Machine learning enthusiasts have long touted the technology’s ability to perform–and sometimes exceed–human mental endeavors, such as identifying objects in images, generating a portrait painting, deciding to grant a loan application, optimizing a route to a destination, and efficiently responding to website visitor or customer queries. In recent years, such computerized “mental processes” have been denied patent protection, a trend underscored by U.S. federal district and Federal Circuit patent decisions issued in the wake of the U.S. Supreme Court’s seminal Alice Corp. v. CLS Bank Int’l opinion in 2014, which provided today’s legal framework for determining whether an invention is patent eligible or unpatentable. See Electric Power Group, LLC v. Alstrom SA, 830 F. 3d 1350, 1354 (Fed. Cir. 2016) (“we have treated analyzing information by steps people go through in their minds, or by mathematical algorithms, without more, as essentially mental processes within the abstract-idea category”) (citing cases).
Ironically, and despite this precedent, a record number of US patent applications have been submitted to the US Patent and Trademark Office (PTO) in recent years involving some aspect of machine learning (as of the end of 2018), and many of those inventions are being patented by the PTO at record numbers. At the same time, many AI-related patent applications have faced high rates of rejection following Electric Power Group. Thus, owners of machine learning patents may be understandably nervous that their US patents could be challenged by third parties under Alice and its progeny and found invalid by courts after further careful scrutiny, which for some companies could put their entire business enterprise in jeopardy. Newer federal district court decisions, however–including one from the Eastern District of Texas and another from the Central District of California discussed below–while maybe not signaling a post Alice sea change, may nevertheless provide added certainty to AI patents, at least those that were carefully drafted.
The first decision comes from Judge Richard Wu of the U.S. District Court for the Central District of California in a case involving Blackberry’s U.S. Patent 8,279,173, which claims a software-enabled device that selects a tag for a photo by displaying a tag entry field for entering a search string, displaying a tag list matching the string, and displaying a tag type indicator for each listed tag indicative of the source of the tag. Blackberry v. Facebook, No. 18-cv-1844-KSx, slip. op. at 35 (C.D. Cal. Oct. 1, 2019). In denying Facebook’s motion for summary judgment under Alice, Judge Wu dismissed Facebook’s “real-world long-standing human practice” argument. Id. “It is not necessarily clear,” he wrote, “that photo tagging, at least in the way described in the ‘173 patent…, has a true pre-computer analogue.” Although the claimed tagging invention can be “traced back to some real-world analogy,” Judge Wu explained, such as a person performing the mental and physical step of pre-labelling and categorizing sticky notes and then searching through them to find the right one to be placed on physical photographs, the claims nevertheless describe “fundamental improvement specific to the technological field of streamlining a process for a social-media-driven function.” Id. at 36 (identifying the use of a software-generated tag entry field on a display to allow searching for tag sources matching a search string, which is useful especially on small display screens).
The second decision comes from Federal District Judge Rodney Gilstrap of the U.S. District Court for the Eastern District of Texas. In denying an Alice-type motion for summary judgment brought by Wells Fargo against patent owner USAA, Judge Gilstrap considered Wells Fargo’s mental process abstraction argument. USAA v. Wells Fargo, No. 18-cv-00245-JRG, slip op. at 12-13 (E.D. Tex. Oct. 28, 2019). At issue were the claims of US Patent 8,977,571, which generally recite a software process involving monitoring a check in the field of view of a camera of a mobile device and capturing the image when “monitoring criterion” are met, and then providing the captured image data to a depository via a communications network. Id. at 3. Using Alice’s 2-part test, Wells Fargo argued that the claims are directed to a mental process of taking a picture and thus are abstract and ineligible for patenting.
Judge Gilstrap disagreed, noting that the use of monitoring criteria in the process is not akin to a human photo evaluation and could not be readily-assessed by a human in real time. This was not a situation, Judge Gilstrap said, where general-purpose computer components are added post-hoc to the simple process of taking a picture, because the human mind is not equipped to capture photographs on its own. To further explain, he wrote:
Finding that an invention is directed to an abstract idea simply because it removed the need for human intervention in a process is precisely the type of “high level of abstraction” that “all but ensures that the exceptions to § 101 swallow the rule.” Taken to its logical conclusion, inventions directed to everything from self-opening doors to self-driving cars would become ineligible for patent protection [under Alice]. Similarly, while the human mind may judge a digital image for subjective, aesthetic criteria, it is not equipped, particularly in real-time while looking at the image on a mobile device screen, to judge objective, technical criteria within the digital image that make the image acceptable for subsequent machine recognition.
Id. at 14, 17. According to Judge Gilstrap, the claims at issue employ a processor to perform a distinct process to automate a task previously performed by humans. But, the processor-enabled system’s use of monitoring criteria did not simply mirror the process previously used by human photographers. Indeed, the objective monitoring criteria were at the heart of the inventive improvement over previous methods.
So how do these two cases help owners of AI patents? While neither the USAA nor Blackberry patents disclosed machine learning techniques at their core (though both patents were asserted against technologies that use or could use machine learning), they demonstrate that software-based process claims can withstand post-grant Alice/Electric Power Group scrutiny even when all or part of the invention to which they are directed automates a mental process or analogous real-world long-standing human practice. The caveat is that the machine-enabled process must be one that involves a function that improves the technology being used as substitute for something humans have performed or reflects a way of doing something that no human could replicate (at least not practically, like determining weights for a neural network using back propagation or a long short term memory network calculation). In the case of Blackberry, the claimed process automated and streamlined the function of labeling things observed in images on a graphical user display. In Wells Fargo, the claimed process used complex criteria to judge the sufficiency of an image before taking a picture, criteria that could not be practically replicated in the human mind.
In those and other situations, the relevant patent eligibility inquiry may boil down to whether a patent’s claims are directed to an improvement in computer technologies or simply a use of computers as tools to enhance the efficiency of fundamental human tasks. See Blackberry at 17 (citing cases). Fortunately, the solution is relatively straightforward: patent practitioners must describe, carefully and fully, how a machine learning-based invention works when describing an invention in a patent application, and write claims that avoid simply expressing a result of mental processes or human performances done by a computer. Rather, claims should at least recite the feature of the AI system that relates to an improvement in a computer-implemented technology used by the system.