PatentNext Takeaway: The USPTO announced its 2024 Guidance Update on Patent Subject Matter Eligibility, particularly focusing on Artificial Intelligence (AI). Effective July 17, 2024, this guidance aims to address examination procedures for U.S. patent applications under 35 U.S.C. § 101, following President Biden’s executive order on the safe development and use of AI. The 2024 AI Guidance is designed to help USPTO personnel apply existing subject matter eligibility rules to AI-related inventions during patent examination, appeal, and post-grant proceedings. It includes case examples from the Federal Circuit, which, although not AI-specific, are relevant for understanding software-related arts. Additionally, it introduces hypothetical examples (new example claims 47-49) illustrating how AI-related patent claims will be analyzed for eligibility. These examples suggest that examiners will scrutinize AI-related claims more rigorously, emphasizing the need for detailed descriptions of how AI features improve technology or technical fields and/or provide a specific medical treatment. While the guidance does not constitute new law, it replaces previous guidance and is expected to be integrated into the MPEP eventually.

****

Introduction

On July 16, 2024, the United States Patent and Trademark Office (USPTO) announced its 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence (the “2024 AI Guidance”). The 2024 AI Guidance addresses examination procedures for U.S. patent applications in view of 35 U.S.C. § 101 (Subject Matter Eligibility). The Guidance is effective as of July 17, 2024.

According to the USPTO, the 2024 AI Guidance was issued as an “update on subject matter eligibility to address innovation in critical and emerging technologies (ET), especially artificial intelligence (AI).” 89 FR 58128. Further, the guidance was issued in view of President Biden’s executive order on the “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.” See PatentNext: Intellectual Property (IP) impacts from President Biden’s Executive Order on Artificial Intelligence (AI)

The 2024 AI Guidance is intended to “assist USPTO personnel in applying the USPTO’s subject matter eligibility guidance to AI inventions during patent examination, appeal, and post-grant proceedings.” 89 FR 58128. 

The 2024 AI Guidance includes a section of case examples from the United States Court of Appeals for the Federal Circuit (the Federal Circuit), which the 2024 AI Guidance describes as “informative.” None of the case examples regard AI-related patents or claims. Instead, the case examples illustrate Section 101 (Subject Matter Eligibility) rulings in the software-related arts more generally, which is somewhat helpful given that AI inventions are fundamentally software inventions.

The 2024 AI Guidance also includes a section describing and linking to three new hypothetical example claims 47-49. See July 2024 Subject Matter Eligibility Examples 47-49. This is likely the most useful section of the guidance as it provides hypothetical examples of AI patent claims intended to illustrate how the USPTO will analyze AI-related patent claims for sufficient subject matter eligibility. While it remains to be seen, these claim examples suggest that examiners will apply increased scrutiny to AI-related claims, where patent practitioners should begin to include (in any AI-related claims) elements regarding how the AI-related features operate and also include in the related patent specification one or more explanations of how the AI-related features improve the functioning of a computer or another technology or technical field (see examples 47 and 48), and/or an explanation of how a specific medical treatment is administered to a given patient (see example 49). Said another way, the new examples suggest that non-technical AI-related claims that simply recite an “AI model” with functional language of how to “apply it” (or equivalent non-operative language) will likely result in a Section 101 rejection from the USPTO.

It should be noted that the 2024 AI Guidance does not constitute new law. Rather it is merely guidance to apply existing law. However, the 2024 AI Guidance does note that its disclosure replaces any previous guidance, and that is expected the 2024 AI Guidance will be incorporated into the MPEP in due time. 89 FR 58131.

The below sections summarize the case law examples and claim elements, focusing on key aspects of the 2024 AI Guidance from the perspective of patentees seeking to protect AI-related inventions. 

Case Examples (2024 AI Guidance, Step 2A, Prong One: whether a Claim “recites” an Abstract Idea)

The 2024 AI Guidance provides several case examples from the United States Court of Appeals for the Federal Circuit (the Federal Circuit). The 2024 AI Guidance describes these cases as “informative” and divides the cases across the three categories of “abstract ideas” routinely cited by examiners for subject matter eligibility (Section 101) rejections, namely: (1) mathematical concepts, (2) certain methods of organizing human activity, and (3) mental processes.

It should be noted that none of these cases include claims that specifically address AI-related claims. Such cases have been rare. But practitioners can expect to see more AI-related cases emerge as AI-patents are asserted in the courts. See, e.g., PatentNext: How the Courts treat Artificial Intelligence (AI) Patent Inventions: Through the Years since Alice.

Nonetheless, each of the cases does relate to software-related inventions. Further, the example cases are intended to update the USPTO’s subject matter eligibility guidance under Step 2A, Prong One analysis, which regards whether a given patent claim “recites” an “abstract” idea. Under Step 2A, Prong One,  if a given claim does not recite an abstract idea, then the analysis concludes, and the claim is found patent eligible. 

The below subsections summarize these case examples provided for Step 2A, Prong One for each of the three categories of abstract ideas. 

Mathematical Concepts

The 2024 AI Guidance defines “mathematical concepts” as “mathematical relationships, mathematical formulas or equations, and mathematical calculations.” 89 FR 58135. A patent claim does not recite a mathematical concept, and thus is not “abstract,” if it is only based on or involves a mathematical concept. Id.

As an example case, the 2024 AI Guidance cites XY, LLC v. Trans Ova Genetics, 968 F.3d 1323, 1330-32 (Fed. Cir. 2020). In XY, the claims included a method of operating a flow cytometry apparatus to classify and sort particles into at least two populations in real time to more accurately classify similar particles. Id. The Federal Circuit determined that the claims did not recite a mathematical concept because the claimed method was not directed to “the abstract idea of using a `mathematical equation that permits rotating multi-dimensional data’ ” even though they may have involved mathematical concepts. Id. 

The case represents the importance of leaving mathematical formulas or equations out of a claim element. That is, even if an invention may use or rely on a newly discovered mathematical formula or equation, it is best practice to describe it in the specification and claim how a given system or method technically operates, even though such operation may involve the mathematical formula or equation. 

Certain Methods of Organizing Human Activity

The 2024 AI Guidance defines “certain methods of organizing human activity” as “concepts related to fundamental economic principles or practices (including hedging, insurance, mitigating risk); commercial or legal interactions (including agreements in the form of contracts, legal obligations, advertising, marketing or sales activities or behaviors, and business relations); and managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions).” 89 FR 58135 (emphasis added). 

The  2024 AI Guidance notes that the term “certain” qualifies this grouping of abstract ideas, and as a result, not all methods of organizing human activity are abstract ideas. Id. Further, “in addition, except in rare circumstances, this grouping should not be expanded beyond the activity within the enumerated sub-groupings of fundamental economic principles or practices, commercial or legal interactions, and managing personal behavior or relationships or interactions between people.” Id.

The 2024 AI Guidance provides three example Federal Circuit cases finding “certain methods of organizing human activity” in the respective claims:

  • Weisner v. Google LLC, 51 F.4th 1073 (Fed. Cir. 2022): The 2024 AI Guidance cites this case as an example of “managing personal behavior or relationships or interactions between people.” 89 FR 58135. In this case, the patent claims recited “collect[ing] information on a user’s movements and location history [and] electronically record[ing] that data” (i.e., “creating a digital travel log”). Id. (citing Weisner, 51 F.4th at 1082).
  • Elec. Commc’n Techs., LLC v. ShoppersChoice.com, LLC, 958 F.3d 1178 (Fed. Cir. 2020): The 2024 AI Guidance cites this case as an example of a “fundamental economic principle or practice.” 89 FR 58135. In this case, the patent claims recited “monitoring the location of a mobile thing and notifying a party in advance of arrival of that mobile thing.” Id. (citing Elec. Commc’n Techs, 958 F.3d at 1181). The Federal Circuit found that the claims amounted “to nothing more than the fundamental business practice of providing advance notification of the pickup or delivery of a mobile thing” and that “business practices designed to advise customers of the status of delivery of their goods have existed at least for several decades, if not longer.” Id.
  • Bozeman Fin. LLC v. Fed. Reserve Bank of Atlanta, 955 F.3d 971, 978 (Fed. Cir. 2020): The 2024 AI Guidance cites this case as an example of a “fundamental economic principle or practice.” 89 FR 58135. In this case, the patent claims recited methods for detecting fraud in financial transactions during a payment clearing process, including determining when there is a match between two financial records, sending a notification to a bank with authorization to process the financial transaction when there is a match, and sending a notification to a bank to not process the financial transaction when there is not a match. Id. (citing Bozeman Fin., 955 F.3d at 978). The Federal Circuit found that the claims were directed to the abstract idea of “collecting and analyzing information for financial transaction fraud or error detection.” Bozeman Fin., 955 F.3d at 981.

The cases illustrate that claims reciting technology at a functional and/or business method-related level are at risk of being found abstract.

Mental Processes

The 2024 AI Guidance defines “mental processes” as “thinking … that ‘can be performed in the human mind, or by a human using a pen and paper’.” 89 FR 58136. This includes “concepts performed in the human mind and explains that claims recite a mental process when they contain limitations that can practically be performed in the human mind, including, for example, observations, evaluations, judgments, and opinions.” Id.

The 2024 AI Guidance provides several case examples that include claims that do recite respective mental processes: 

  • Trinity Info Media, LLC v. Covalent, Inc., 72 F.4th 1355, 1362 (Fed. Cir. 2023): A claim to a method of “(1) receiving user information; (2) providing a polling question; (3) receiving and storing an answer; (4) comparing that answer to generate a `likelihood of match’ with other users; and (5) displaying certain user profiles based on that likelihood” was found directed to a “mental process” because it could practically be performed in the human mind (i.e., “[a] human mind could review people’s answers to questions and identify matches based on those answers”). 89 FR 58136.
  • In re Killian, 45 F.4th 1373, 1379 (Fed. Cir. 2022): A claim to “the collection of information from various sources (a Federal database, a State database, and a case worker) and understanding the meaning of that information (determining whether a person is receiving SSDI benefits and determining whether they are eligible for benefits under the law)” was found directed to a “mental process” because “[t]hese steps can be performed by a human, using observation, evaluation, judgment, [and] opinion, and because they involve making determinations and identifications, which are mental tasks humans routinely do.” Id.
  • PersonalWeb Techs. LLC v. Google LLC, 8 F.4th 1310, 1316-18 (Fed. Cir. 2021): Claims to “the use of an algorithm-generated content-based identifier to perform the claimed data-management functions,” which include limitations to “controlling access to data items,” “retrieving and delivering copies of data items,” and “marking copies of data items for deletion,” were found directed to a “mental process” because the claims cover “a medley of mental processes that, taken together, amount only to a multistep mental process,” where such steps can be practically performed in the human mind. Id.

The cases illustrate that claims reciting terminology (e.g., determining, observing, etc.) that can be arguably performed in the human mind are at risk of being found abstract.

In contrast, the 2024 AI Guidance describes that claims do not recite a mental process when they contain limitations that cannot practically be performed in the human mind, for instance when the human mind is not equipped to perform the claim limitations. Notably, concerning AI-related claims, the 2024 AI Guidance mentions that “claim limitations that only encompass AI in a way that cannot practically be performed in the human mind” do not recite a “mental process.”  The 2024 AI Guidance provides a single case example that includes a claim that does not recite a mental process: 

  • ADASA Inc. v. Avery Dennison Corp., 55 F.4th 900 (Fed. Cir. 2022): The example illustrates a claim that does not recite a mental process because it cannot be practically performed in the human mind. In particular, the claim-at-issue included: “a specific, hardware-based RFID serial number data structure” (i.e., an RFID transponder), where the data structure is uniquely encoded (i.e., there is “a unique correspondence between the data physically encoded on the [RFID transponder] with pre-authorized blocks of serial numbers”). 89 FR 58136 (citing ADASA, 55 F.4th 900 at 909).

This case illustrates that claims reciting technical steps and/or physical components that perform technical functions that the human mind cannot perform are better positioned for not being found abstract.

Case Examples (2024 AI Guidance, Step 2A, Prong Two: whether the Claim as a whole integrates the Abstract Idea Into a “Practical Application”)

The 2024 AI Guidance provides an overview of Step 2A, Prong Two, which provides an additional test if a claim is found abstract under Step 2A, Prong One. 89 FR 58136. The test requires examiners to consider the claim as a whole and effectively asks whether, even though the claim may recite an “abstract idea” (as determined per Step 2A, Prong One), the claim as a whole integrates the abstract idea into a “practical application.” Id.

The 2024 AI Guidance focuses on one type of “practical application,” namely whether “a claimed invention improves the functioning of a computer or improves another technology or technical field.” 89 FR 58137. Since the Supreme Court’s decision in Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014), U.S. Courts have routinely identified such an “improvement” as a sufficient “practical application” for demonstrating patent eligibility pursuant to Section 1. 

With respect to AI, the 2024 AI Guidance recognizes 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.” 89 FR 58137. 

And the 2024 AI Guidance cautions that “[a] key point of distinction to be made for AI inventions is between a claim that reflects an improvement to a computer or other technology described in the specification (which is eligible) and a claim in which the additional elements amount to no more than (1) a recitation of the words “apply it” (or an equivalent) or are no more than instructions to implement a judicial exception on a computer, or (2) a general linking of the use of a judicial exception to a particular technological environment or field of use (which is ineligible).” Id.

The 2024 AI Guidance describes that, in order to demonstrate an “improvement,” AI-related claims should indicate of how the AI-related invention is technically implemented:

“An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome.” [citing MPEP 2106.05(a)]. AI inventions may provide a particular way to achieve a desired outcome when they claim, for example, a specific application of AI to a particular technological field ( i.e., a particular solution to a problem). In these situations, the claim is not merely to the idea of a solution or outcome and amounts to more than merely “applying” the judicial exception or generally linking the judicial exception to a field of use or technological environment. In other words, the claim reflects an improvement in a computer or other technology.

Id. (emphasis added).

The 2024 AI Guidance provides a list of cases that demonstrate claimed “improvement” (and thus a “practical application”) for purposes of demonstrating sufficient subject matter eligibility pursuant to Section 101. These cases are presented below. 

  • ADASA Inc. v. Avery Dennison Corp., 55 F.4th 900 (Fed. Cir. 2022): This is the same case as provided above regarding “mental processes.” With respect to a claimed “improvement,” the claims included “specific, hardware-based RFID serial number data structure” (i.e., an RFID transponder), where the data structure is uniquely encoded ( i.e., there is “a unique correspondence between the data physically encoded on the [RFID transponder] with pre-authorized blocks of serial numbers”), such that it is “a hardware-based data structure focused on improvements to the technological process by which data is encoded. 89 FR 58137 (citing ADASA, 55 F.4th at 909). Further, with respect to an identified “improvement,” the Federal Circuit found that “[c]ritically, once a block [i.e., a sector of bits] is allocated to an encoder, there is no need to reconnect to a central database until the unique numbers within the block have been exhausted. … Thus, in the previous example, 224, or approximately 16.8 million, RFID tags could be commissioned before reconnection to a central database is required. And by eliminating the need for a continuous connection to the database, the attendant delays are reduced and the commissioning process is improved.” ADASA, 55 F.4th at 905 (emphasis added). Thus, when considered as a whole, and in view of the specification, the Federal Circuit found that the claim was not directed to an abstract idea. Id. at 908.
  • Cal. Inst. of Tech. v. Broadcom Ltd, 25 F.4th 976, 988 (Fed. Cir. 2022): The claims regarded performing error correction and detection encoding where information bits appeared in a variable number of subsets and were directed to an improvement of encoding data that relied in part on irregular repetition, and, thus, were not found directed to an abstract idea. 
  • Packet Intel. LLC v. NetScout Sys., Inc., 965 F.3d 1299, 1308-10 (Fed. Cir. 2020): The claims regarded a packet monitor to identify disjointed connection flows as belonging to a same conversational flow and were found comprising an improvement in computer technology, and, thus, were not found directed to an abstract idea.
  • Uniloc USA, Inc. v. LG Elec. USA, Inc., 957 F.3d 1303, 1305, 1307-08 (Fed. Cir. 2020): The claims regarded a primary station for use in a communication system, where an additional data field was added to enable the primary station to simultaneously send inquiry messages and poll parked secondary stations. Thus, the claims were found comprising an improvement in computer functionality, namely the reduction of latency experienced by parked secondary stations in communication systems and, thus, were not found directed to an abstract idea.
  • CardioNet, LLC v. InfoBionic, Inc., 955 F.3d 1358, 1368-69 (Fed. Cir. 2020): The claims regarded a cardiac monitoring device that analyzed the variability in the beat-to-beat timing for atrial fibrillation and atrial flutter to more accurately detect the occurrence of these cardiac conditions. Thus, the claims were found comprising an improvement in cardiac monitoring technology, and, thus, were not found directed to an abstract idea.
  • Koninklijke KPN N.V. v. Gemalto M2M GmbH, 942 F.3d 1143, 1150-51 (Fed. Cir. 2019): The claims regarded varying the way check data is generated by modifying the permutation applied to different data blocks. The claims were found comprising an improvement in a technological process for detecting systemic errors in data transmission and, thus, were not found directed to an abstract idea.

While none of these cases include claims that specifically address AI-related elements, they nonetheless describe software-related inventions (which AI-related elements fundamentally are) and thus illustrate how to claim and describe an improvement in a given patent application.

AI-related Claim Examples (Hypothetical Claims 47-49)

The 2024 AI Guidance provides three new examples of AI-related claims (i.e., new example claims 47-49). The 2024 AI Guidance titles these the “July 2024 Subject Matter Eligibility Examples 47-49” (the “July 2024 SME Examples”).  The examples provide hypothetical subject matter eligibility analyses under 35 U.S.C. 101.

The examples are intended to assist USPTO personnel and the public in understanding the proper application of the USPTO’s subject matter eligibility guidance in certain fact-specific situations, such as whether a claim recites an abstract idea or whether a claim integrates the abstract idea into a practical application because the claimed invention improves the functioning of a computer or another technology or technical field and thus is not “directed to” a given abstract idea. 

The 2024 AI Guidance reminds examiners that “it is not necessary for a claim under examination to mirror an example claim to be subject matter eligible.” July 2024 SME Examples at 1.

Important takeaways from the examples include: (1) if the invention can operate on specific hardware, be sure to recite such specific hardware in at least one claim; (2) do not simply recite generic AI technology or terminology (e.g., an “AI model”) followed by simple functional elements that amount to nothing more than “apply it” (or equivalent) because such a functionally claiming strategy will likely result in AI-related claim elements being found abstract;  (3) instead, describe, in at least one or two elements of the claim, how the AI-related elements technically operate (e.g., specific data used to train an AI model, specific steps of how the AI model is trained, and/or specific ways that the AI model’s output is used), and describe in the specification how such technical operate improves the functioning of a computer or provides improvements to another technology or technical field with respect to the prior art.

The examples come from different technical fields, each as summarized below.  

Claim Example 47 (“Anomaly Detection”)

Example 47 illustrates the application of the eligibility analysis to claims that recite limitations specific to AI, particularly the use of an artificial neural network (ANN) to identify or detect anomalies. According to the example, the use of specially trained ANNs to detect anomalies realizes a number of improvements over traditional methods of detecting anomalies, including more accurate detection of anomalies. The example further provides methods for training an ANN that lead to faster training times and a more accurate model for detecting anomalies.

Example 47 includes three hypothetical claims, each of which claim use of an ANN for anomaly detection. The three claims are reproduced below. 

Claim 1 (Eligible) Claim 2 (Not Eligible) Claim 3 (Eligible) 
[Claim 1] An application specific integrated circuit (ASIC) for an artificial neural network (ANN), the ASIC comprising: 
a plurality of neurons organized in an array, wherein each neuron comprises a register, a microprocessor, and at least one input; and 
a plurality of synaptic circuits, each synaptic circuit including a memory for storing a synaptic weight, wherein each neuron is connected to at least one other neuron via one of the plurality of synaptic circuits.
[Claim 2] A method of using an artificial neural network (ANN) comprising: 
(a) receiving, at a computer, continuous training data; 
(b) discretizing, by the computer, the continuous training data to generate input data; 
(c) training, by the computer, the ANN based on the input data and a selected training algorithm to generate a trained ANN, wherein the selected training algorithm includes a backpropagation algorithm and a gradient descent algorithm; 
(d) detecting one or more anomalies in a data set using the trained ANN; 
(e) analyzing the one or more detected anomalies using the trained ANN to generate anomaly data; and 
(f) outputting the anomaly data from the trained ANN. 
[Claim 3] A method of using an artificial neural network (ANN) to detect malicious network packets comprising:
(a) training, by a computer, the ANN based on input data and a selected training algorithm to generate a trained ANN, wherein the selected training algorithm includes a backpropagation algorithm and a gradient descent algorithm;
(b) detecting one or more anomalies in network traffic using the trained ANN;
(c) determining at least one detected anomaly is associated with one or more malicious network packets;
(d) detecting a source address associated with the one or more malicious network packets in real time;
(e) dropping the one or more malicious network packets in real time; and
(f) blocking future traffic from the source address. 

As summarized below, claims 1 and 3 are found patent eligible. Claim 2 is found to be ineligible

  • Claim 1 (Eligible): Claim 1 is patent eligible because it includes a plurality of neurons, which are specific hardware components comprising a register and a microprocessor, and a plurality of synaptic circuits which together form an ANN. July 2024 SME Examples at 5. The claim does not recite any abstract ideas, such as mathematical concepts, mental processes, or certain methods of organizing human activity. Id. While ANNs may be trained using mathematics, there is no mathematical formula recited in the claim. Id. Thus, the claim is eligible under Step 2A, Prong One (whether an abstract idea is “recited” by the claim). Thus, no analysis is required under Step 2A, Prong Two (regarding a “practical application”).
  • Claim 2 (Not Eligible): Claim 2 is not patent eligible because, under Step 2A, Prong One, it recites abstract ideas including, for example, mental concepts (e.g., rounding data values) that can be performed in the human mind, as well as mathematical concepts (e.g., a backpropagation algorithm and a gradient descent algorithm for training of the ANN). July 2024 SME Examples at 7. Further, Claim 2, even when considered as a whole under Step 2A, Prong Two, fails to include a “practical application” because Claim 2 recites generic computer hardware that simply recites the abstract ideas with the words “apply it” (or an equivalent), that amount to nothing more than mere instructions to implement an abstract idea on a computer without placing any limits on how such steps are performed. Id. at 8-9 (emphasis added). For example, even though Claim 2 includes AI-related elements such as “detecting one or more anomalies in a data set using the trained ANN” and “using a trained ANN,” such elements merely recite the outcome and fail to describe any details about how the elements are accomplished. Id. at 9 (emphasis added). Finally, Claim 2 is further not eligible under Step 2B because the Claim 2 elements amount to nothing more than well-understood, routine, and conventional activity in the field of computing. Id.
  • Claim 3 (Eligible): Claim 3 is patent eligible because it includes specific steps describing how the claimed ANN is used to detect and drop computer network packets. July 2024 SME Examples at 11. Even though under Step 2A, Prong One, Claim 3 recites abstract ideas (e.g., the mathematical concept of backpropagation and the mental process of “detecting” anomalies), under Step 2A, Prong Two, Claim 3 is nonetheless found eligible because it includes a “practical application” (i.e., an “improvement”), where the disclosed system detects network intrusions and takes real-time remedial actions, including dropping suspicious packets and blocking traffic from suspicious source addresses. Id. at 12. The example further explains that the disclosed system enhances security (i.e., provides an improvement) by acting in real time to proactively prevent network intrusions. Id. Thus, the claim as a whole integrates the abstract idea into a practical application (an “improvement”) such that Claim 2 is not abstract. Id.

Claim Example 48 (“Speech Separation”)

Example 48 illustrates the application of the eligibility analysis to claims that recite AI-based methods of analyzing speech signals and separating desired speech from extraneous or background speech. 

Example 48 includes three hypothetical claims, some of which include a deep neural network (DNN), and each of which seek to address the problem of detecting speech amid background noise, which can be experienced when a user attempts to use a smartphone, gaming console, or other electronic device for capturing voice commands in a noisy environment. 

Example 48 includes three example claims, as reproduced below. 

Claim 1 (Not Eligible) Claim 2 (Eligible) Claim 3 (Eligible) 
[Claim 1] A speech separation method comprising: 
(a) receiving a mixed speech signal x comprising speech from multiple different sources sn, where n ∈ {1, . . . N}
(b) converting the mixed speech signal x into a spectrogram in a time-frequency domain using a short time Fourier transform and obtaining feature representation X, wherein X corresponds to the spectrogram of the mixed speech signal x and temporal features extracted from the mixed speech signal x; and
(c) using a deep neural network (DNN) to determine embedding vectors V using the formula V = fθ(X), where fθ(X) is a global function of the mixed speech signal x
[Claim 2] The speech separation method of claim 1 further comprising: 
(d) partitioning the embedding vectors V into clusters corresponding to the different sources sn
(e) applying binary masks to the clusters to create masked clusters; 
(f) synthesizing speech waveforms from the masked clusters, wherein each speech waveform corresponds to a different source sn
(g) combining the speech waveforms to generate a mixed speech signal x’ by stitching together the speech waveforms corresponding to the different sources sn, excluding the speech waveformfrom a target source ss such that the mixed speech signal x’ includes speech waveforms from the different sources sn and excludes the speech waveform from the targetsource ss; and 
(h) transmitting the mixed speech signal x’ for storage to a remote location.
[Claim 3] A non-transitory computer-readable storage medium having computer-executable instructions stored thereon, which when executed by one or more processors, cause the one or more processors to perform operations comprising: 
(a) receiving a mixed speech signal x comprising speech from multiple different sources sn, where n ∈ {1, . . . N}, at a deep neural network (DNN) trained on source separation; 
(b) using the DNN to convert a time-frequency representation of the mixed speech signal x into embeddings in a feature space as a function of the mixed speech signal x
(c) clustering the embeddings using a k-means clustering algorithm; 
(d) applying binary masks to the clusters to obtain masked clusters; 
(e) converting the masked clusters into a time domain to obtain N separated speech signals corresponding to the different sources sn; and 
(f) extracting spectral features from a target source sd of the N separated speech signals and generating a sequence of words from the spectral features to produce a transcript of the speech signal corresponding to the target source sd

As summarized below, Claim 1 is found to be ineligible. Claims 1 and 3 are found patent eligible because they each integrate an abstract idea into a “practical application.”  

  • Claim 1 (Not Eligible): Claim 1 is not patent eligible because, under Step 2A, Prong One, it recites abstract ideas including, for example, mathematical concepts (e.g., “converting the mixed speech signal x into a spectrogram in a time-frequency domain using an [short-time Fourier transform] STFT, and obtaining feature representation X …”). July 2024 SME Examples at 19. Further, Claim 1, even when considered as a whole under Step 2A, Prong Two, fails to include a “practical application” because Claim 1 recites generic computer hardware and functionality that simply recites the abstract idea. Importantly, Claim 1 fails to include details about a particular DNN (e.g., how it is trained) or how the DNN operates to derive the embedding vectors other than that it is being used to determine the embedding vectors. Id. at 20. That is, The DNN is recited generally as applying the abstract idea (i.e., perform the mathematical calculation using the recited mathematical equation) without placing any limitation on how the DNN operates to derive the embedding vectors as a function of the input signal. While the specification includes this detail, as well as detail regarding how the invention offers an improvement over existing speech-separation, none of these features that result in the improvement are recited in Claim 1. Finally, Claim 1 is further not eligible under Step 2B because the Claim 1 elements amount to nothing more than well-understood, routine, and conventional activity in the field of computing technology. Id.
  • Claim 2 (Eligible): Claim 2 is patent eligible because it includes specific technical steps describing how the speech detection algorithm works. July 2024 SME Examples at 22-23. This is true despite the fact that Claim 2 does not explicitly recite an AI-related model (e.g., a DNN-related model). In particular, even though under Step 2A, Prong One, Claim 2 recites abstract ideas (e.g., mathematical concepts including specific formulas), under Step 2A, Prong Two, Claim 2 is nonetheless found eligible because it includes a “practical application” (i.e., an “improvement”) by reciting a particular speech-separation technique that solves the problem of separating speech from different speech sources belonging to the same class, while not requiring prior knowledge of the number of speakers or speaker-specific training. Id. at 24. The claim reflects the improvement discussed in the example disclosure by reciting details of how the DNN aids in the cluster assignments to correspond to the sources identified in the mixed speech signal, which are then synthesized into separate speech waveforms in the time domain and converted into a mixed speech signal, excluding audio from the undesired source. Id. Importantly, the claim features align with the improvement described in the specification. Id. Thus, the claim as a whole integrates the abstract idea (e.g., a mathematical concept) into a practical application (an “improvement) such that Claim 2 is not abstract. Id.
  • Claim 3 (Eligible): Claim 3 is patent eligible because it includes specific technical steps describing how the speech detection algorithm works, and places specific limits on the otherwise generic AI-related claims. July 2024 SME Examples at 26-27. In particular, even though under Step 2A, Prong One, Claim 3 recites abstract ideas (e.g., certain claim elements include mathematical concepts), under Step 2A, Prong Two, Claim 3 is nonetheless found eligible because it includes other elements (e.g., elements (e) and (f)) that do not include any mathematical formula, and thus, fall outside of the “mathematical concept” grouping of abstract ideas (as well as the mental process and certain methods of human activity groupings). Id. at 27. Further, even though a DNN is recited at a high level of generality, Claim 2 nonetheless includes a “practical application” (i.e., an “improvement”) because Claim 2 includes details of how the DNN trained on source separation aids in the cluster assignments to correspond to the sources identified in the mixed speech signal, which are then converted into separate speech signals in the time domain to generate a sequence of words from the spectral features, thereby making individual transcription of each separated speech signal possible. Id. at 28. Such features are also described in the detailed description as an improvement in the field of speech-to-text technology. Thus, the claim as a whole integrates the abstract idea (e.g., a mathematical concept) into a practical application (an “improvement) such that Claim 2 is not abstract. Id.

Claim Example 49 (“Fibrosis Treatment”)

Example 49 illustrates analysis of method claims reciting an AI model that is designed to assist in personalizing medical treatment to the individual characteristics of a particular patient.

In the hypothetical example 49, the patent applicant developed a new anti-fibrotic drug, “Compound X,” that effectively reduces scarring around a microstent implantation site in glaucoma patients at high risk of “post-implantation inflammation” (PI”) after microstent implant surgery without the undesirable side effects of a known “drug A.” 

The patent disclosure also includes a computer-implemented machine learning model (referred to as “the ezAI model”) and its clinical applications. Given an input of a patient’s genotype dataset, the ezAI model calculates a weighted polygenic risk score (PRS) from informative single-nucleotide polymorphisms (SNPs) in the dataset—using multiplication to weight corresponding alleles in the dataset by their effect sizes and addition to sum the weighted values. Using the same weights and informative SNPs, the ezAI model improves upon the base PRS model by determining a risk score and providing a classification in less time. 

Example 49 includes two example claims, as reproduced below, where Claim 2 depends from 

Claim 1 (Not Eligible) Claim 2 (Eligible) 
[Claim 1] A post-surgical fibrosis treatment method comprising: 
(a) collecting and genotyping a sample from a glaucoma patient to a provide a genotype dataset; 
(b) identifying the glaucoma patient as at high risk of post-implantation inflammation (PI) based on a weighted polygenic risk score that is generated from informative single-nucleotide polymorphisms (SNPs) in the genotype dataset by an ezAI model that uses multiplication to weight corresponding alleles in the dataset by their effect sizes and addition to sum the weighted values to provide the score; and 
(c) administering an appropriate treatment to the glaucoma patient at high risk of PI after microstent implant surgery.
[Claim 2] The method of claim 1, wherein the appropriate treatment is Compound X eye drops.

As summarized below, Claim 1 is not patent eligible. Claim 2 is patent eligible because it integrates the abstract idea of Claim 1 into a “practical application.” 

  • Claim 1 (Not Eligible): Claim 1 is not patent eligible because, under Step 2A, Prong One, it recites abstract ideas including, for example, a mental process (e.g., “identifying the glaucoma patient as at high risk of PI based on a weighted PRS …”) and a mathematical concept (e.g., “a weighted PRS that is generated from informative SNPs in the genotype dataset by an ezAI model that uses multiplication to weight corresponding alleles in the dataset by their effect sizes and addition to sum the weighted values to provide the score”). July 2024 SME Examples at 31. Further, Claim 1, even when considered as a whole under Step 2A, Prong Two, fails to include a “practical application” because Claim 1 recites generic computer hardware and functionality that simply recites the abstract ideas at a high level, e.g., “collecting …” and “administering ….” Importantly, with respect to the AI-related claim elements, both the specification and Claim 1 fail to include details about the ezAI model with respect to an improvement to the functioning of a computer or any other technology. Id. at 33. Rather, the specification, at best, simply describes an improvement to the determination of a patient risk score and thus is considered an improvement to the abstract idea itself of improving risk scores, which is insufficient to demonstrate a “practical application.” Id. Finally, Claim 1 is further not eligible under Step 2B because the Claim 1 elements amount to nothing more than well-understood, routine, and conventional activity in the given field. Id.
  • Claim 2 (Eligible): Claim 2 is patent eligible because it includes the additional element of “the appropriate treatment is Compound X eye drops.” July 2024 SME Examples at 34. In particular, even though under Step 2A, Prong One, dependent Claim 2 recites the abstract ideas of Claim 1 (e.g., mental processes and mathematical concepts), under Step 2A, Prong Two, Claim 2 is nonetheless eligible because claim elements, when considered as a whole, recite identification of a patient as belonging to a specific patient population (glaucoma patients at high risk of PI), and where the patient is then administered a treatment (Compound X eye drops instead of any common anti-fibrotic treatment, such as drug A, after microstent implant surgery) that is particular to that specific patient population (glaucoma patients at high risk of PI). Id. at 35. Relying on the determination of patient risk to administer Compound X eye drops to glaucoma patients at high risk of PI after microstent implant surgery is, therefore, a particular treatment for a medical condition such that the claim as a whole integrates the judicial exception into a practical application. Thus, the claim as a whole integrates the abstract ideas (e.g., mental processes and mathematical concepts) into a practical application (e.g., a particular medical treatment) such that Claim 2 is not abstract. Id.

Taken together, new claim examples 47-49 suggest that examiners will apply increased scrutiny to AI-related claims. Patent practitioners should now include (in any AI-related claims) elements regarding how the AI-related features operate and also include in the related patent specification one or more explanations of how the AI-related features improve the functioning of a computer or another technology or technical field (see examples 47 and 48), and/or an explanation of how a specific medical treatment is administered to a given patient (see example 49). Said another way, the new claim examples suggest that non-technical AI-related claims that simply recite an “AI model” with functional language of how to “apply it” (or equivalent non-operative language) will likely result in a Section 101 rejection from the USPTO.

****

Subscribe to get updates to this post or to receive future posts from PatentNext. Start a discussion or reach out to the author, Ryan Phelan, at rphelan@marshallip.com (Tel: 312-474-6607). Connect with or follow Ryan on LinkedIn.