The Role of Patent (In)Eligibility in Promoting Artificial Intelligence Innovation

Abstract

The artificial intelligence (“AI”) revolution sweeping across virtually every industry is set to transform and impact our world in ways not seen since the invention of electricity. Not surprisingly, Congress, the United States Patent and Trademark Office, academics, judges, practioners, and businesses are suddenly looking to understand whether our laws are ready for the challenges that are sure to follow.

When it comes to AI, patent eligibility—and its judicially created exceptions—is an area of the law that needs further attention. As with other areas of technology, companies and inventors are heavily relying on the patent system to offer protection to promote innovation in AI. With exponential growth in patent applications for AI inventions in the last decade, it is clear that patents will play a key role for innovation in the field of AI. But the current unresolved debate surrounding § 101 and its exceptions for abstract ideas, laws of nature, and natural phenomena are at the center of the question whether the current law on patent eligibility serves to promote or stifle innovation.

Before we can attempt to answer this question, however, we must first consider this question: What is an AI invention? The short answer, actually, is that we don’t know—and neither does the USPTO. There is no agreed upon definition for “AI.” As such, it should come as no surprise that there is no single definition for an “AI invention.” For example, in its attempt to define “AI invention” in a recent report, the USPTO defined AI as “comprising one or more of eight component technologies.” In this respect, I believe it is critical that we first recognize the unique considerations for determining patent eligibility of “AI inventions.” Rather than proceed with an unbounded, unclear definition of “AI,” or try to dissect AI into dozens of discrete types of inventions, I propose a taxonomy for addressing eligibility of AI inventions based on the three fundamental levels of AI systems: data, software, and hardware systems. All of them fall within the umbrella of “AI inventions,” and each presents distinct issues and levels of eligibility for patent protection. Thus, the first goal of this article is to explore differences in how we define an “AI invention” and offer a more structured and clear approach for determining how patent eligibility impacts AI innovation.

The second part of this project explores the criticisms to the current state of the law for § 101 and argues that any changes to § 101 and its related exceptions must be done with a scalpel rather than a machete. In particular, the article analyzes four key metrics of innovation for AI (patent and patent applications, publications, investment in privately held sector, and public R&D) to demonstrate that AI innovation is alive and well. While we must make progress to clarify the law that emerged through the Mayo/Alice framework set forth by the U.S. Supreme Court nearly a decade ago, there is no need for drastic measures, such as abolishing the judicially created exceptions or opening the floodgates to weaker patents by lowering the threshold for patent eligibility. Broad patents and functional claiming can do more to harm innovation than promote it. Particularly in the types of inventions at issue for AI, a meaningful subject matter eligibility threshold is critical to allow for further innovation. While narrower patent eligibility may reduce some private investments in this space--impacting smaller companies and startups, primarily--the existing restrictions to patent eligibility relating to AI offer critical opportunities for open innovation by reducing the risks that broad exclusive monopolies will impede efforts to advance the "basic building blocks" of AI innovation.