Building and Using Generative Models Under US Copyright Law

Abstract

As advancements in artificial intelligence continue to expand our capabilities, generative machine learning (ML) tools have sparked considerable interest and debate. These tools, powered by a core "model" trained on massive amounts of data, are capable of creating new text, code, images, and music that rival human capabilities. However, the use of copyrighted materials for training ML models raises fundamental questions regarding copyright infringement and fair use.

This article provides a comprehensive analysis of the copyright issues associated with machine learning in two parts. First, it provides a clear and correct description of machine learning technology accessible to non-scientists. It explains machine learning models, how they are trained, and how they are used to generate new works.

Second, the article analyzes applicable copyright law in relation to the factual foundation developed in part one. Comparing ML to technologies scrutinized in previous cases, the article finds robust support in case law to argue that the development and use of generative ML models in most cases falls outside the scope of copyright or constitutes fair use.

The article serves as a definitive guide to machine learning practices for legal practitioners, scholars, and policymakers seeking to navigate the complex intersection of machine learning and law. By promoting a common factual and analytical framework, the article aspires to provide a common foundation for future legal discourse, even among those who may disagree with its conclusions.

Keywords: AI, artificial intelligence, ML, machine learning, model, copyright, fair use