Mcle Self-study Article

• Jurisdiction: United States,Federal
• Author: Agatha H. Liu
• Publication year: 2019
• Citation: Vol. 44 No. 3

MCLE Self-Study Article

Agatha H. Liu

Hickman Palermo Becker Bingham LLP

DO I HAVE A PATENTABLE TECHNOLOGY IN ARTIFICIAL INTELLIGENCE?

(See end of this article for information on receiving 1.0 hour MCLE self-study credit.)

* * *

This article presents a systematic, tiered approach to assessing patentability of technologies related to artificial intelligence ("AI"). This article first describes the recent increasing developments in AI and the need for guidance in assessing patentability of eligible AI-related technologies. Next, this article shows a systematic, tiered approach, based on a classification of AI-related technologies. This article further offers advice on effective approaches to assessing patentability of AI-related technologies.

Surge of AI-Related Technologies

AI has attracted tremendous attention both in technical research and the popular media. AI typically refers to the apparent intelligence exhibited by a computer that has been programmed to simulate natural functions of the human brain. The main reason that AI has become so popular is the desire for a "smarter" world that requires less human intervention and the explosion of digital data that triggers and further fuels such a desire. Right now, a main approach towards achieving a smarter world is to deploy computer programs that automatically update data models and thus "learn" from large amounts of data that capture human behavior that can be derived from interaction with online services, applications, or databases. As time goes on, inventors and developers likely will focus more efforts on simulating aspects of human intelligence more advanced than learning, such as reasoning or problem solving.

AI has been a research field in computer science since at least the 1950s,¹ and machine learning has been a fundamental branch of AI since the research field's inception.² Without knowing precisely how humans convert external signals into internal information, the research field has modeled the human learning process at various levels. Interestingly, knowledge-based approaches representing the human learning process at a higher, conceptual level, such as rule-based inference engines, did not dominate until later,³ while symbolic methods representing the human learning process at a lower, physiological level, such as artificial neural networks, came into existence early on.⁴ While the knowledge-based approaches might be more human-understandable, the symbolic methods might better capture the intricacies of human cognition, which partly explains why they are regaining popularity.

Uncertainty in Determining Patentability of AI-Related Technologies

Machine learning typically involves creating and applying a data model that takes an input set of items, each having a list of features, and produces an output set of items indicating classifications of or relationships among the input set of items.⁵ The input set of items may correspond to existing human actions, and the output set of items may correspond to a prediction of human behavior as a result of a machine learning from the human actions. As noted above, the data model may represent the human learning process at one of different levels. The treatment of a machine learning method generally involves a design phase followed by a building (creation) phase and an execution (application) phase. An AI-related technology may be an invention that fits anywhere within the overall paradigm for machine learning, as further discussed below.

As AI becomes more involved in technological advances, the prospect of patent protection for an AI-related technology may have increased interest. Beyond the enthusiasm, however, the question of how to seek patent protection for AI-related technologies also arises more frequently. One reason for such uncertainty is that because AI involves simulating natural intelligence, whether an AI-related technology merely replicates existing human behavior (and thus possibly ineligible under 35 U.S.C. § 101) or goes beyond replication to deserve the right to exclude others from using that technology may be unclear. In particular, people start to conceive AI-related technologies to resolve all sorts of issues as they begin to realize the potential of AI or specifically machine learning. While these technologies might sometimes be novel, whether they are obvious (and thus possibly unpatentable under 35 U.S.C. § 103) remains a question that is often not easily answered. The fact that AI may be involved to different extents in AI-related technologies tends to further complicate the situation, as further discussed below.

[Page 13]

Inventors have filed close to one thousand applications with the United States Patent and Trademark Office (USPTO) for AI-related technologies (class 706) since 2015, including close to four hundred specifically related to machine learning (subclasses 12, 14, 16, and 25). The Federal Circuit has not heard many cases involving AI-related technologies but recently upheld...