infringement, registration, and plagiarism.

GamesBeat: How many lawyers do we have in the room? That’s appropriate. This is an interesting topic. Some of it comes from the launch of Palworld this year. It was widely described as “Pokemon with guns.” Nintendo took notice of this. They said they were looking into the issue of IP rights. That was a rare statement for them to make. We’re going to get into that. But that maybe captures some of this topic, what we’ll talk about today.

Palworld has formed a joint venture for merchandise licensing.
Palworld has formed a joint venture for merchandise licensing.

One of the things that–I don’t know if this is apocryphal or not. I believe it’s true. Someone who used to work at Rockstar Games told me that years ago, they could not get the rights to cars for their games. It’s kind of obvious why. They were crashing these cars. You could use those cars in robberies or to run people over. This was not the kind of thing that brand licensors were interested in back in those days. I remember there were a lot of car games where you just couldn’t crash a car. You couldn’t damage them.

Rockstar’s rule, reportedly, was that if they created a car on their own, they wanted to make it look like a car from the real world. But it had to be at least 60% different from the car they were thinking about copying. What do you think about this? Do we have some parallels?

Sunner: Approaching this question as far as whether it’s true or not, what percentage of the work is different from the original one, that part is a myth. There’s no such thing as 60% original, or 10% or whatever. It’s not the percentage that differs from the original work. It’s what elements of the original work have been taken and used in a new game. How much of that is substantial to the original work versus the new one?

There was a case in the EU that I can comment on. The CJEU ruled in 2019 that–there was a case involving the music of Kraftwerk. The defendant used two seconds of a Kraftwerk song. That was enough for the court to say that it was infringing. Those two seconds were so distinctive that anyone who heard those two seconds would immediately recognize the original work. The same logic applies to video games.

To answer Dean’s question, then, it’s thoroughly possible that even if you make your own game that’s 90% different from a previous game, if you use elements that are similar to the core elements of the original game, it could be enough for infringements to happen.

Sunner: I agree with that approach. Even from a U.K. perspective, we have similar ideas, even after Brexit. We have substantial takings as well. It’s a qualitative test, not a quantitative test. It’s really about, what is the main thing that makes something distinctive within that work? What makes it stand out?

When it comes to cars, this is an area I have a lot of expertise in, because I’ve spent a long time working with games companies that do racing games. Formula One, WRC and so on. When it came to the designers of these games, when I was in-house working for some of these companies, what I would always say is, “Don’t just take the badge off the BMW and think that’s enough.” Think about it as if you’re a car manufacturer and you’re trying to design a car. From a practical point of view that’s the safest thing. Internally there were some systems we had where we looked at specific things like an engine in a certain way, or the exhaust or whatever. It can be like that where you have time and resources. Most people don’t, though, so they end up making things that look like Ferraris or Lamborghinis.

Back in the really old days, when I was just getting in the industry, things looked a lot blurrier. One thing I would say about things that were blurry–now some of those games have been remastered. Now they look more infringing than they did. Infringement can creep up on you when you least expect it. Being lazy is not an excuse. When designers are making these things, they need to think about the potential hazards of infringement. The car itself, just taking cars as an example–a lot of time and effort went into the designs of those cars. The branding, the marketing. If you just slavishly copy it, some of these very large companies, particularly with some very famous marques, will come after you. It’s a dangerous area to be developing games.

Andrea Dufaure: Similarly, in France, looking at two games and assessing whether there’s a copy or not–there’s no 60% rule. I heard a client say they try to make at least seven differences between two games. But there are no rules. It just helps. If something is 60% different, you have more arguments to put before a judge. If there are a lot of dissimilarities, that helps as well. But I agree that it’s qualitative. A judge will look at what is very original in one game, what distinguishes it from its competitors. Did someone take those specific elements? It’s a case by case basis, and it varies a lot.

Koornhof: What’s also important to understand is that when we talk about IP, we don’t talk about one thing. Different forms of a game can be protected in different ways. When you look at the code, the visual assets of the game, those things are protected by copyright. Copyright protects not an idea, but the original expression of that idea.

Let me give an example. I’m thinking of a game that has a bunch of cute little creatures running around that you can catch, train, and use to battle with other people. What game am I talking about?

Digimon Original Digivice.
Digimon Original Digivice.

Audience: Digimon!

Koornhof: Yes! That’s an idea. That idea in itself can be abstract, can be generic, and be originally expressed in different ways. That’s how copyright works. Cars are usually protected by design. You’re not looking at protecting the function, but the aesthetic form. Then you can go on top of that. Sometimes when we develop engines or when we have nominal mechanics, those things can be potentially – not in all countries – protected by patent, because there’s something there that’s new and never been done before. There, to a certain extent, you can protect the idea, because you’re the first one that came up with it.

Last but not least, when it comes to franchising and reputation management, is trademark. Trademarks can be visual. They can be words. You’re linking your brand, your reputation to a particular thing, and because you’re doing so you want to protect that. These things will often overlap in games, because we build IP. We build a game that has mechanics, but we also want, if it’s good, to get a TV show or a movie. We want to make sequels, a franchise, merchandise. There are layers of complexity. That makes it easier to protect, but also sometimes easier to get away with certain things.

Dufaure: These are taken from real cases in France. We’ve had several disputes in the past year, especially around hypercasual mobile games, initiated by Voodoo, a French mobile games publisher. I have four cases. I’ll give a brief background and ask whether, according to you, it’s infringing or not and what the court would say.

This is a mobile game. It’s about woodworking. Here, in this case, Voodoo alleged it had created a unique game experience for its players about woodcutting and woodworking. You would choose the log you wanted, the wood you wanted, and then use your tool – a chisel here in this case – and paint and polish the carved piece after that. That’s the background. You can see that the defendant’s game had a similar system. You could also cut the wood, use a chisel, carve it, polish it, paint it. What do you think?

Sunner: The thing is, when you look at it from one perspective–these cases are always very difficult. They’ve kind of flipped it over. The layout is so similar. But when it comes down to–there was a case that happened in the U.K. involving pool games. In the game of pool you have various pockets, cues, things like that. There are so many generic elements to the game of pool that it’s difficult to make another pool game that doesn’t look the same. The question is, are there any elements within that that are infringing? Most of the time, most generic games won’t infringe on other generic games. That’s where I’d go with this.

GamesBeat: The tree kind of bugs me, as well as the chisel, the wood, and the arrow. It feels way too similar.

Donkey Kong

Sunner: These guys were really lazy. They could have been a bit more inventive. But that’s not enough to win in court sometimes.

Koornhof: The history of video games is filled with people stealing, but doing it in the right way. Were it not for a failed Popeye licensed game being turned into Donkey Kong, which resulted in a lawsuit around the use of a character that resembled King Kong, and a man who subsequently argued that this was a generic thing – a big ape, everyone’s used it – his name was John Kirby, and he inspired a character himself as a result. Lawyers are big fluffy marshmallows that eat everything in sight and then gain superpowers.

I would say yes, if you look at certain aspects of this as generic and commonplace, it probably isn’t infringement. But if you go further than that and look at the visual layout–remember, when we make a game, it’s not just the code. It’s also the assets and the mechanics. You can create confusion, and confusion can be relevant, because in certain countries, creating something to intentionally confuse, intentionally take away custom, and intentionally use the goodwill someone else created, that can be the cause of liability. That’s the case in France, is it not?

Pekala: And the EU as well. Potentially it’s not a copyright infringement, but it may be an infringement. There is an argument to be made here there’s no infringement on the copyright level because the original work didn’t meet the basic criteria of originality, which is a requirement for a work to be protected by copyright. If the court finds that Voodoo’s original work lacks that element, anyone else who copied those elements could not infringe. The protection on that level was never initiated.

Sunner: To wrap, part of the problem with making video games is that they’re international. The biggest issue, particularly if you’re working for a global games publisher–even if it’s a small publisher, they might be selling across the world. We have a risk profile for various countries. We know that in certain territories it goes beyond what the law would say in, for example, the U.K. You take an assessment based on that, whether you would carry on or not sell in that territory.

In the U.K., certain things may be held not infringing. But there are these other things we call passing off, which is a very complex claim. That isn’t copyright. It’s more like a trademark claim than a copyright claim. Because of all these complexities, we get these arguments. Sometimes you can be infringing.

Koornhof: Sometimes, when a lawyer says “It depends,” it really does depend. We’re not just being facetious.

Dufaure: The court ruled that there was no copyright infringement. Part of what we do is also looking at prior video games, what came before. Were people doing this before Voodoo? How does a woodworking game work in general? You find these elements that they’re claiming a monopoly on. The court ruled that there was no copyright infringement. The elements were much too simple. The judge said, “How do you do a woodworking game if you don’t have a lathe and a chisel?” Voodoo was simply translating into a video game what would happen in real life. You’re correct on that.

You’re also correct that the defendant was nevertheless sentenced on the grounds that you discussed, of unfair competition or free-riding. That was regarding elements that were specific to the case. In particular, the defendant had changed the version of its games many, many times to resemble or to differ from the prior game, which made the judge suspicious.

Let’s go to the second one. In this second one, it’s important to know that the defendant was not represented. That could play a part. They didn’t have the opportunity to defend themselves. This was the cube surfing case. In both games you can find the similar concept of cube surfing. The cubes are yellow. The obstacles are similar. The character is wearing green in both cases. You have a point system on the top of the screen. What do you think?

Stephan Reichart, head of Devcom.
Stephan Reichart, head of Devcom.

Sunner: To touch on this, the way copyright works is that the more detailed the work becomes, that’s when the chances of infringement increase. With the woodworking game, the lathe and the chisel are very familiar. As the designs get more complex, like they are here, they’re getting more copyrightable. Even the angle at which things are pointing. Why did they choose that angle? Is it functional? Is it necessary? That’s what you have to think about when you look at infringement. As things get more detailed, the chances of infringement can increase.

Dufaure: In this case the audience is correct. It was deemed a copy. It’s important that the defendant wasn’t represented because the judge is not going to do the defendant’s work for them. Maybe there were other games that also featured similar details, but nobody was there to show them to the judge. When you’re not represented, it means the court only hears one side.

This was a sniper game. It was also Voodoo initiating the action for this one. Voodoo claimed a monopoly on several elements. One, it was a first-person shooter. Two, the fact that you could not see the shooter or their hands or anything, but just the sniper. The fact that the player was shooting from the top of a building. The fact that the coins were on the top right. The fact that there was a three-step shooting process – touch and hold, touch and drop, touch and release to shoot. The fact that there was a zoom on a particular target. The confetti that would spray when you won at the end. It wrote “Head Shot” when you scored a head shot. The fact that living characters were displayed in red and dead characters in black. Those were all similarities put forward by Voodoo.

GamesBeat: The shooter genre is so vast. I have to believe this has all been done before.

Sunner: Did they play the games?

Dufaure: That’s a very good question. It depends on jurisdiction. In France we have specialized judges in IP, but they’re not specialized in video games. A part of our job is to explain to them and get them interested. Usually judges are a bit older. They have their own perspectives on video games. The common way to do it would be to show pictures, but that doesn’t always reflect the gameplay. How do you explain that to a judge? We would always file some exhibits with video, hoping that they would at least watch that, if not actually play the game. In this case I don’t know because I wasn’t part of the action.

Sunner: One of the interesting things here is that when you look at something static, as opposed to while it’s moving, and actually being involved in that movement–the sequence of events is exactly the same. Is that infringing? Copyright doesn’t protect that in its essence. There have been cases where they’ve tried to say it’s a dramatic work and other elements like that, but it’s a very tricky part of the law.

Koornhof: What he’s saying there–what a video game is legally classified as, that’s interesting. In copyright, do you think there is a legal category for video games under the law? Usually copyright protects a category of original work. There’s film, books, music, artistic works, dramatic works. There is no category for video games, because as she just told you, judges are old. Parliamentarians are terribly slow. They’re nice people notwithstanding. But things are slow to change.

When we have to put a video game in one of those boxes, which box does it get put into? It depends on the country, really. Some countries would say a video game is a film, because it’s a bunch of moving pictures on a screen. Back in the day when these judgments were made, we put cartridges in a console, just like you put VHS tapes in a VCR. It made sense to people at the time. Others will say that a video game is software, which means it’s protected as a book in some parts of the world. That’s what code is. It’s just writing. As a result, it gets interesting.

Sunner: It makes you think about the constituent parts of a game. You have the software. You have the art. You have the sound and music. All these layers are separate rights in themselves. All of them can be protected by copyright. They’re all copyright works. On top of that you may have, going back to the cars–there was something I used to have to check every year, which was the ghost car patent. When you race in a game you’ll see the ghost of your lap. That was protected by an Atari patent until that expired. There are all these layers. That’s one thing you have to think about when you’re thinking about infringement.

Dufaure: The French judges said this was a copy. The list of elements I gave you was what the judge held as being decisive. That might be quite surprising. Some seem individually very common in a first-person shooter game. But put all together, it could be classified as a copy. Also, the judge said that they could have done things differently. He didn’t believe that this was all a coincidence to have all these things in the same game.

Also, the Voodoo game was released, and then the defendant’s game arrived just a few weeks later. That can go both ways of course. You might say it’s not a copy because it’s part of a larger trend behind a certain type of game. Alternatively, depending on the lawyer, one could argue that they couldn’t get a copy out so quickly after the first game entered the market.

SAG-AFTRA is on strike against game companies over AI use of talent.
SAG-AFTRA is on strike against game companies over AI use of talent.

One last case. These are both games where you can build your town. Voodoo argued that the similarities were that in both games you had a character that had to chop wood. That character had an axe. That character had a backpack in which he could put logs. You could trade logs for points. Those are the similarities. There was a big difference in the scenario, meaning that one game was created where you had to cut the logs and build, and in the other one you had to defend yourself from enemies. What do you think?

Pekala: I think it wasn’t. Or at least it shouldn’t be.

Koornhof: You can copy some things, but not others. That’s what I’d say. Think about a collage, where you take a bunch of–this is how we can finally get to AI, I suppose. If you take a bunch of pictures that other people took, that they composed, and they own the copyright, and you make your own image with them, make a new picture, you were the person that created this original expression of that idea. That collage is owned by you, potentially. But also, you’ve infringed a hell of a lot of copyright to get there. You can have a situation where you can do both.

I’d say you can probably take some of it, but the rest is fine. Am I right? I’m usually wrong on these things. That’s why they put me in, to make everyone else look good.

Dufaure: The answer is no. Indeed, the judge deemed that there were too many differences in the games. Just having a backpack with wood inside can’t be sufficient for it to be a copy. Thanks for playing!

GamesBeat: Who wants to take us into AI? Can we get a primer based on what we’ve discussed so far?

Dufaure: Would any of these rulings be different if the games were created with AI?

Abstract AI image showing a desktop Mac monitor displaying a sheet emitting orange flame effect against a purple red backdrop
Credit: VentureBeat made with Midjourney

Sunner: Let’s start with what people think is the actual issue with AI. We can talk about what we’ve been looking at. I was doing a talk last year on AI. When a work is created by generative AI, is it actually protectable? There’s that issue. There’s the data set itself, the data collected to build the AI and generate the work. That’s an issue. There’s the person who prompts the model. Do they have any rights? There’s a number of these layers. There’s the output that’s been created. Is that infringing anyone else?

One thing I would say, before we answer all of those different questions. When I was in-house at various game companies, marketing people were sending me things. Design teams were sending me things. I was seeing stuff that made me think, “That looks just like so-and-so’s game.” You still have that test, regardless of whether AI made it. The actual output is still governed by the question of whether it infringes someone else’s work. Having AI has just created more problems from a legal point of view.

Some countries don’t even recognize AI output, because they say it’s not made by a human. In the U.K. we have a category called computer-generated works, but when you look at the Copyright Act it comes from, which is from 1988, it’s to do with the person who built the system. The person who built the system would own the output. We don’t want that to happen. Most companies want the people who create the prompts to own it. The actual data set, no one really thought about that. In the U.K. we had an act that dealt with satellite data that was then generated through a computer system. That didn’t belong to anyone. Now we have a situation, though, where data sets potentially take data from lots of people without any consent.

GamesBeat: Let’s answer that first question you raised. Can AI output be protected by copyright?

Pekala: I don’t think it’s answered that easily. One important thing to note is that different countries have different approaches. The majority of countries, at least so far, say that AI-generated content is not copyright protected, with some distinctions. However, saying that AI-generated content is not copyright protected doesn’t mean that the same content cannot infringe other works that were created before it. That’s very important.

The output of generative AI is a result of a data set that’s been populated and prompts that query it. What’s been put in the data set, that’s the first thing to consider. Where did the data set come from? Was it legally obtained? Do you have permission to use it, or do you need permission to use it? Does it come from the public domain? If any of those permissions are not given, there’s potential for a problem in the future.

Koornhof: Putting aside the question of whether an AI asset or code will infringe, because I think the short answer there is it depends on what the AI is trained on–the question of whether an AI can create copyrighted IP–well, again, it depends. Here’s the question. What does the word “original” mean? Again, it depends on the country. Different countries ascribe different meanings to originality.

In some countries, expending the sweat of your metaphorical brow means it’s sufficiently original. You spent time and effort and skill. Other countries, and the EU, say it needs to be your own intellectual creation. If I’m good at telling an AI to make something unique, something cool, and I’m not just putting in a generic instruction, but putting in paragraphs of prompts in great detail, I would argue that you could be using it as a tool to create. Just like we use any other toolkit to create.

Dufaure: This is a hot topic because for most studios your copyright over your game is your biggest asset. A lot of legal teams are doing guidelines for developers, but also struggling to see them enforced. The reason is that generally, all over the world, to have copyright you have to be human, which an AI is not.

That type of question is not new, to some extent. There are cases in France where, when digital cameras first came out, everyone wondered if the camera was taking the picture, or was it the person behind it? Similarly, if you compose music with electronic assistance, is that doing it yourself? Each time these questions have been raised, in the end a judge said that there is someone behind all this ordering the machine.

I won’t give the “it depends” answer, but kind of? With AI it depends on what you’re prompting the AI to do. If you have pages and pages of prompts which are super detailed, where the AI just helps you get to what you had in mind faster, that could survive the test and be protected by copyright. On the other hand, if you just say, “Create a fun character,” the output won’t be something that you thought up. If the AI is doing that much work, it would be hard to survive the test.

In France we don’t have a copyright office. There are no cases yet. In the U.S. there have been. That’s been the test applied to figure it out. Did you work on it in Photoshop or another piece of software afterward? What did you ask the AI to do? Those are concrete questions that the office has asked.

Sunner: Part of this is whether the data set is infringing, but let’s say it’s a clean data set. The actual output that’s created–we’ve had a lot of discussion going on. If the AI has created it, but then you alter it and work on it, a human worked on that. Then we’re back to the question of how much work went into it. If it was done as a concept, the AI creating a concept – which happens all the time in games now – and you create a work based on that concept yourself, that’s when you’re starting to get back into normal copyright law. A human is doing something inspired by something that was generated by a computer.

What we don’t have is an answer across the world about how much effort that is. It stems from the problems we have with the law where in some countries originality means one thing and in others it means something else. I’m sorry that we can’t answer that question, because I’ve gotten it many times.

GamesBeat: I’ve heard that some big companies have banned the use of generative AI in their studios. They’re afraid that something might come out that is infringing. Do you find that to be the case for a lot of the bigger gaming companies? They’re afraid to use AI now? Or are they going forward?

AI cartoon style art of shaggy brown haired man typing on computer in garden of strawberries
Credit: VentureBeat made with ChatGPT

Pekala: I can confirm that a lot of video game companies specifically forbid their developers to use any AI for generative content. They can use it for support activities, but not for anything that’s meant to be creative. The thinking is twofold. One, as you said, it’s because of the risk of infringing on the rights of others. The other issue is the lack of ownership of the content you might create.

The debate will keep going for some time. I’m afraid that at least in Europe, without a concrete new approach to legislating from the EU, we won’t receive any clear answer from the courts. They don’t have the tools to assert an answer. It might happen in the U.S. and other countries. As a result, a lot of developers have decided to just drop AI for the time being. But they also advocate that generative content is still not at the same level of quality as human-made content.

There was a recent interview with the lead quest designer for Cyberpunk 2077. He said that they would never use AI for creating quests, even if they could, because it’s not the same level of creativity as a human could provide. Out of all the quest concepts created by humans at CD Projekt Red, he explained, they only take five percent to develop and discuss further.

GamesBeat: Does this mean that game companies aren’t using AI at all? Or can they use it for internal-facing purposes, like concept art, which is never going to directly end up in a finished product?

Pekala: Out of all the clients that I work with, there are none that aren’t using AI at the moment. Everyone uses AI to some extent. Depending on the company they might use it more or less. There are specific tools these days that can’t work without AI, to support finding bugs for instance. This is very helpful. It helps the QA team pick up on those loopholes.

At the same time, there are developers that use AI for creative output to make their lives easier, because it’s easier to just ask the AI to make assets. It also happens that developers don’t disclose that information. They keep that information from others and from the public, because at least for now, it’s very difficult to distinguish some elements that have been created by AI from elements that have been created by humans. There’s also this element of not only having good intentions when using AI. But it’s like any new technology.

Sunner: Going back to what you actually see, what you actually create–for example, if you have a scene with lots of very small stones, pebbles on the shore, sand, who is going to sue you and say, “That’s my sand”? I had a case, very early in my career, where I got a letter from the forestry commission. They said that the trees in our game infringed upon the trees in their protected land. That obviously went away, but–I said, “Okay, if you can point to which tree in which forest ended up in our game…”

Koornhof: It’s very rare that plant breeding rights come into video games, but sometimes they can.

Sunner: When you take Getty images from their bank of things and you license a photograph of a person, you still have to get the rights to that person. You don’t have to get the rights to a plant. Unless you can show exactly that that photograph you’ve taken is an exact copy and the rights are owned. There’s technology now that I think is being used in universities to detect plagiarism, but also for showing whether generative AI has created an artifact.

GamesBeat: I wanted to ask about the issue of fair use. If I take a picture of this bottle, which has a brand on it, I still created that picture. If I put that in my data set, am I infringing on the brand in doing that?

Sunner: The problem is copyright and trademark. In a game, normally if I was going to put that bottle in the game, then that would be quite an issue. The design of the bottle is unique. Sometimes shapes are protected, like a Coca-Cola bottle. If this is a generic bottle that’s fine. But then on the bottle you have a label with a name. You need to change the name, or you need to get a letter of consent.

Sometimes people will say, “Yes, I’d love to be in the game.” You’d have hundreds of these consent letters. Even tiny little things on uniforms. This has all been done in the past. But when you have AI, the AI doesn’t do it. It just trawls through and pockets things. A clean data set is the way to go, and you have to do that yourself. You can do that with your own game. You can take buildings you built in your first game and use them in a data set for the sequel, using generative AI to create similar buildings. That will happen.

GamesBeat: We’re about to go into this world of user-generated content. Roblox is already there. Are we going to restrict what users can do?

Koornhof: You’ll find, in the terms of conditions–now we’re going a bit off-topic I suppose. But with UGC, the terms and conditions will first say that if there’s something IP-infringing and we find out about it, because we’re potentially liable, we’ll remove it. If we need to pass that liability on to you, by the way, we’ll do that too. But also, some of these licenses – not all of them, but some – will say that you give us a license and we’ll use this for our own stuff later on down the line, thank you very much.

It’s a big risk. When you have a game that allows for UGC, potentially someone could put Spider-Man in the game. That’s Marvel’s. You can’t do that. There are risks which you need to guard against. There are also laws in certain countries that say you’re just a platform, you can’t control everything that’s put there, and you’re not actually telling people to go out and infringe copyright. If you remove it when you know about it, then you’re safe. But some countries don’t have those laws. If there’s something on your platform that infringes, you’re potentially liable. It’s a tricky one.

AI golden age sci-fi style art of humanoid robot wearing glasses repairing watch with small. jeweler's screwdriver
Credit: VentureBeat made with OpenAI ChatGPT

GamesBeat: Does anyone have questions we haven’t asked so far?

Question: I wanted to ask if you’d looked into the subject of zero knowledge proofs. In cryptography there’s ongoing research to prove computation mathematically, including the training of AI models. If that progresses, we could reach a point where you can have audits, where someone has to generate a mathematical proof of how they trained their AI model to comply with regulation. If you don’t do that, you’ve violated some variation of copyright automatically.

Sunner: I’ve been speaking to a couple of companies that are looking at identifying whether something has been generated by the AI. I don’t know the technicalities behind it, but I assume that would be what it is. All of these things may get used. The problem is always, is it viable? Will that be regulated itself, to show that it’s a viable tool to check someone else’s work?

Pekala: Until we have introduced and established standards we can refer to, that the industry will agree to abide by, a lot of these cases will just get away with it for the time being. We’re at a very early stage in using AI. Some will take advantage of that, while some others may potentially be caught even years after, if new technology emerges to allow for checking AI-generated content. Anything is possible at this stage.

Koornhof: Commercially speaking, people are probably already working on that. It might go quicker because–a lot of the people here at Devcom that I’ve talked to, they’re all trying to solve cheating, griefing, toxicity. These are all issues in games. Using things you shouldn’t be using through generative AI is also becoming a problem for companies. There will be a reputational risk, a legal risk. There already is. Those tools are being worked on as we speak. I would be surprised to hear if there aren’t people in quiet rooms here or during Gamescom already talking about how to go about that.

GamesBeat: Zero knowledge proofs are also becoming very interesting in blockchain games.

Question: I wonder about a game where the artwork is playing on popular culture or memes, things like that. Things that are parodic, but maybe still infringing. Does that also depend on whether a game is free or paid?

Koornhof: In one jurisdiction and one jurisdiction only, it may make a difference. Dean mentioned fair use. Fair use is a term that’s commonly used when talking about copyright law, but it’s actually only applied in the United States. Some other countries are now thinking about bringing in fair use to a limited extent. In the rest of the world we use set lists of limitations and exceptions, things you can and can’t do.

In the context of the fair use test, the question of whether or not you’re doing it for profit can be relevant. It’s not always relevant, but it can be. The short answer is no, it doesn’t really make a difference if it’s free or not. Even if it’s free, there are other ways of making money with things in software and games, as we all know.

Dufaure: I had a case involving a card game. Some card designs are very, very old. The first tarot cards, things like that. Those models are usually in the public domain and you can use them. But companies are smart. Every 20 or 30 years they change their models a bit to meet a new copyright term. You have to figure out which models or designs you’re taking inspiration from. Are they old ones or more recent ones?

Abstract gradient artwork blue and red tones, dark silhouette of woman with long hair leaping in profile off a GPU fan
Credit: VentureBeat made with Midjourney

Question: You talked about companies banning the use of AI. But what about the use of an offline model, so it doesn’t take work from the web or send my work to others? It’s only based on my work. What do you think about the risks of that approach?

Pekala: If you’ve taken your own work, you give permission to yourself to use it for AI-generated content. That’s fine. This is the clearest example of a clean data set that’s possible. You reuse your own work to make something new with AI.

Question: As a designer I do a lot of reference-gathering on projects. Previously we would scour the internet ourselves for hours looking for images, movies, TV, animation, stuff like that. With prompts and AI generation, obviously I can feed in a prompt instead of doing a search and produce a bunch of different concepts. The reference might be more accurate to what I want to achieve in a game. That goes to another department who will create the prop. Is there any potential liability there, because the eventual created prop or asset or level might be an original creation, but the concepts haven’t come, necessarily, from public domain assets, because I didn’t gather the reference directly myself?

Sunner: We touched on this a bit. There, what you’re doing–you’d be doing that anyway. You’re saving time by using a device, a tool to do the research for you. Again, it depends on the prompts. If you ask for a cute game involving an Italian plumber, it’s obvious what it’s going to generate. We’ll still look at what you’ve done in the end and see what that infringes. The element of taking the generative AI output and starting to work on it–it’s already been a step removed, because you’re just using it as reference. It’s further away from infringement. That looks okay at the moment.

Dufaure: Inspiration is fine. Also, to be a little less alarming, for the output to be infringing you have to put in a prompt that targets something very specific. The AI trains on so many different prior works that it will likely be a mix of many prior works. It’s less likely to produce something exactly infringing. You have to ask for that.

Sunner: Having said that, there were some tests done–I can’t remember which conference I was at. But many Sonic characters appeared. Donkey Kong appeared. It just goes to show you how these models are supposed to be clever, but they’re really not.

GamesBeat: If you use a data set that’s clean and one that’s dirty, so to speak, and give them both the same prompt, does the clean one produce something that’s legally protected?

Dufaure: If it’s based on your own work, it could be a derivative work of your own. Potentially.

Koornhof: It depends.

Disclosure: Devcom paid my way to Cologne where I moderated two sessions at the event.



Source link

You might also like

Comments are closed, but trackbacks and pingbacks are open.