ABSTRACT
Generative AI has taken the world by storm, promising to revolutionize any industry it enters. Creating a generative AI model is a complex and costly process, necessitating the consumption of a gargantuan amount of human-created expressive content. Yet, the creators of these expressive works have not consented to this usage, nor been compensated for it. Balancing the interest of human creators against the potential advancements offered by generative AI has become a prominent topic of debate among legal scholars and policymakers. Presently, this issue is predominantly explored through the lens of copyright law.
Unfortunately, the application of copyright law to the dilemma of generative AI training proves futile, leading to one of two extreme solutions. First, courts may find that generative AI training infringes on the copyright of human creators. If this is the case, human creators will be protected through injunctions and statutory damages, but this protection will spell the end of generative AI training. Second, courts can hold that generative AI training does not constitute copyright infringement but should instead be considered fair use. Under this alternative, human creators receive nothing. Ironically, this outcome is also detrimental to the further development of generative AI. If human creators are not compensated for their labor, and have insufficient incentive to create, generative AI will have less human-generated data to train on. Over time, if generative AI is increasingly trained on AI-generated content, this can lead to model collapse. Important to note, the fact that copyright law offers poor solutions to the generative AI dilemma is not surprising, as generative AI was nothing more than science fiction when copyright law was developed.
Against this backdrop, this Article is the first to offer an alternative legal response to the dilemma of generative AI training, through the law of unjust enrichment. Unjust enrichment law can offer a middle ground solution between the two extreme responses offered by copyright law. Unjust enrichment law can provide better-tailored remedies that establish a layer of protection for human creators without paralyzing the market for generative AI. This can be done by providing human creators liability rule protection instead of property rule protection, meaning that human creators are compensated moderately, without being given the power to stop the training and development of generative AI. More broadly, unjust enrichment, as a residual and flexible legal framework which is not context-specific, is uniquely suited to be applied to new technologies and emerging problems. Unjust enrichment law also enjoys an advantage over copyright in terms of its interjurisdictional applicability. Thus, any limitation set on AI development under United States copyright law will mainly disadvantage American firms. Conversely, if American courts adopt liability under unjust enrichment for AI training, this regime can be enforced also against foreign firms. The Article studies additional advantages, implications, and limitations of the proposal to apply unjust enrichment doctrine to generative AI training, including preemption, the challenges created by lack of provenance, and the diversity of plaintiffs under this cause of action.
Kaplan, Yotam and Gordon-Tapiero, Ayelet, Generative AI Training as Unjust Enrichment (March 27, 2024), Ohio State Law Journal, volume 86, 2025.
Leave a Reply