George R.R. Martin sues OpenAI, cites AI-generated Game of Thrones prequel “A Dawn of Direwolves”

This week, we learned that A Song of Ice and Fire author George R.R. Martin was teaming up with several other prominent authors to sue OpenAI, the company behind ChatGPT. ChatGPT is at the forefront of generative AI; you can prompt the program with a few simple words and it will generate long, legible streams of text in response.

ChatGPT and programs like it work by reading vast amounts of text written by humans, synthesizing it, and then rearranging it in answer to human prompts. That text includes copyrighted works written by the likes of Martin and others. The authors do not appreciate their work being used in this way without their permission or remuneration. They’re striking back, alleging “mass-scale copyright infringement” on the part of OpenAI.

One of the things that Martin and his co-plaintiffs have to prove is that OpenAI actually did hoover up their books without permission. I thought it might be interesting to get into the details of how they’re going about that, focusing on Martin and A Song of Ice and Fire in particular. The full complaint is available to read online.

George R.R. Martin cites AI-generated versions of The Winds of Winter and A Dream of Spring in lawsuit against OpenAI

To show that OpenAI has absorbed Martin’s books, the complaint points to works of generative AI that could not have been made had that not happened, including (and I’m quoting directly from the complaint here):

  • “In July 2023, Liam Swayne used ChatGPT to generate versions of The Winds of Winter and A Dream of Spring, intended to be the final two books in the series A Song of Ice and Fire, which Martin is currently writing.”
  • “When prompted, ChatGPT accurately generated summaries of several of the Martin Infringed Works, including summaries for Martin’s novels A Game of Thrones, A Clash of Kings, and A Storm of Swords, the first three books in the series A Song of Ice and Fire.”
  • “When prompted, ChatGPT generated an infringing, unauthorized, and detailed outline for an alternate sequel to A Clash of Kings, one of the Martin Infringed Works, and titled the infringing and unauthorized derivative A Dance With Shadows, using the same characters from Martin’s existing books in the series A Song of Ice and Fire.”
  • “When prompted, ChatGPT generated an infringing, unauthorized, and detailed outline for a prequel book to A Game of Thrones, one of the Martin Infringed Works, and titled the infringing and unauthorized derivative A Dawn of Direwolves, using the same characters from Martin’s existing books in the series A Song of Ice and Fire.”

The complaint also mentions an experiment conducted by researchers at the University of California, Berkeley into which works ChatGPT has been memorizing. The researchers found that Martin’s novel A Game of Thrones “ranked 12th with respect to the degree of memorization.” So not only is ChatGPT using Martin’s work, it’s using it rather a lot.

Game of Thrones creator vs ChatGPT

It seems pretty obvious to me that ChatGPT has indeed absorbed the works of authors like Martin without their permission or remuneration. Next, Martin and his co-plaintiffs have to prove that OpenAI has violated their copyright and deserve redress.

The most likely defense that OpenAI can mount is fair use. Essentially, they could admit to using copyrighted work to power their programs, but argue that this is still allowed under the law. Coverage of this case has focused on the plaintiffs’ argument that fair use does not apply because ChatGPT could be used to produce new work that will affect the market value of the original work. In other words, they think ChatGPT could produce a version of, say, The Winds of Winter that people could pay to read instead of paying to read Martin’s version. In that case, a fair use defense wouldn’t work, since fair use only applies if it doesn’t displace the market for the original work.

I have some issues with this argument, since I’m not at all convinced that generative AI is at a point where it can write anything that could serve as a satisfying substitute for human writing. The complaint mentions those AI-generated versions of The Winds of Winter and A Dream of Spring. I’ve read some of those, and they are brutally repetitive, nonsensical and bland. I very much doubt anyone in their right mind would accept those as a worthy substitute for whatever Martin is working on.

This argument, at least, seems kind of theoretical to me. It might be possible for AI to generate works that will compete with the originals, but I have yet to see any examples. Works like A Dawn of Direwolves and A Dance With Shadows seem to be stuff the plaintiffs generated to prove up their case, not examples of infringing works that are out there and crowding out the market for the original books.

It suggests a comparison to fan-fiction. Fan-fic authors have been writing infringing works based on all kinds of books, movies and TV shows for decades, but authors mostly leave them alone provided fan-fic writers don’t try to make money off their infringing works. Although it’s never been definitively tested in court, it’s assumed that people aren’t reading the fan-fic in place of the original work. Thus, fan-fic is allowed to continue to exist under the law.

So is AI-generated fan-fic like normal fan-fic? Does it deserve the same arm’s-length “you can exist but don’t try anything funny” treatment? Or is AI-generated fiction something new that needs to be governed by new rules and boundaries?

Those are the questions the courts will be grappling with as this and other cases involving generative AI make their way through the system. We’ll be watching.

To stay up to date on everything fantasy, science fiction, and WiC, follow our all-encompassing Facebook page and sign up for our exclusive newsletter.

Get HBO, Starz, Showtime and MORE for FREE with a no-risk, 7-day free trial of Amazon Channels