AI Patents: A Data Driven Approach

Abstract

The global technology market exceeds $12 trillion. The market’s fastest growing niche is artificial intelligence (AI). Yet, while the literature on technology patents is theoretically robust - the literature on AI patents is relatively uncharted. As a consequence, lawyers, scholars, and commentators often refer to AI as a black box – arguing not even advanced computer scientists understand how it works. But all AI technology is written with formal logic, mathematics, and computer code. Thus, all AI systems are syntactically describable, repeatable, and explainable. In other words, there is no black box. This Article empirically analyzes the unique intellectual property strategy decisions technology firms face by introducing a dataset including four specific types of machine learning patents: deep learning, reinforcement learning, deep reinforcement learning, and natural language processing. Dataset charts, models, and graphs, provide insight into market alcoves, while analysis of each machine learning technology shines a light through the “black box.” Further, patent claims analysis reveals significant overlap in patented AI technologies. In sum, this Article draws on a growing body of informatics, intellectual property, and technology scholarship to provide novel patent analysis and critique.