Introduction
Artificial Intelligence (AI) has significantly impacted daily life and work, particularly in the realm of intellectual property law. Court cases in the U.S. and China are questioning whether Generative AI (GenAI) tools violate copyright laws. These AI systems use large datasets to generate content like text, images, and videos, raising concerns about potential infringement if the data includes copyrighted or protected material. This article explores the challenges of copyright infringement related to AI models like Stable Diffusion and Midjourney, which use copyrighted content for training and content generation.
Copyright Infringement in the Age of AI
The rapid advancement of AI has introduced new intellectual property considerations, particularly around copyright infringement, influencing both the types of claims raised and the court’s reasoning. In jurisdictions with a rising number of GenAI copyright cases, creators have sued AI models over the replication of their copyrighted content or its use in training these models. Claimants usually assert multiple forms of infringement, including direct infringement during the training phase; claims that AI-generated output is a derivative work; and vicarious infringement.
Direct Copyright Infringement
In a direct infringement claim, the claimant argues their copyrighted work was used without permission to train AI models. This is governed by Section 36 (a) of the Copyright Act 2022. In 2023, artists Sarah Anderson, Kelly McKernan, and Karla Ortiz sued Stability AI, DeviantArt, Inc., and Midjourney, Inc., claiming their copyrighted images were used to train AI systems, resulting in infringing outputs
To succeed, claimants must prove valid copyright ownership and so, in determining which of Sarah Andersen’s works could be rightly claimed for in the case cited above, the court limited the works to only those that had been registered with the Copyright Office. Claimants must also demonstrate how their work was copied or used in training. Thus, the court ruled that Stability’s use of the plaintiffs’ copyrighted works to train its AI model, Stable Diffusion, amounted to direct infringement. However, claims against DeviantArt and Midjourney were dismissed, as the plaintiffs failed to show how the AI-generated images resembled the original works or what training, if any, occurred. The court allowed the plaintiffs to amend their claims to clarify facts showing how the developers had improperly used the plaintiffs’ copyrighted images.
Indire]ct Infringement
Generative AI is a self-evolving tool, which continues to learn and expand its training data set through user prompts, beyond the control of the developer who merely trains it to understand “how to learn”, a volitional requirement is unlikely to be met especially when the developer has taken all objectively plausible, practically feasible, and technologically available steps to ensure that the model does not bootleg output that is substantially similar to any of its input training data.
When direct infringement claims fail due to lack of causation/volition, the focus often shifts to the indirect infringement standard. Section 36 (e) of the Copyright Act provides that permitting for profit, any place for communications that infringe copyright would constitute indirect infringement, unless the operator was unaware or had no reasonable ground for believing such communication to be infringing.
Copyright Infringement through Derivative Works
Derivative works are covered by section 9 of the Copyright Act 2022 which are exclusive to the owner of such rights subject to the exceptions under Part II of the Act. Under the Nigerian Copyright Act, copyright holders have exclusive rights to reproduce, adapt, and distribute their works, and any unauthorized reproduction or adaptation constitutes infringement.
Creators also allege copyright infringement when AI-generated outputs produce what the creators deem to be derivative works of their copyrighted content. In their nature, GenAI models require large datasets for training, and when prompted, they may reproduce parts or entire works of copyrighted material. In cases involving derivative works, courts typically require claimants to prove that the AI outputs are direct copies or substantially similar to their copyrighted works. Failure to establish this often results in dismissal of the claims.
In 2023, authors Richard Kadrey, Sarah Silverman, and Christopher Golden filed a class action against Meta, accusing the company of using their copyrighted books to train its AI model, LLaMA, and claiming copyright infringement in the model’s outputs. The plaintiffs asserted direct and vicarious infringement, unfair competition, negligence, and unjust enrichment. However, the court dismissed most claims, including those related to the outputs, ruling that the plaintiffs failed to prove that the outputs were actual copies or substantially similar to their works to qualify as derivative works.
In early 2024, the Guangzhou Internet Court in China ruled in favour of a character licensing agency, affirming the rights of creators in a copyright infringement case involving AI. The claimant, holding the exclusive license for the Ultraman series in China, argued that an AI service provider’s model generated images identical to Ultraman works based on prompts related to “Ultraman.” The court upheld the claimant’s exclusive rights, ruling that the AI-generated images had some adaptation of the original work which constituted unauthorized derivative work.
If the output generated by the Generative AI model is materially similar or a mild alteration, of a work, it may infringe the rights of the copyright owner. However, who will be liable? The user or the Model developer? both? or neither? These questions are essential to consider given the Model itself lacks legal personhood for imputing any intent, liability, or damages – and responsibility for infringement has to be attributed to a human/corporation. The User’s interaction with the model by means of prompts outside the control of the developer is also a crucial factor to be considered when imputing liability.
The Defence of Fair Use
In defending against copyright infringement claims, AI developers often cite fair use, which allows the unlicensed use of copyrighted works under certain conditions to promote freedom of expression. In Nigeria, the term is cited as “fair dealing,” and is permitted for purposes like private use, parody, satire, or caricature. The Nigerian Copyright Act outlines conditions for fair dealing, including: (i) purpose and character of the use, (ii) nature of the work, (iii) amount and substantiality used, and (iv) effect on the market or value of the work. These conditions are also part of U.S. Copyright Law, where an essential element is the transformative use of the work, meaning it is used in a new way or for a different purpose, and thus does not infringe its holder’s copyright.
In Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, which involved Warhol’s Prince Series based on a photograph of the musician Prince by Lynn Goldsmith, Warhol argued that his alterations to the photograph through his paintings were transformative. However, the court held that minor alterations to a copyrighted work are not transformative under fair use where altered work was used commercially for substantially similar purpose as original.
Also, in June 2024, the Recording Industry Association of America (RIAA), on behalf of Sony Music Entertainment, Universal Music Group’s UMG Recordings, and Warner Records Inc., sued Suno and Udio, two AI models that produce AI-generated music. The plaintiffs alleged mass infringement of copyrighted sound recordings copied and exploited without permission to train the generative AI models. Admitting to using copyrighted songs to train their AI models, the defendants claimed that it was nothing more than a “fair” use of tools already available on the market. As the case progresses, it remains to be seen how the courts will interpret the law in reaching a decision.
Read also: Nigeria taps artificial intelligence to accelerate growth
Class Action Lawsuits in GenAI Copyright Infringement Cases
Another important aspect to consider is the use of class action lawsuits by claimants in copyright infringement cases. Given that GenAI can replicate works from numerous creators, class actions offer an efficient way for these creators to collectively seek relief. However, questions arise as to the potential for success if Nigeria were to face similar cases, particularly regarding the challenges of filing class actions and the dearth of cases in this regard. Expanding our jurisprudence on class actions, especially in technology-related disputes, would be beneficial.
Obtaining Licences
A potential solution to address copyright owners’ concerns could be requiring GenAI developers to obtain licenses from copyright holders before training their models. While this may seem like a straightforward remedy for copyright infringement claims, the amount of data used to train generative AI makes this approach challenging. For instance, recent models like GPT are trained on trillions of words sourced from diverse materials. Obtaining licenses for such vast amounts of content, along with the associated costs, could hinder innovation and discourage the development and use of AI models, potentially stalling progress in the field.
The European Union Act on AI takes a proactive stance on copyright ownership, recognizing both the innovation potential of generative AI and the challenges it poses to creators. The Act requires authorization from copyright holders for any use of protected content, unless specific exceptions apply, such as for text and data mining. It also mandates AI model providers to ensure compliance with copyright laws, including transparency about the data used in training models, and to publicly disclose detailed summaries of the training content.
Conclusion
The intersection of AI and intellectual property law presents complex challenges, particularly in the context of copyright infringement. Although Nigeria has not yet ruled on such cases, this presents a timely opportunity for legislators and policymakers to take a proactive approach in updating the legal framework. It is essential to address emerging issues, such as the criteria for determining copyright infringement by AI and establishing policies to guide generative AI developers on sourcing data. At the same time, it is crucial to create provisions that foster investment in AI development while striking a balance between encouraging innovation and safeguarding the rights of creators.
CONTRIBUTORS
Tilewa Oyefeso – Partner at Wigwe & Partners ([email protected])
Emaediong Lawrence – Associate at Wigwe & Partners ([email protected])
DISCLAIMER
This article is for informational purposes only and does not constitute legal advice or establish a lawyer-client relationship. For specific legal advice, please consult a qualified legal professional.
Join BusinessDay whatsapp Channel, to stay up to date
Open In Whatsapp
