The evolution of artificial intelligence (AI) technologies has triggered a surge in the filings of patent applications, from machine learning models to applications of those models. See USPTO, Artificial Intelligence (AI) trends in U.S. patents (June 29, 2022) at 7. But because AI technologies rely on machine learning algorithms and computing power, AI inventions face challenges in light of the legal standards under Section 101 of the Patent Act.
Section 101 sets forth eligibility requirements for patents, providing that they may be granted only for a “new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof.” 35 U.S.C. § 101. In Alice Corporation v. CLS Bank International, the U.S. Supreme Court set forth a two-part test to determine if a claimed invention is eligible for patent protection: (1) whether the claim of an invention is directed to a patent-ineligible concept, such as an abstract idea; and (2) if so, whether the claim recites an “inventive concept” sufficient to transform the abstract idea into a “patent-eligible application.” 573 U.S. 208, 217-18 (2014).
In a recent case in the District of Delaware, the court found that the plaintiff’s patents claimed no more than an abstract idea “using known generic mathematical techniques.” See Recentive Analytics, Inc. v. Fox Corporation, No. 22-cv-1545-GBW, 2023 WL 6122495, *8 (D. Del. Sept. 19, 2023) (appeal pending). The two patents both involved machine learning techniques, directed to methods for (1) generating television network schedules, and (2) optimizing event schedules by considering a variety of parameters. Id. at *1–2.
Regarding the first Alice prong, the court found that the patents recited collecting and analyzing information, a “familiar class of claims directed to a patent-ineligible concept.” Id. at *9. The court distinguished the patents from those in McRO, Inc. v. Bandai Namco Games America Inc. See 837 F.3d 1299 (Fed. Cir. 2016) (claims reciting software for animating lip synchronization and facial expressions of animated characters were not directed to an abstract idea). First, the television network maps produced by Recentive’s methods were “less tangible than the created animated characters” in McRO. Recentive at *8. Second, “changing a process where both humans and algorithms are trying to maximize TV ratings” was different from transforming “a traditionally subjective process performed by human artists into a mathematically automated process” in McRO. Id. Finally, the rules Recentive claimed were “conventional machine learning techniques,” unlike the “specific and unconventional” ones in McRO. Id.
Regarding the second Alice prong, the court held that the patents did not involve an “inventive concept.” Id. at *12. Instead, “[t]he machine learning limitations [were] described only in broad functional terms and provide[d] little guidance on model parameters or training technique[.]” Id. For example, the patents vaguely disclosed “any suitable machine learning technique” and “using either a neural network or a support vector model and iteratively training it.” Id.
Recentive sought to distinguish the claimed inventions from the way human brains might process the same information, emphasizing that “humans process data qualitatively, rather than quantitatively,” and that “[t]he number of possible solutions is far beyond what a human could process.” Id. The court rejected these arguments, reasoning that it is irrelevant whether a human brain would process the information the same way as AI. Rather, “[t]he relevant question is whether the machine learning processes are mathematical algorithms.” Id. The court found that no amount of data or computing power made a meaningful difference: “humans can engage in the mathematical techniques to perform machine learning (albeit slowly).” Id.
Claims directed to novel methods for improving prior art neural networks, however, can still be eligible. The court noted that a “Method for Training a Neural Network for Facial Detection” identified in the USPTO’s guidance (M.P.E.P. § 2106.04(a)(1)) described such methods as “an expanding training set using mathematical transformations and the minimization of false positives using a distinctive training method.” Recentive, 2023 WL 6122495 at *6. Recentive’s claims were not analogous. Instead, they merely “relate[d] to the application of machine learning techniques to a manual process.” Id.
With Recentive in mind, patentees should be prepared to articulate what makes their AI claims patentable, for example, by focusing on specific and unconventional machine learning techniques claimed in the patents.