Over the past year, artificial intelligence enjoyed its mainstream breakthrough.
The instant success of ChatGPT and follow-up releases of other large language model-based tools kickstarted what many believe is a new revolution.
By now it is clear that AI offers endless possibilities. At the same time, however, it has ignited many new worries. Copyright holders, in particular, are concerned that their work is being used as training models but without permission.
‘Piracy-Trained’ AI Models
Over the past few months, we have seen a variety of copyright lawsuits, many of which were filed by writers. These cases target ChatGPT’s OpenAI but other platforms are targeted as well. A key allegation in these complaints is that the AI was trained using pirated books.
For example, several authors have just filed an amended complaint against Meta, alleging that the company continued to train its AI on pirated books despite concerns from its own legal team.
This clash between AI and copyright piqued the interest of the U.S. Copyright Office which launched an inquiry asking the public for input. With more than 10,000 responses, it is clear that the topic is close to the hearts of many people.
The tone of the responses isn’t hard to guess either. Copyright holders defend their rights and interests, often suggesting keeping a tight leash on AI training, while AI developers and tech companies prefer as few barriers to innovation as possible.
Science Fiction and Fantasy Writers
It’s impossible to summarize all opinions without AI assistance, but one submission stood out to us in particular; it encourages the free sharing of books while recommending that AI tools shouldn’t be allowed to exploit this generosity for free.
The submission was filed by the Science Fiction and Fantasy Writers Association (SFWA), which represents over 2,500 published writers. The association is particularly concerned with the suggestion that its members’ works can be used for AI training under a fair use exception.
SFWA sides with many other rightsholders, concluding that pirated books shouldn’t be used for AI training, adding that the same applies to books that are freely shared by many Science Fiction and Fantasy writers.
“SFWA acknowledges the problem of generative AI scraping pirated material published as copy-protected ebooks by professional publishers, but SFWA additionally has the unique position of representing many authors who have fought to make their work available for free for human readers.
“[M]any science fiction and fantasy authors of short fiction have embraced the open Internet, believing that it is good for society and for a flourishing culture that art be available to their fellow human beings regardless of ability to pay,” SFWA adds.
Many of the authors strongly believe that freely sharing stories is a good thing that enriches mankind, but that doesn’t automatically mean that AI has the same privilege if the output is destined for commercial activities.
The SFWA stresses that it doesn’t take offense when AI tools use the works of its members for non-commercial purposes, such as research and scholarship. However, turning the data into a commercial tool goes too far.
“The current content-scraping regime preys on that good-faith sharing of art as a connection between human minds and the hard work of building a common culture. The decision to publish creative work online to read and share for free […] is a trade-off of many factors including piracy, audience, and the simple (albeit elusive) ability to make a living.
“Bluntly, many authors do not want their work taken for this purpose, and that cannot be ignored,” SFWA insists.
Frodo and Gandalf faced the Balrog in Moria
AI freeloading will lead to unfair competition and cause harm to licensing markets, the writers warn. The developers of the AI tools have attempted to tone down these concerns but the SFWA is not convinced.
On one hand, AI-generated Science Fiction and Fantasy publications may vastly increase the availability of content. It could lead to a “flood of trash” publications that make it harder for human writers to get noticed and reach an audience.
AI-powered works could also compete within the licensed market, by using popular concepts, that are clearly inspired by existing works.
Meta previously said that AI models learn by taking basic concepts from a variety of sources. However, the writers counter that an AI’s knowledge about Frodo and Gandalf is much more specific.
Meta’s discussion in their lengthy section entitled, ‘How Large Language Models “Learn”’ uses the sentence, “Susan’s aunt planted the flower in the garden” to illustrate how these tools would draw from a wide variety of sources to learn individual words. This enables their tools to fluently use words like ‘flower’ and ‘Susan’; however, their explanation holds less true in how these tools would learn the words in a sentence like, “Frodo and Gandalf faced the Balrog in Moria.” Learning the words in that sentence both requires a much more targeted input data set, and—with due respect to Susan’s aunt—seems likely to be more commercially interesting.
Finding a Balance
The writers want to protect their rights but they don’t believe in the extremely restrictive position of some other copyright holders. They don’t subscribe to the idea that people will no longer buy books because they can get the same information from an AI tool, for example. However, authors deserve some form of compensation.
SFWA argues that all stakeholders should ultimately get together to come up with a plan that works for everyone. This means fair compensation and protection for authors, without making it financially unviable for AI to flourish.
“Questions of ‘how’ and ‘when’ and ‘how much money’ all come later; first and foremost the author must have the right to say how their work is used,” their submission reads.
“So long as authors retain the right to say ‘no’ we believe that equitable solutions to the thorny problems of licensing, scale, and market harm can be found. But that right remains the cornerstone, and we insist upon it,” SFWA concludes.
—
A copy of the Science Fiction and Fantasy Writers Association’s submission to the U.S. Copyright Office’s inquiry is available here (pdf).