Federal Circuit Rules On § 101 Eligibility of AI Machine Learning Patents

The subject matter eligibility of patents claiming AI technologies or using AI has been the topic of substantial discussion in the IP community in recent years. The volume of AI-related U.S. patent applications has more than doubled in the past two decades.¹ At the same time, the Supreme Court’s 2014 decision in Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014), and lower courts’ application of Alice have cast doubt on the eligibility under 35 U.S.C. § 101 of many patents directed at software, algorithms, and automating human activities or business methods. The Alice decision set forth a two-step test for evaluating eligibility, with step 1 evaluating whether a patent’s claims are directed to an abstract idea and, if so, proceeding to step 2, which evaluates whether the claims disclose an inventive concept that supports eligibility.² Alice led to a significant increase in successful § 101 challenges: in 2008-14, before Alice, there was an average of 19 successful § 101 challenges per year; between 2015-22, after Alice, that average has shot up over ten-fold to 217 per year.³ Given these competing trends, the U.S. Patent Office last year published guidance on eligibility of AI-related patents, in an attempt to provide clarity on when AI-related inventions are patent-eligible or not.⁴

Last week, the Federal Circuit weighed in, issuing a unanimous, precedential opinion in Recentive Analytics, Inc. v. Fox Corp., a case that squarely addresses the eligibility of patents directed to using machine learning.⁵ The opinion affirmed the ineligibility of four patents under § 101 and Alice. It suggests a challenging outlook for the eligibility of AI-related patents, though the opinion’s concluding remarks seek to narrow its holding, and the case’s facts were particularly unfavorable to the patentee, Recentive, under recent § 101 precedents.

Recentive’s four patents recited using machine learning to optimize scheduling of live events and TV programming. Two of the patents recited using a “ML [machine learning] model,” with steps of: collecting input parameters for live events; iteratively training the ML model using the event parameters; using the ML model to output an optimized schedule; and updating the schedule based on changes in the parameters.⁶ The two other patents related to constructing a “network map” (i.e., information describing what programs to broadcast in certain geographies and times) using “machine learning.” The claims recited steps of: collecting input data; using machine learning to analyze the input data and output an optimized network map; and then updating network map based on changes to input data.⁷

In rejecting the four patents as patent-ineligible, the Federal Circuit first concluded that the patents did not claim any improvements on machine learning technology.⁸ Rather, as Recentive’s counsel and the patents’ specifications admitted, the patents used conventional, generic machine learning models.⁹ Although the claims recited a seeming specific technical feature of “iterative training,” that feature is a necessary and inherent requirement of any machine learning model and thus failed to support eligibility.¹⁰ Also unavailing was the patents’ application of machine learning to activities such as event scheduling or network maps, given that those activities predated computers and could be done manually by humans.¹¹ 

After rejecting the four patents, the opinion concludes by attempting to cabin its holding and its potential future impact on AI-related patents, stating:

Machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology. Today, we hold only that patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.¹²

The Recentive opinion suggests a challenging environment for those seeking to patent AI-related technologies and enforce them in litigation in the future. To be sure, the Federal Circuit’s concluding remarks suggest AI-related inventions can be patented if specific “improvements” to the underlying AI technology (e.g., machine learning) are disclosed. Recentive also presented a perform storm of facts that weighed against Recentive’s patents. First, as the Federal Circuit noted, the claims simply invoked a generic, conventional “ML model” as a black box that performed the claims’ key steps. Second, the ML model was applied to automating common human activities, like event scheduling. In past cases, the Federal Circuit applying Alice has stated that patents that take a human activity and simply claim “do it on a computer” or carrying out the activity using the Internet are ineligible.¹³ In the same vein, Recentive’s patents claim event scheduling and “do it using machine learning.” It may therefore be unsurprising that the patents were found ineligible under the Federal Circuit’s past § 101 jurisprudence, and the opinion’s closing remarks are consistent with past decisions explaining that patents claiming to “improve the functioning of the computer itself” may be eligible.¹⁴

On the other hand, it is unclear how describing improvements to machine learning could confer eligibility under other § 101 precedents. Long before the Alice decision in 2014, pure algorithms have been considered an ineligible category of subject matter under § 101.¹⁵ Thus, elaborating on the steps of a new, improved machine learning algorithm may not result in eligibility.

As another possibility, applying AI technologies to more physical, hardware environments—as opposed to intangible environments or human activities (like event scheduling)—may give patentees an alternative path to eligibility. For example, in Ocado Innovation, Ltd. et al. v. AutoStore AS, et al, patent claims related to optimizing a robotic warehouse system survived a motion to dismiss because the elements described may be innovative.¹⁶ The district court reasoned that the “control system described in the [patent]” could not be “performed ‘entirely’ in the human mind and is thus not persuaded by [defendant’s] analogies to human activities.”¹⁷

The eligibility of AI patents, therefore, remains uncertain after Recentive, given its particularly unfavorable facts for the patentee. With the substantial increase in AI patents in recent years, the Federal Circuit will likely have additional opportunities to elucidate the boundaries of eligibility of such patents. O’Melveny will continue to monitor and provide updates on the Federal Circuit’s patentability decisions relating to AI patents.