We’ve always been discussing AI in this class, and I recently had an experience that left me both fascinated and concerned and made me want to share my thoughts. I was in a WeChat group chat with some Chinese entrepreneurs who were discussing their AI products. One product, in particular, caught my attention: it compiles multiple AI models together to create a coherent system that users can leverage to easily generate content, including entire novels from scratch. The mechanism asks users to input any novel they like into the chat box, and the AI will extract the plot progression. It can then combine this with characters and descriptive styles from another book while allowing users to specify their desired writing style and genre/theme.
What’s amazing about this multi-AI model is its ability to consume over 100,000 words per turn, allowing it to generate a complete novel at once. This model typically sources material from online novels that are readily accessible. Interestingly, some online novel websites in China even notify creators that by posting their work on the platform, they consent to its use by AI models, for example, Fanqie has launched its AI policy that explicitly allows the platform to do that.[1]
There is an “AI Training Supplementary Agreement” on that platform that refers to Article 3.2.10 newly added by Tomato Novel in their contracted work cooperation agreements in late 2023. It states: “Party A (Tomato Novel) may use all/part of the contracted work’s content and related information (such as work title, synopsis, outline, chapters, characters, author’s personal information, cover images, etc.) as data, corpus, text, materials, etc. for annotation, synthetic data/database construction, AI artificial intelligence development, machine learning, model training, deep synthesis, algorithm development, and other current or future technological development/application fields. [2]Some writers are really worried about it.[3] However, it’s unfortunate that many writers don’t seem too worried about their work being utilized by AI, perhaps assuming it can’t be easily leveraged. But they are sorely mistaken. This raises some thorny questions around the human effort involved in the creative process and how copyright should be attributed.
screenshot of the AI model, translated by Google
The part that worries me is that the AI-generated work appears to be of high quality, drawing in readers and incentivizing platforms to promote AI usage. There is even a Japanese writer who made their novel partly by AI and got a prize in a competition.[4] This is drastically altering how we create and consume content. While it may not be entirely distinct from other AI-related challenges, I want to pay close attention to the coherent approach employed by this multi-AI model to see if it introduces any novel issues.
According to Canadian copyright law, for a work to be considered original, it must involve an exercise of skill and judgment beyond mere mechanical effort. The work must also be more than a simple copy, though it doesn’t have to be novel or unique.[5]
This AI model guides users through a series of prompts, like selecting the genre, writing theory, and style to emulate. The key question is whether this process entails sufficient human effort and originality. Compared to graphic design, where you simply point and click to generate, composing an engaging AI novel seems to require mastery of the prompting process.
I’m inclined to think that the skill and judgment involved in prompt engineering likely don’t meet the originality standard on its own. However, the issue of writing prompts raises some complex questions, as we touched on in class. Beyond that, it could introduce potential co-authorship issues.
Writing the prompt itself is a copyrightable activity. However, the final work is produced by the AI using the sourced materials and prompts provided by various people in the group chat. If such works were ever deemed copyrightable, the prompt writer could theoretically be considered a co-author of thousands of novels. Just imagine the implications if those novels started generating profits – which is already happening based on the data shared in the group chat.
About the prompt, some US standards say whether AI-generated content can obtain copyright registration depends on the specific circumstances of creation, especially how the AI tool was operated and how it was used to create the final work. If the traditional creative elements of a work are produced by machines, then that work lacks human creativity, and the copyright office will not register it. For example, when a technical tool only generates derivative works based on human prompts, the creative elements of that work are machine-made, not human; If human artists select or arrange AI-generated materials in a sufficiently creative way, or if artists modify AI-generated materials, and these modifications meet copyright standards, then the copyright office can protect the human-created portions of these works. [6]
This situation is distinct from just creating a short story or novella because generating an entire novel with that volume of text fundamentally changes the nature of the creation. Profiting from this approach could seriously harm the original copyright owners.
Things get even messier when you consider that this AI novel writing industry doesn’t solely rely on AI generation. Among the AI-written novels, some are purely AI-generated with the human simply copying and pasting, while others require extensive editing before they’re publishable. It’s really difficult to draw clear lines in this space.
Trying to apply the concept of “effort” to determine originality becomes problematic in this context. There has been some recent legislation attempting to address this. In August, the European Artificial Intelligence Act, the world’s first comprehensive AI regulation, officially took effect. Regarding licensing and exceptions, the “AI Act” mandates that providers of general-purpose AI models obtain authorization for any use of copyright-protected material during model development and training unless such use falls under the EU copyright regime’s limitations on rights.[7]
I think this will be extremely difficult to enforce in practice because for our AI model, the sources are just words – and words are ubiquitous. You simply can’t control that. So how do we effectively tackle this issue? Furthermore, is creativity truly the distinguishing factor?
And how is this different from AI-assisted homework corrections? Perhaps it’s similar to the CCH case, where grammatical errors being fixed in judicial reasons were considered too trivial to warrant copyright protection, as these additions are more properly characterized as mere mechanical exercises.[8]
I’m starting to believe that we need better frameworks for how AI generates work. For instance, maybe we should require specific activities, like actually “writing” original material, for a piece of fiction to be copyrightable, rather than just mastering a prompt. While prompt engineering undoubtedly requires human effort, it’s not the same as writing. The mere presence of human effort may no longer be sufficient.
Perhaps we could create new legal categories specifically for AI-generated art. I know some online platforms are already requiring authors to tag their work as “AI-generated” if applicable, though it’s not mandatory. Maybe the law needs to recognize this distinction as well.[9]
Ultimately, I don’t have any easy answers, but I think we must grapple with these questions around AI-generated content and copyright. Technology is evolving so rapidly, and our legal frameworks are struggling to keep pace. We need to think critically about what constitutes originality and creativity in the age of AI, and how we can fairly protect the rights of human creators while still allowing for technological innovation.
Reference
[1] https://fanqienovel.com/writer/zone/article/7327136545129906238
[2] https://www.sohu.com/a/797863051_121429370
[3] https://medium.com/@giant_chen1688/the-ai-revolution-in-web-novels-how-authors-can-protect-their-rights-and-thrive-a1d1ac80541a
[4] https://aibusiness.com/nlp/author-reveals-using-chatgpt-for-award-winning-novel#close-modal
[5]CCH Canadian Ltd v Law Society of Upper Canada, 2004
[6] Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, US Copyright Office, Mar 16, 2023, https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence
[7] https://artificialintelligenceact.eu
[8] CCH Canadian Ltd v Law Society of Upper Canada, 2004
[9] https://www.pixiv.net/info.php?id=8834&lang=en
Copyright & Social Media
Communications Law