PatentNext Summary: Because AI is a relatively newer technology, court cases analyzing AI-related patents have been few in number. Given the increased numbers of AI patent filings, as recently reported by the USPTO, we can expect to see future court cases involving AI-related patents. We can also expect that courts will analyze these AI-related patents per the two-part Alice with the same scrutiny as we have seen for more general software-related patent cases.
In the last quarter of 2020, the United States Patent and Trademark Office (USPTO) reported that patent filings for Artificial Intelligence (AI) related inventions more than doubled from 2002 to 2018. See Office of the Chief Economist, Inventing AI: Tracking The Diffusion Of Artificial Intelligence With Patents, IP DATA HIGHLIGHTS No. 5 (Oct. 2020).
But what about the courts? How does such an increase in the number of patent filings translate to the number of related court decisions? And given that AI fundamentally relies on software, how do courts analyze AI-related inventions for patentability under the Supreme Court’s two-part patent eligibility test as described in Alice Corp. v. CLS Bank International (“Alice”)?
This article briefly explores these questions in view of the (few) post-Alice cases that have considered AI-related patent inventions.
Alice Primer on Software-Related Inventions and its Applicability to AI
Following Alice, courts have routinely applied the Supreme Court’s two-part Alice test to analyze general software-related inventions. It comes as no surprise, therefore, that courts have applied, and continue to apply, this test to AI-related inventions.
This is because AI is fundamentally a software-related technology. For example, an AI-related invention typically comprises some form of AI architecture and its related use. This can include a training dataset, an AI algorithm or technique (such as machine learning or a neural network) that inputs the training dataset to produce an AI-based model, and execution of that AI-based model by a computing device to achieve an end purpose.
While the Federal Circuit has yet to analyze an AI-related patent (having AI claim elements) with the two-part Alice test, it has analyzed more general software-related patents with this test. For the purposes of this article, a brief Alice primer as to software-related inventions follows.
Under Alice step 1, courts ask whether a given AI or software-related invention “directs to” an “abstract idea.” An “abstract idea” is one type of “judicial exception” to the “inventions patentable” as listed by 35 U.S.C. 101 (regarding “patent eligibility”). And because “abstract ideas” comprise “mental processes” and “mathematical concepts,” courts typically analyze software-related inventions (including AI-related inventions) under the “abstract idea” rubric (with other “abstract ideas” being “laws of nature” and “natural phenomenon,” which rarely arise for software-inventions).
If a given claim is not “directed to” an abstract idea per Alice step 1, then the invention is patent-eligible.
If an abstract idea is found, however, then the analysis proceeds to Alice, step 2. Under step 2, a court asks whether a given AI or software-related invention nonetheless possesses an “inventive concept.” That is, do the claim(s), when considered as a whole, include additional limitations amounting to “significantly more” than the abstract idea itself? If so, then the invention is patent-eligible. Otherwise, it is not.
Early Court Treatment of AI (following Alice)
In the early years following Alice (since mid-2014), courts analyzing AI-related patents (and, more generally, software-related patents) typically applied Alice at the early/pleading stages of a case.
Many such cases found respective patents-at-issue invalid for reciting high-level mathematical formulas and/or mental processes as performed by generic computing “modules” or by general-purpose computers. For example, in the few months following Alice, Judge Mayer, writing in a concurring opinion, encouraged this approach. See I/P Engine, Inc. v. AOL Inc., 576 Fed. Appx. 982 (Fed. Cir. 2014) (per curiam; unpublished) (Mayer, J., concurring) (stating that “[f]rom a practical perspective, there are clear advantages to addressing section 101’s requirements at the outset of litigation” and that “unnecessary litigation … could have been avoided [here]” if the claims-at-issue had been invalidated under the two-part Alice test, where such claims described a system for filtering “information for relevance to a user’s query using combined content and collaborative data,” where the patent owner argued that such collaborative data could be combined with other data via the use of a “complex neural network function,” but where the claims lacked such element, Judge Mayer stating that “[the] claimed invention, which describes a system which filters information for relevance to a user’s query using combined content and collaborative data … does not pass muster under section 101.”)
District courts also followed suit. For example, in Vehicle Intelligence & Safety LLC v. Mercedes-Benz USA, LLC, the court applied the two-part Alice test to find patent claims invalid on a Rule 12(c) motion for judgment on the pleadings. 78 F. Supp. 3d 884 (N.D. Ill. 2015). The patent-at-issue generally described “expert systems” used “to screen equipment operators for impairments, such as intoxication, physical impairment, medical impairment, or emotional impairment” and to selectively test the equipment operators and control the equipment (e.g., automobiles, trucks, [etc.]) if impairment of the equipment operator is determined.” While the patent-at-issue did not expressly define an “expert system,” the court acknowledged that an expert system exemplified an “application of [AI].” Nonetheless, because the claims failed to “recite any new or improved computer technology or provide new physical components” (claim 8) beyond the mere replication of an “expert system” and its methods could be “performed entity in the human mind” (e.g., claim 15), then such claims were found directed to mere “abstract ideas.” Each of the claims also did not possess an “inventive concept” because the remaining claim elements comprised mere generic computer components without “significantly more.” Thus, the claims were invalid under 35 U.S.C. 101. See also Blue Spike, LLC v. Google Inc., 2015 WL 5260506 (N.D. Cal. Sept. 08, 2015) (also finding claims invalid via the two-part Alice test on a Rule 12(c) motion for judgment on the pleadings, with patent owner arguing that such invalidation could render “breakthroughs in artificial intelligence” categorically unpatentable, but where the claims failed to recite any artificial intelligence element, and where the court found that the claims “merely discuss[ed] using routine computer components.”). See also Neochloris, Inc. v. Emerson Process Mgmt. LLLP, 140 F. Supp. 3d 763 (N.D. Ill. 2015) (also finding claims invalid via the two-part Alice test on summary judgment although at least one claim reciting “an artificial neural network module,” the court finding that “it is not even clear [from the specification or claim itself] what [that term] refers to besides a [generalized] central processing unit—a basic computer’s brain.”).
In Purepredictive, Inc. v. H2O.AI, Inc., the court applied the two-part Alice test to find patent claims invalid on a Rule 12(b)(6) motion to dismiss. 2017 WL 3721480 (N.D. Cal. Aug. 29, 2017). The patent-at-issue related to “an automated factory for predictive analytics.” The claims-at-issue recited a method that included generating “a plurality of learning functions” and involved selecting the most effective learned functions to create a rule set for further data input. The patent owner argued that its company (and related technology presumably embodied in the patent) employs “artificial intelligence to provide insight into a business’s data through the use of predictive modeling,” the court, however, found the claims-at-issue directed to “a mental process” (an abstract idea), merely using “mathematical algorithms to perform predictive analytics.” Because the claims failed to “described specific system architecture” and instead made reference to generic “modules,” the court found the claims lacked an “inventive concept” and, therefore, held the claims ineligible pursuant to Section 101. See also Power Analytics Corp. v. Operation Tech., Inc., 2017 BL 475384 (C.D. Cal. July 13, 2017) (finding claims invalid via the two-part Alice test on a motion for partial summary judgment, where certain claims recited a “machine learning engine,” but where the court found that the patent merely described that term in functional terms without purporting to add any particular inventive implementation, the court finding the claims invalid as abstract and generic that merely “describe[d] desired functions or outcomes, but [did] not, individually or in combination, constitute ‘inventive concepts.’”)
More Recent Court Treatment of AI
While the Federal Circuit has yet to analyze an AI-related patent with the two-part Alice test, at least some judges have signaled caution in getting questions about “important inventions” (such as AI) wrong.
In Smart Sys. Innovations, LLC v. Chicago Transit Auth., Judge Linn, writing in dissent, described the uncertainty of patent eligibility review using the two-part Alice test and the dangers inherent to new age technology, such as Artificial intelligence, of arriving at the wrong result. 873 F.3d 1364, 1378 (Fed. Cir. 2017) (Linn, J., dissenting and concurring in part) (“Despite the number of cases that have faced these questions and attempted to provide practical guidance, great uncertainty yet remains. And the danger of getting the answers to these questions wrong is greatest for some of today’s most important inventions in computing, medical diagnostics, artificial intelligence, the Internet of Things, and robotics, among other things.”).
Similarly, in Athena Diagnostics, Inc. v. Mayo Collaborative Servs., Judge Moore, writing in dissent, expressed her concern as to the impact of Alice (and related case Mayo). LLC, 927 F.3d 1333 (Fed. Cir. 2019) (per curiam). The Athena Diagnostics court had denied a petition for rehearing en banc regarding claims directed to a “method for diagnosing neurotransmission or developmental disorders.” The court, issuing the opinion per curiam, expressed its obligation (and concern) in holding the claims ineligible. Id. (Lourie, J.) (“I concur in the court’s decision not to rehear this case en banc. In my view, we can accomplish little in doing so, as we are bound by the Supreme Court’s decision in Mayo. … Some of us have already expressed our concerns over current precedent”). In dissent, Judge Moore added: “I do not fault my colleagues, who under protest have concluded that they have no choice but to hold the claims in Athena ineligible because of Mayo.” Judge Moore cited several amicus and senate hearings, including a discussion concerning AI-related inventions in the biotech space. See id. (citing The State of Patent Eligibility in America, Part II, 116th Cong. 9 (2019) (written testimony of Henry Hadad, President, IPO) (“[C]onfusion about what is patent-eligible discourages inventors from pursuing work in certain technology areas, including discovering new genetic biomarkers and developing diagnostic and artificial intelligence technologies. [This] uncertainty disincentivizes the enormous investment in research and development that is necessary to fuel the innovation cycle.”)).
More recent district court decisions show that the Federal Circuit’s cautionary statements may have trickled down, with some courts seemingly reluctant to grant case dispositive motions, at least at the early stages of a case.
For example, in Singular Computing LLC v. Google LLC, 2020 U.S.P.Q.2d 10708 (D. Mass. 2020), the court denied a 12(b)(6) motion to dismiss, where the defendant alleged that the asserted claims were patent-ineligible per the 2-step Alice test. The patent-at-issue generally related to “a processor or other device” that included “processing elements designed to perform arithmetic operations . . . on numerical values of low precision but high dynamic range [i.e., LPHDR].” The patent owner had argued that such architecture allows “for more efficient use of a computer’s transistors, which improves computer performance in certain applications such as artificial intelligence software.” In denying the motion, the court focused on the alleged improvement (via LPHDR technology) to existing computer architecture, finding that such improvement made the claimed invention plausibly inventive, which was at least enough for the patent claims to survive at the pleading stage where allegations as to inventiveness were accepted as true.
Still, other courts continue to find AI-related patents invalid at the outset, especially where the respective claims lack any AI-specific elements or architecture. For example, in Applied Predictive Techs., Inc. v. Marketdial, Inc., the court granted a Rule 12(b)(6) motion to dismiss where the claims lacked AI-specific elements and were instead “directed to the abstract concept of optimizing parameters for business initiative testing.” 2020 Us Dist Lexis 221981 (D. Utah Nov. 25, 2020). In Applied Predictive Techs, even though the specification described the possible use of a “neural network” to perform the claimed “testing,” the court found that “no specific, technical details regarding the exact sort of analysis” was provided, and, therefore, the claims were considered to cover “history data analysis” broadly. “To grant a patent on the concept as it is articulated in the [patent-at-issue] would be to preempt any method of analyzing historical data to determine how certain parameter settings influence that data.” Thus, the court granted the motion finding that claims-at-issue patent ineligible. See also Kaavo Inc. v. Amazon.com Inc., 323 F. Supp. 3d 630 (D. Del. 2018) (finding claims invalid for lack of subject matter eligibility using the two-part Alice test on a motion for summary judgment, where certain claims-at-issue recited methods of “forecasting” but lacked any AI-specific elements, even though the specification explained that such “forecasting” may be performed by “neural networks” or by “Load Forecasting”/”Pricing” modules, but where the court found that “claim language [did] not require the use of these modules.”).
Clues to Navigating AI/Software-Related Patents in Future Cases
With few AI-related cases issued to date, we can expect that courts (or parties) in future cases will look to examples of AI-related claims and cases as provided by the USPTO, at least for guidance.
For example, along with its 2019 Revised Patent Subject Matter Eligibility Guidance (the “2019 PEG”), the USPTO provided several example patent claims and respective analyses under the two-part Alice test. See Subject Matter Eligibility Examples: Abstract Ideas. One of these examples (“Example 39”) demonstrated a patent-eligible artificial intelligence invention. In particular, Example 39 is labeled “Method for Training a Neural Network for Facial Detection” and includes claim elements for training a neural network across two stages of training set data so as to minimize false positives for facial detection. The USPTO’s analysis informs that the claim of Example 39 is patent-eligible (and not “directed to” an abstract idea) because the claim does not recite any mathematical concept, mental process, or fundamental economic concept.
As a further example, the Patent Trial and Appeal Board (PTAB) more recently applied the 2019 PEG (as revised) in an ex parte appeal involving an artificial intelligence invention. See ex parte Hannun (formerly Ex parte Linden), 2018-003323 (April 1, 2019) (designated by the PTAB as an “Informative” decision). In Hannun, the patent-at-issue related to “systems and methods for improving the transcription of speech into text.” The claims included several AI-related elements, including “a set of training samples used to train a trained neural network model” as used to interpret a string of characters for speech translation. Applying the two-part Alice test, the Examiner had rejected the claims finding them patent-ineligible as merely abstract ideas (i.e., mathematical concepts and certain methods of organizing human activity without significantly more.) The PTAB disagreed. While the PTAB generally agreed that the patent specification included mathematical formulas, such mathematical formulas were “not recited in the claims.” (original emphasis). Nor did the claims recite “organizing human activity,” at least because the claims were directed to a specific implementation comprising technical elements, including AI and computer speech recognition. Finally, and importantly, the PTAB noted the importance of the specification describing how the claimed invention provides an improvement to the technical field of speech recognition, with the PTAB specifically noting that “the Specification describes that using DeepSpeech learning, i.e., a trained neural network, along with a language model ‘achieves higher performance than traditional methods on hard speech recognition tasks while also being much simpler.’”
Because AI is a relatively newer technology, court cases analyzing AI-related patents have been few in number. Given the increased numbers of AI patent filings, as recently reported by the USPTO, we can expect to see future court cases involving AI-related patents. We can also expect that courts will analyze these AI-related patents per the two-part Alice with the same scrutiny test as we have seen for more general software-related patent cases. See Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016) (analyzing a self-referential database software-related patent under the two-part Alice test). Courts (and parties) may also look to USPTO cases and examples for guidance.