Copyright for AI Generated Content

Previously, we explored the growing concerns around AI-generated content and whether it posed a threat to human authors. While the influx of AI-generated books, articles, and other creative works is undoubtedly changing the landscape, the more pressing issue may be how to navigate the complex and developing copyright implications of these new technologies.

As AI systems become increasingly adept at generating human-like content, the question of who owns the rights to these works has become a subject of intense debate and legal uncertainty. Different countries are considering a range of approaches, from denying copyright protection altogether to granting it to the human operators or the AI developers. Meanwhile, the ethical considerations around the use of AI-generated content are also coming into sharp focus.

By understanding the current state of the law and anticipating future developments, writers can position themselves to thrive in an increasingly AI-driven creative landscape.

Before we dig right in, we also want to remind you that we are based in the United States and are not practicing attorneys. As you'll shortly see, local rules can vary, and this is an area of law that will hopefully see some change in the not too distant future, so please just use this as a starting point to understand where AI and copyright intersect at this particular point in time.

The Current State of Copyright for AI

There are a few ways that countries are considering for how they will treat the copyright of AI-generated works.

1. No Copyright Protection: Copyright requires human authorship, and under this model, there is no copyright protection extended to anything generated by a machine. This is especially relevant when there is little to no direct human involvement in its creation.
2. Granting Copyright to the Human Operator/User of the AI System: Provided that there is a sufficient level of human intervention and creativity, this model would grant full copyright protection to the human directing and editing the AI-generated work.
3. Granting Copyright to the AI System's Trainer/Developer: An alternative approach is to vest the copyright of AI-generated works with the individuals or organizations responsible for training and developing the underlying AI technology, rather than the human user who operates the system to create new content. This model recognizes the AI system itself as the true “author” of the generated output, and grants the copyright to the entities who imbued the AI with the skills and knowledge to produce the creative works. Proponents of this view argue that the AI trainer/developer is the true “creator” deserving of copyright protection, as they are the ones who enabled the AI's generative capabilities.
4. Copyright for the Arrangement or Display of AI-Generated Content: This model allows for copyright protection for the end product that is generated using AI without extending full copyright protection over the underlying words, images, or other elements. The only protections would be for the specific arrangement, curation, or display of the AI-generated content. This more limited copyright protection would apply if somebody tried to copy your entire work, but would not protect from them plagiarizing bits or pieces of it.

Current State of Copyright

Overall, there is a lack of international consensus on how to treat copyright protections. As an author or publisher, you need to know that different areas of the world may treat your use of a work differently than other parts.

For example, at the time of initially writing this (May 2024), the United States officially has a stance that there are no copyright protections for AI-generated works because of the lack of human involvement.

However, the copyright office has assigned copyrights to individual works on a case-by-case basis, with most of the specific examples I've seen falling into one of two categories:

1. The filing acknowledges that some of the content has been AI-generated and edited into the human-generated content, granting full protection to the latter while providing no protection to the AI-generated content. However, there is no indication specifically in the filing about what is human and what is AI generated, so until a case lands in front of a court or in front of the Copyright Claims Board (for small claims involving disputes up to $30,000) there is no way to know how these situations may actually be handled if infringement should happen.
2. Copyright has been granted to the arrangement and curation of a work, offering protection against someone reselling your work as a whole. How well that might protect somebody from only copying a portion of your work is also untested.

In other parts of the world, the European Union has proposed rules that would grant copyright protection if there is a sufficient level of human intervention and creativity, but where that line is drawn has not been settled yet.

The United Kingdom's Intellectual Property Office has recommended that copyright be granted to the human operator or user of the AI system, although the UK's laws have not yet been updated to reflect that recommendation.

China and India have been exploring the copyright implications of AI, including creating sui generis rights for AI-generated content, but I have not yet found any clear or unified policies that have been gaining steam.

Sui Generis Rights for AI-Generated Content

Sui generis rights refer to a specialized form of intellectual property protection that is distinct from traditional copyright, patent, or trademark laws. (The term “sui generis” comes from Latin and means “of its own kind” or “unique.”)

In AI-generated content, the concept of sui generis rights involves creating a new, tailor-made legal framework to address the unique challenges and considerations around the copyrightability of works produced by artificial intelligence systems. This is in contrast to trying to fit AI-generated content into existing copyright laws, which may not adequately account for the non-human authorship.

Sui generis rights would be separate from traditional intellectual property rights and would establish a distinct form of protection rather than trying to apply copyright, patent, or other existing frameworks.

These rights could cover different areas of AI-generated content, such as the underlying training data, the algorithms and models used to generate the new content, as well as the final output itself.

Ownership and allocation of the rights would need to be determined between the AI developer, the end-user, the copyright holders of the underlying training data, or some combination of the different parties.

The advantage of these new rights would be by allowing different limitations and exceptions that are specific to AI-generated content without being beholden to the rules governing copyright, and would need to balance the need for innovation and public access.

Establishing sui generis rights would be an uphill battle despite the benefits of being tailor made for the new technologies, as it would require a significant amount of international cooperation and such a system may not be feasible in a reasonable frame of time.

The Nuances of Copyright Protection

The primary focus of most of the ongoing discussions and policy proposals is on adapting existing copyright frameworks rather than creating new ones. However, even within the realm of traditional copyright law, there are complex nuances and considerations that still need to be addressed.

The key distinction is between works that are generated solely by an AI system with no human intervention or with minimal intervention such simple prompting, versus content where a human has played a significant role in editing, curating, or shaping the AI-generated output.

The level of human creativity and “authorship” involved can impact whether these works qualify for copyright protection. If the human contribution is judged as being sufficiently transformative, it may qualify for its own copyright, apart from the underlying AI-generated elements.

Similarly, when humans create new works that are derived from or based on AI-generated content, questions arise around the copyrightability of these derivative works. There will need to be a definition for the boundaries between the rights of the AI “creator” and the human “derivative work” author.

Ultimately, policymakers will need to carefully balance providing appropriate copyright incentives for human creativity involving AI, while also ensuring reasonable public access and use of AI-generated content. Overly broad copyright protection could stifle innovation, while too narrow of an approach may undermine the rights of human authors and artists.

Ethical Considerations of AI-Generated Content

The growing prevalence of AI-generated content has raised several moral and ethical questions around the appropriate application of copyright law. At the heart of the debate is the fundamental quandary of whether AI systems can be true “authors” deserving of copyright protection, or if such works should remain in the public domain.

One of the primary ethical concerns centers on the source and legitimacy of the training data used to create these AI models. There are fears that unscrupulous actors could leverage AI to mass-produce content that infringes on the rights of human creators by training their systems on copyrighted works without permission. This raises the specter of “digital plagiarism” – the unauthorized use of protected material to generate new content.

Even where the training data is ostensibly public or unprotected, there are debates around whether the act of scraping and aggregating this information constitutes plagiarism.

Proponents argue we should recognize and compensate the human creativity and effort involved in curating and organizing data, instead of simply allowing AI systems to exploit these resources freely.

Others view accessing public information and using it as training for their models is “fair use” and often point to the Authors Guild vs Google lawsuit from 2015 as setting precedent, where Google's scanning and cataloging books for their search engine was deemed to be a fair use by transforming it into a new form of content.

In my opinion, accessing public data or using data that you have legal access to should be considered a fair use for these training models, but using pirated content illegally should not be.

Another ethical contention is the questionable creativity and “authorship” inherent in AI systems, and granting copyright would be a misguided attempt to artificially bestow human-like rights upon non-human technology.

The counterargument is that AI-generated content should remain freely available to the public to foster innovation and ensuring broad access to these novel forms of creative expression.

Protecting Author Rights

As the prevalence of AI-generated content continues to grow, it is crucial for authors to understand the steps they can take to safeguard their intellectual property and ensure fair compensation for their creative works. While the legal landscape remains in flux, there are proactive measures writers can employ to protect their rights.

First, authors should maintain a log of their work. Authors should start by maintaining a journal that tracks their progress towards a specific work and summarizes what they have worked on. Even better, would be backups of any chat logs with the various generative AI models the author uses to assist their work. Best would be regular backups of the work in progress that show individual changes.

Depending on the software that you use to write your books, you could copy the individual files regularly, such as at the end of a workday or end of a week. Some writing software, such as Google Docs, automatically keeps a complete and real-time revision history for you. I tend to write in Scrivener, which has a snapshot option that lets me define specific revision points.

Should you ever need to defend your copyright, having evidence of what you produced, when you produced it, and how it was produced can go a long way towards proving not only that you are the original author but also help differentiate between what is human generated versus what is AI generated.

Second, for key publications, it may be worth registering copyrights. This can bolster the legal standing over your work and make it easier to enforce your rights.

There are two schools of thought for monitoring for unauthorized use or reproduction of one's work. The first school of thought is that you should enforce your copyright to the best of your ability and make sure that anybody infringing on your work takes it down immediately. The second is that as authors we suffer far more from obscurity than we do from piracy, and that piracy is more likely to lead to new and future paying readers than it is to lead to actual lost sales as pirated books were unlikely to have been from somebody who would have purchased the book, anyway.

I fall in the middle but more towards the latter for my work. I keep an eye out for anybody infringing upon my work, but I don't go to a lot of effort to stop the pirates. If somebody is reselling my book or claiming it as theirs, then that is when I would be more likely to get the book removed from where it's online or would consider starting a lawsuit, though I've yet to encounter a situation where that would have been worth my while up to this point.

As for using AI-generated content as part of a work, we will have to wait and see what laws and legal interpretations are over the next 5 to 10 years before we will be able to say for certainty what can and can not be done. With that said, there are some assumptions we could consider along the risk versus reward scale to decide what we are comfortable using. Some of the biggest issues that are worth addressing are:

• Using copyrighted output: One concern is that the output created by an AI model will consist of copyrighted work. As long as you are not intentionally trying to generate exact replicas of a work based on a model's training data, it's possible, but unlikely, that it will output something exactly as it was trained on. You can run the output through a plagiarism checker to find any instances of infringement.
• Using models trained on unlicensed copyrighted content: A larger concern is that the underlying model is illegally using copyrighted content. As I mentioned earlier, I expect AI models will eventually be legally recognized as transformative fair use, minimizing the legal exposure for end users of these models. More likely, the companies that are creating the training models may have to retrain their models using content that is properly licensed or may have to pay licensing fees and fines, but will then continue on with business as usual.
• Using “public domain” AI-generated content: Assuming that content that you want to use is considered to be in the public domain because it's not eligible for copyright, as appears likely to be the case for the United States, then you can include that content in any form in your own work. Others could as well, but that does not prevent you from using it and selling it.

Conclusion

As the prevalence of AI-generated content continues to grow, the need for clear, consistent, and author-friendly copyright policies has never been more urgent. The legal landscape remains in flux, with countries taking divergent approaches and ongoing debates about the treatment of works produced by artificial intelligence.

Ultimately, striking the right balance will be crucial – one that provides meaningful protections for human creativity while also fostering innovation and public access. Policymakers, technology companies, and the creative community will all need to collaborate to navigate these complex issues.

For authors, the key is to proactively monitor their local laws and regulations, and even to advocate for fair policies, while exploring how they can integrate some of these tools into their own processes. By embracing the potential of these technologies as collaborative tools, rather than viewing them as threats, authors can unlock new avenues for artistic expression and commercial success.

The rise of AI may irrevocably change the future of authorship, but the human touch, the personal connection, and the unique creative vision of individual writers will always hold intrinsic value. AI will not replace authors, but they may lose market share to other authors who embrace AI in their own systems.

Next time, we will explore some of the innovative ways that authors are using AI. Even for authors who aren't interested in generating content directly, these tools can still be used to enhance creativity and aid in marketing our books.

In the meantime, if you are interested in more information about copyright and how it affects you as an author, with and without the use of AI, you may be interested in checking out Training #475 which covers Copyright for Authors.