To be the devil’s advocate (or GRRM’s attorney), I see the merit of his and other authors’ concerns. Chat GPT makes it feasible to generate short stories in their world and with their characters, which can easily replace their licensed products. This is not just their main work, but also other products that generates them some revenue stream.
Example: A friend of mine is using Chat GPT to generate short bedtime stories for his daughters. A typical request is something like this: “Please write a five paragraph story where Elsa from Frozen meets Santa Claus. Together, they fly in Santa’s sleigh over the world, and Elsa is magicking snow on all Christmas trees.” Normally, you’d buy a Disney-licensed book of short Christmas stories (I have one for my kids), but Chat GPT is more flexible and free.
Same goes for GRRM. He doesn’t write Children stories, but one can still prompt Chat GPT to produce stories from the universe, which scratch the ASOIAF itch. This substitutes the officially licensed products and deprives the author of additional revenue stream. Just for the fun of it, I prompted Chat GPT: “Hello GPT-3.5. Please write a four paragraph story set in the Game of Thrones universe. In this story, Jon Snow and Tyrion Lannister go fishing and catch a monster alligator, which then attacks them.” It produces a surprisingly readable story, and if I were a fan of this universe, I can imagine myself spending a lot of time with different prompts and then reading the results.
(On a side note,AI-generated content already has at least one group of victims: the authors of short fiction. Magazines like Clarkesworld were forced to close submissions of new stories, as they became overwhelmed with AI-generated content.)
i don’t see why ChatGPT would be at fault itself here. Taking the rest of your argument as granted, Chat GPT is more like a tool or service that provides “snippets” or previews, such as a Google image search or YouTube clips or summaries. The items being produced are of a fundamentally different quality and quantity and cannot really be used to copy a work wholesale. If someone is dedicated enough to price together a complete story, I would think their role in producing it is more important than the use of ChatGPT
copywrite law itself is broken and too broad as it is, I don’t think we should be stretching it even further to protect against personal use of AI tools. An argument can be made if an individual uses ChatGPT to produce a work which is then commercialized (just like any other derivative work), but the use of the tool by itself seems like a ridiculously low bar that benefits basically no one
You are right, especially regarding the copyright law. My argument here, however, was the same argument as companies are using against non-genuine spare parts or 3D printing (even though the latter seems to be a lost battle): people who are able to generate substitutes based on the company’s designs (you can say their IP) are eating into their aftermarket profits. That’s not even taking into account planned obsolescence (my kids toys are prime examples) or add-ons to products (I printed my own bracket for my Ring doorbell). With AI, I don’t need to buy short story books for my kids to read; I’ll generate my own until they are old enough to use Chat GPT themselves.
Yea, I mean I get why automated tools are bad for companies, I just don’t have any sympathy for them, nor do I think we should be stretching our laws beyond their intent to protect them from competition. I think the fair-use exceptions for the DMCA (such as for breaking digital copy-protection for personal use) are comparable here. Under those exceptions for example, it’s considered fair use to rip a DVD into a digital file as long as it’s for personal use. An IP holder could argue that practice “eats into their potential future profits” for individuals who may want a digital version of a media product, but it’s still protected. In that case, the value to the consumer is prioritized over a companies dubious copyright claim.
In my mind, a ChatGPT short story is not a true alternative to an original creative work (an individual can’t use GPT to read ASOIAF, only derivative short stories), and the work that GPT CAN produce are somewhat valueless to an individual who hasn’t already read the original. Only if they were to take those short stories and distribute them (i.e. someone ripping a DVD and sharing that file with friends and family) could ‘damages’ really be assumed.
I think the outcome of these lawsuits can help inform what we should do, also: LLMs as a tool will not go away at this point, so the biggest outcome of this kind of litigation would be the inflation of cost in producing an LLM and inflation of the value of the “data” necessary to train it. This locks out future competitors and preemptively consolidates the market into established hands (twitter, reddit, facebook, and google already “own” the data their users have signed over to them in their TOS). Now is the time to rethink copyright and creative compensation models, not double-down on our current system.
I really hope the judges overseeing these cases can see the implications here.
How will it know what grrm is if it hasn’t read the book or is aware of the content? Pretty sure it does need to read the book in order to generate content similar to the authors style.
Right, but would that be pursued if a human did the same thing? Because there’s a vast amount of fanfiction churned out by human hands, and it’s safe as long as they don’t try to monetize it. Seems like most of the fear is the existential threat that it might suddenly begin writing good stories, and destabilize the entire writing industry as people can just ask for exactly the sort of story they want. For free. (But not actually, because corporations will own all the AI and data).
Well, I mean we kinda are, capitalism and all that. There are thousands of authors of Patreon, Kofi, and the like that you can pay to write you the fanfiction you want. Further, if you don’t know the provenance of a fanfic, how do you tell which ones are the copyright violation? The only way to do so is if you have records of its birth, especially as generative AI improves.
I’m not blind to the plight of creators here, but isn’t the issue that a machine can, in theory, out compete the authors at their own style? If a random human can write Stephen King’s style better than Stephen King, it’s forgiven because that took time, effort, and talent, where a machine doing it alarms us. No author has ever been sued because they read a book and were influenced in their writing, unless they outright plagiarized without attributing. I just think that there needs to be a significant frame shift, since artificially limiting generative AI to protect the current business model instead of allowing it to reshape how people produce and consume media isn’t realistic. The issue is figuring out how creators are still compensated for their work.
People are already building small generative AI projects, so there’s no containing it, and it’s only going to grow.
To be the devil’s advocate (or GRRM’s attorney), I see the merit of his and other authors’ concerns. Chat GPT makes it feasible to generate short stories in their world and with their characters, which can easily replace their licensed products. This is not just their main work, but also other products that generates them some revenue stream.
Example: A friend of mine is using Chat GPT to generate short bedtime stories for his daughters. A typical request is something like this: “Please write a five paragraph story where Elsa from Frozen meets Santa Claus. Together, they fly in Santa’s sleigh over the world, and Elsa is magicking snow on all Christmas trees.” Normally, you’d buy a Disney-licensed book of short Christmas stories (I have one for my kids), but Chat GPT is more flexible and free.
Same goes for GRRM. He doesn’t write Children stories, but one can still prompt Chat GPT to produce stories from the universe, which scratch the ASOIAF itch. This substitutes the officially licensed products and deprives the author of additional revenue stream. Just for the fun of it, I prompted Chat GPT: “Hello GPT-3.5. Please write a four paragraph story set in the Game of Thrones universe. In this story, Jon Snow and Tyrion Lannister go fishing and catch a monster alligator, which then attacks them.” It produces a surprisingly readable story, and if I were a fan of this universe, I can imagine myself spending a lot of time with different prompts and then reading the results.
(On a side note,AI-generated content already has at least one group of victims: the authors of short fiction. Magazines like Clarkesworld were forced to close submissions of new stories, as they became overwhelmed with AI-generated content.)
Couple things:
i don’t see why ChatGPT would be at fault itself here. Taking the rest of your argument as granted, Chat GPT is more like a tool or service that provides “snippets” or previews, such as a Google image search or YouTube clips or summaries. The items being produced are of a fundamentally different quality and quantity and cannot really be used to copy a work wholesale. If someone is dedicated enough to price together a complete story, I would think their role in producing it is more important than the use of ChatGPT
copywrite law itself is broken and too broad as it is, I don’t think we should be stretching it even further to protect against personal use of AI tools. An argument can be made if an individual uses ChatGPT to produce a work which is then commercialized (just like any other derivative work), but the use of the tool by itself seems like a ridiculously low bar that benefits basically no one
You are right, especially regarding the copyright law. My argument here, however, was the same argument as companies are using against non-genuine spare parts or 3D printing (even though the latter seems to be a lost battle): people who are able to generate substitutes based on the company’s designs (you can say their IP) are eating into their aftermarket profits. That’s not even taking into account planned obsolescence (my kids toys are prime examples) or add-ons to products (I printed my own bracket for my Ring doorbell). With AI, I don’t need to buy short story books for my kids to read; I’ll generate my own until they are old enough to use Chat GPT themselves.
Yea, I mean I get why automated tools are bad for companies, I just don’t have any sympathy for them, nor do I think we should be stretching our laws beyond their intent to protect them from competition. I think the fair-use exceptions for the DMCA (such as for breaking digital copy-protection for personal use) are comparable here. Under those exceptions for example, it’s considered fair use to rip a DVD into a digital file as long as it’s for personal use. An IP holder could argue that practice “eats into their potential future profits” for individuals who may want a digital version of a media product, but it’s still protected. In that case, the value to the consumer is prioritized over a companies dubious copyright claim.
In my mind, a ChatGPT short story is not a true alternative to an original creative work (an individual can’t use GPT to read ASOIAF, only derivative short stories), and the work that GPT CAN produce are somewhat valueless to an individual who hasn’t already read the original. Only if they were to take those short stories and distribute them (i.e. someone ripping a DVD and sharing that file with friends and family) could ‘damages’ really be assumed.
I think the outcome of these lawsuits can help inform what we should do, also: LLMs as a tool will not go away at this point, so the biggest outcome of this kind of litigation would be the inflation of cost in producing an LLM and inflation of the value of the “data” necessary to train it. This locks out future competitors and preemptively consolidates the market into established hands (twitter, reddit, facebook, and google already “own” the data their users have signed over to them in their TOS). Now is the time to rethink copyright and creative compensation models, not double-down on our current system.
I really hope the judges overseeing these cases can see the implications here.
The thing is you can tell an AI to make a story like grrm and the AI doesn’t even have to read grrm. This is a losing battle.
How will it know what grrm is if it hasn’t read the book or is aware of the content? Pretty sure it does need to read the book in order to generate content similar to the authors style.
Right, but would that be pursued if a human did the same thing? Because there’s a vast amount of fanfiction churned out by human hands, and it’s safe as long as they don’t try to monetize it. Seems like most of the fear is the existential threat that it might suddenly begin writing good stories, and destabilize the entire writing industry as people can just ask for exactly the sort of story they want. For free. (But not actually, because corporations will own all the AI and data).
A human isn’t a product for use by other humans. It’s not the same.
Well, I mean we kinda are, capitalism and all that. There are thousands of authors of Patreon, Kofi, and the like that you can pay to write you the fanfiction you want. Further, if you don’t know the provenance of a fanfic, how do you tell which ones are the copyright violation? The only way to do so is if you have records of its birth, especially as generative AI improves.
I’m not blind to the plight of creators here, but isn’t the issue that a machine can, in theory, out compete the authors at their own style? If a random human can write Stephen King’s style better than Stephen King, it’s forgiven because that took time, effort, and talent, where a machine doing it alarms us. No author has ever been sued because they read a book and were influenced in their writing, unless they outright plagiarized without attributing. I just think that there needs to be a significant frame shift, since artificially limiting generative AI to protect the current business model instead of allowing it to reshape how people produce and consume media isn’t realistic. The issue is figuring out how creators are still compensated for their work.
People are already building small generative AI projects, so there’s no containing it, and it’s only going to grow.