The U.S. Patent and Trademark Office (USPTO) has updated its patent subject-matter eligibility guidance to address inventions involving artificial intelligence (AI). Our Intellectual Property Team examines how the revised guidance will impact AI patent strategies.
- The guidance treats AI inventions largely as it treats other complex computer-implemented inventions
- The AI guidance standardizes existing USPTO practices and summarizes precedent for all examiners to follow
- New patent applications should focus on the practical, nonmathematical technological improvements of AI
On July 17, 2024, the U.S. Patent and Trademark Office (USPTO) issued an update to its patent subject-matter eligibility guidance to specifically address artificial intelligence (AI) inventions. The AI guidance is effective immediately and aims to clarify the types of AI-related inventions that are eligible for patenting under Section 101 of the Patent Act in light of recent technological advancements and court decisions. These court decisions prohibit the patenting of inventions that fall into certain excluded categories, including abstract ideas (e.g., mathematical calculations, mental processes, and methods of organizing human activity); laws of nature; and natural phenomena. This new guidance affects to varying degrees patent strategies involving AI, machine learning (ML), and other software concepts.
Background
The AI guidance is certainly different from the typical USPTO examiner’s interpretation of AI-related inventions as recently as a few years ago. It may seem like a shift in the interpretation of Section 101, but the AI guidance is not the cause or leading edge of the shift. Instead, the AI guidance reflects a solidification of the USPTO’s gradual policy change to unify treatment of Section 101 across the office and align the USPTO’s examination with recent precedent.
Since the Supreme Court Alice and Mayo decisions—Alice Corp. v. CLS Bank International and Mayo Collaborative Services v. Prometheus Labs—in the early 2010s, the USPTO has made several attempts to establish concrete guidance for examiners to analyze inventions on the edge of patent eligibility. These efforts culminated in a relatively comprehensive set of guidance documents and examples published in 2019. The 2019 guidance described a compartmentalized subject-matter eligibility analysis mirroring the Alice/Mayo framework established by the Supreme Court. A major contribution of the 2019 guidance was the USPTO’s collecting and organizing the existing precedent on “abstract ideas” into a set of three buckets: (1) mental processes; (2) mathematical concepts; and (3) certain methods of organizing human activity (including “fundamental economic practices” and similar business concepts).
By most accounts, the 2019 guidance was a very patentee-friendly interpretation of Section 101 precedent and the high-water mark of the post-Alice eligibility treatment. However, coupled with a general shift in the makeup of the Federal Circuit, the more recent precedent has not generally followed the 2019 guidance. The Federal Circuit has continued to be skeptical of cases that don’t offer something akin to a “technical solution” to a “technical problem” (whether because the claims are too high level or because the claims are purely directed to improvements of an underlying abstract idea) with a heavy reliance on a patent’s specification to weed out creative arguments/shifts in the focus of claims. Free Stream Media Corp. v. Alphonso Inc. provides a good example of specification’s role in a subject-matter eligibility decision under Section 101.
More recently, our anecdotal experience is that the USPTO has been standardizing their examination and shifting in the same direction as the Federal Circuit, with informal, unwritten changes being made nearly continuously in recent months.
Overview of the AI Guidance
The AI guidance does not replace the 2019 guidance and doesn’t make sweeping changes to the scope of the abstract idea groupings or the overall subject-matter eligibility analysis the USPTO currently uses. Instead, it focuses on summarizing existing caselaw and introducing new hypothetical examples to map the 2019 guidance and the new examination standards to AI- and other software-related issues.
Generally, the AI guidance’s treatment of AI is consistent with recent public comments that view AI/ML as a subset of computer-implemented inventions. The USPTO generally addresses AI-related concepts under the “mental process” or “mathematical calculation” frameworks—with the possibility of implicating “organizing human activity” or “fundamental economic practice” with the particular use cases of AI. When referencing AI/ML, the AI guidance focuses heavily on the practical application analysis of the 2019 guidance, and notes that “[m]any claims to AI inventions are eligible as improvements to the functioning of a computer or improvements to another technology or technical field.”
The AI guidance provides insights into both AI/ML and other software-related inventions (e.g., fintech) for both prongs of the second step of the two-step subject-matter eligibility analysis and is accompanied by hypothetical examples that attempt to contextualize these insights.
Handling of various types of abstract ideas
Organizing Human Activity. The guidance cites the Federal Circuit’s 2020 decision in Bozeman for the proposition that financial issues (e.g., fraud detection) continue to fall within the category of organizing human activity (fundamental economic principles being a subset) without another eligibility hook (e.g., another non-abstract idea that is improvement by a claimed invention). This is true even where basic, generic AI/ML is deployed to perform the financial activity.
Mental Processes. The guidance reaffirms that the “mental processes grouping is not without limits, and as such, claim limitations that only encompass AI in a way that cannot practically be performed in the human mind do not fall within this grouping.” While initially promising (albeit duplicative of past precedent and guidance), the only additionally provided example of such a case is a claim directed to a specific radio frequency identification (RFID) data structure that is uniquely encoded based on a physical RFID transponder as addressed in the ADASA case. Like other eligibility examples that deal with sound (e.g., the new hypothetical example 48) or images (e.g., the ubiquitous hypothetical example 39 from the 2019 guidance), these applications of AI are clearly not mental processes because the type of media/signal would not lend itself to mental processing at the granularity required by the claim, but they don’t provide a lot of instruction on edge cases (e.g., similarly complex processing on textual data) or much confidence in claims that relate solely to complex processing.
Mathematical Processes. The guidance is brief regarding the category, but it is addressed in the hypothetical examples. Example 47, in particular, states that conventional ML techniques, such as back propagation of errors, include a series of mathematical operations that fall within the mathematical processes bucket. This characterization of mathematical processes may be a trap into which a lot of pure AI/ML inventions fall without some underlying technical hook outside the model and is something we have been monitoring for some time.
Practical applications of abstract ideas
In some instances, a practical application of an abstract idea may still be eligible for patenting. The AI guidance gives some examples of how to analyze whether an AI-related invention is integrated into a practical application (and thus patent eligible) by comparing (1) the improved haplotype phase determination using a Hidden Markov Model (see In re Board of Trustees of Leland Stanford Junior University (held ineligible)) with (2) rule-based animation of lip synchronization (see McRO Inc. v. Bandai Namco Games America Inc. (held eligible)). The AI guidance indicates that the USPTO will focus on whether the AI technology is merely an abstract idea “applied” on a computer or an improvement to the computer or other technology, which again is consistent with how software cases have been handled previously. The glaring hole left by the AI guidance is any insight into whether AI/ML itself is a technology that could be improved to integrate an abstract idea into a practical application.
New hypothetical examples
The new hypothetical examples share the deficiencies of the AI guidance and do not, in our view, go sufficiently in depth around improvements to AI itself or address any issues not already squarely resolved by the Federal Circuit. In our view, the new example 47 comes the closest to addressing AI by describing (via claim 2) that “training an [artificial neural network (ANN)] using a selected algorithm” comprising “a backpropagation algorithm and a gradient descent algorithm” is an abstract mathematical calculation (i.e., not eligible) and that simply training the model is insufficient to overcome Section 101 but describes (via claim 3) that using the trained model to improve a downstream technology by preventing network intrusions is eligible. Example 47 (claim 1) purports to show the eligibility of a more structurally defined ANN and while it seems to open the door to low-level descriptions of AI inventions, we view it as a red herring whose eligibility will disappear as soon as any context or functionality is added to the claim without also claiming a new, fundamental improvement to the ANN.
Takeaways
A fair characterization of the AI guidance is that AI is treated much the same as other complex computer-implemented inventions in that conventional AI alone will not take a previously ineligible concept and transform it into eligible subject matter. Overall, arguments based on “preemption” or “complexity” in isolation remain unlikely to succeed, and permutations of “receiving,” “analyzing,” “displaying,” or “storing,”—regardless of the degree of technology/AI involved—will continue to be treated with extreme skepticism by the USPTO. However, downstream technical use cases; unique, granular improvements to the computer’s ability to execute the AI model; and other atypical technological steps or effects before, during, or after the modeling remain strong options for pursuing patent protections in the United States.
The AI guidance’s most significant development is simply the standardization of practices that had been quietly used within the USPTO for some time and the summarization of existing precedent in a way that all examiners can follow. This will provide concrete examples to help minimize inconsistencies between examiners and will be particularly useful for examiners having less experience with subject-matter eligibility issues. As might have been expected, the AI guidance does not tread new ground or resolve previously uncertain issues one way or the other.
Impact on Patent Strategies
In short, while the guidance provides a helpful summary and helpful reference materials when negotiating with the USPTO, we see little impact to current patent strategies for AI-related technology based on the AI guidance.
Fintech, insurtech, health care, and other business-related technologies involving AI have been, and will likely always be, pinched between the “software” treatment of AI and the general aversion to business use cases in current precedent. Recommended patent strategies for these AI use cases should focus on the improvement to the computer technology itself as a bare minimum, and if available, focusing on improvements to “other technology” facilitated by the AI (e.g., not purely improvements to the abstract ideas facilitated by AI but improvements to technology that do not have a mental or historical business equivalent). Consistent with these strategies, new patent applications should be drafted with consideration of the practical, nonmathematical, technological improvements brought about by improved AI models/modeling techniques.
[View source.]
