"Digital Dialectic: Why Every “AI-Generated” Work Has a Human Author" by Lea Shaver

Lea Shaver, Indiana University - Robert H. McKinney School of LawFollow

Abstract

Are ChatGPT and Midjourney tools or creators? Ownership of billions of AI-assisted creative outputs hangs in the balance. Copyright scholars have long debated whether an autonomous artificial intelligence could qualify as an author, but this remains a hypothetical question. Despite widespread application of the term “AI” to software products of the 2020s, autonomously creative artificial intelligence still does not exist. Today’s commercial AI products—such as ChatGPT, Midjourney, Dall-E, Copilot, Gemini, Claude, Suno, Perplexity, and Lumo—are simply the newest generation of computer software. They do not qualify as “artificial intelligence” in either the scientific or science fiction senses. They are mere computer programs, tools that act only upon instructions of human creators. Under basic principles of copyright law, long applied to computer software, the human user of the software program is entitled to protection of the output. Every “AI-generated” work has a human author. To prove that point, this Article’s centerpiece is a Socratic dialogue, presented as a conversation between myself and a large language model (LLM). Following long tradition, this dialogue is structured to lead readers to a logical conclusion; namely, that every “AI-generated” work has a human author. By design, the narrative presents as an encounter between two minds. In reality, there is only one intelligence at work—mine. This dialogue serves several purposes. Of interest to scholars in any discipline, it demonstrates how we can use digital dialectic to explore hard problems, identify related literature, develop theories, and achieve new insights. Thematically, it explains how LLM software works, contrasts this with common misconceptions about “artificial intelligence,” and reflects on the intellectual work we do when we use LLM software and what this means for the authorship question in copyright law. It also models how scholars—and lawyers—can leverage LLM AI tools to conduct more powerful research, thinking, and analysis. Finally, it offers copyright scholars a welldocumented specific instance of AI-assisted writing, so that we can move from hypothetical debate to empirical analysis of generative authorship. The dialogue thus serves as argument, methodology, and demonstration— piercing the illusion of artificial intelligence to offer a more realistic understanding of today’s generative software as tools, not creators. The Article also surveys over four decades of copyright scholarship about authorship and artificial intelligence, distinguishing between hypothetical and empirical AI scholarship. Copyright issues surrounding AI have long been explored hypothetically. In my view, there is now an urgent need for empirical AI scholarship. The new school of generative empiricists ground their legal analysis in real-world study of how artists, musicians, and writers use the new tools to explore, play, study, and create. The longhypothetical copyright debate presumed an autonomously-acting artificial intelligence. Empirical AI scholarship emphasizes that today’s generative technology does not fit this description. Midjourney and ChatGPT are software tools, subject to the same rules of copyright that govern Photoshop and Microsoft Word. Under the Supreme Court’s Feist standard of originality, software users are entitled to copyright ownership of the output whenever they contribute at least a minimal spark of creativity.