_edited.jpg)
Chris Emerson
This article was developed by the Kentucky Bar Association (KBA) Intellectual Property (IP) Section. To learn more about the KBA IP Section, visit https://www.kybar.org/page/intellectualproperty.
Roadmap
The United States Patent and Trademark Office (USPTO) issued a Guidance Update on subject matter eligibility for artificial intelligence (AI) inventions.
The USPTO Guidance may indicate how patent attorneys can draft patent applications to avoid subject matter eligibility rejections and may indicate how examiners may evaluate patent applications in determining whether subject matter eligibility rejections are appropriate.
This article will discuss the Guidance Update, and specifically, the examples included therein.
What is Subject Matter Eligibility?
Subject matter eligibility is a determination (in view of statute 35 U.S.C. 101) of whether an invention that is disclosed in a patent application is new and useful and falls into one of the following categories of process, machine, manufacture, and composition of matter.[1] Subject matter eligibility is largely guided by statute, case law, and a USPTO Examiner’s judgement.
Other patent application concerns, such as novelty[2] (35 U.S.C. 102) and obviousness [3] (35 U.S.C. 103) are largely guided by “prior art” that a USPTO Examiner searches for during the application examination phase.
AI subject matter eligibility should not be confused with AI inventorship. While inventions that use AI may be eligible for patent protection, inventions made (at least solely) by AI are not eligible for patent protection. [4]
What is the current status of AI Subject Matter Eligibility?
Currently, inventions using AI may be eligible for patent protection. Inventions using AI are subjected to the same subject matter eligibility test as any other invention. A flowchart of the test excerpted from the Manual of Patent Examination Procedure (MPEP) 2106, may be found at the end of this article.
The first step of the test asks whether a patent application claim (e.g., not necessarily the invention itself) is directed to a process, machine, manufacture, or a composition of matter.
Many claims for AI-focused applications may satisfy the first step of the test by being a claim to a process and/or machine.
The second step of the test asks whether the claim is directed to “judicial exceptions,” including a law of nature, a natural phenomenon (product of nature), and/or an abstract idea.
If the claim is directed to a judicial exception, then it may not be eligible for a patent unless the claim satisfies third step (subsequently discussed).
Many AI-focused claims may run into issues at the second step, for example, because they are directed to an abstract idea such as mathematical equations, organizing human activity, or mental processes.
The third step of the test is a determination of whether the claim “recite[s] additional elements that amount to significantly more than the [abstract idea].” If the claim does recite the additional elements, then it is eligible. If the claim does not recite the additional elements, then it is not eligible.
These tests are simple enough on their face but leave a lot of room for individual (and arguably arbitrary) decisions by the Examiner of whether terms like an “abstract idea” and “additional elements that amount to significantly more” apply to a pending claim. Frequently, Applicants and Examiners have opposing interpretations of these terms and their application to a pending claim.
USPTO examples of AI cases (which Examiners are encouraged to consider) provide guidance to Applicants and Examiners on the applicability of a subject matter eligibility (35 U.S.C. § 101) based rejection to a pending claim.
What examples of AI Subject Matter Eligibility does the 2024 Guidance provide?
The 2024 Guidance identifies three new examples 47-49 specifically “intended to assist examiners in applying the USPTO’s subject matter eligibility guidance to AI inventions during the patent examination process.”[5]
The 2024 Guidance acknowledges that “USPTO personnel must draw a distinction between a claim that ‘recites’ an abstract idea (and thus requires further eligibility analysis) and one that merely involves, or is based on, an abstract idea” [6] and that “many claims to AI inventions are eligible as improvements to the functioning of a computer to another technology or technical field.” [7]
Regarding examples 47-49 specifically, which were “developed [as] examples for AI inventions,”[8] the 2024 Guidance indicates that recitation of an AI element in a claim, alone and in itself, does not cause the claim to fall under the judicial exception.
Example 47
In example 47, an application-specific integrated circuit (ASIC) for an artificial neural network (ANN) is claimed. Independent claim 1 also specifies inclusion of a plurality of neurons synaptic circuits. Dependent claim 2 specifies a method to be performed relating to retrieval and processing of first data using the ANN, and that is used to generate second data. Dependent claim 3 specifies a method to train an ANN, detect anomalies using the ANN, and perform an action based on detection of the anomalies.[9]
Independent claim 1 was found eligible based on the claimed limitations being directed to hardware and structural features.[10] This is in contrast to software and method-based claims, which sometimes recite limitations that could arguably be performed by a human mind (and thus fall into the realm of an “abstract idea” and incur a 101 rejection).
Dependent claim 2 was found ineligible based on the claimed limitations being directed to a method of data retrieval and processing which could have been performed by a human mind.[11]
Put another way, dependent claim 2 was found to be directed to a judicial exception to patentability—e.g., as being performable by a human, and being an abstract idea.
Additional limitations in dependent claim 2 were found to be directed to limiting the method to generic computer hardware, and limitations necessary for generic data retrieval and processing, and therefore was found to not amount to significantly more than the judicial exception.
Dependent claim 3 included limitations directed to the abstract idea of mental grouping, such as determining anomalies and non-anomalies.[12] However, dependent claim 3 included additional limitations of identifying addresses from which the anomalies were being sent, dropping packets associated with the anomalies, and blocking the address, while also specifying that these limitations were performed by a computer. These additional features were found to amount to significantly more than the judicial exception.
In summary, Example 47 indicates that structural limitations (in contrast with methodical limitations) may render a claim eligible in view of the second step of the subject matter eligibility test. Example 47 also indicates that data-based actions amounting to more than just retrieval and processing of data (e.g., identifying specific addresses, blocking the addresses, and dropping data from the addresses) may render a claim eligible in view of the third step of the subject matter eligibility test.
Example 48
In example 48, a speech separation method is claimed. Independent claim 1 includes limitations for receiving signals, converting signals using a mathematical formula, and processing the signals using an AI model. Dependent claim 2 includes limitations relating to categorizing features of the signals, processing features of the signals, generating a specific signal based on the signals, and transmitting the generated signal. Dependent claim 3 includes limitations relating to receiving the transmitted signal, processing and categorizing features of the transmitted signal, and extracting features of the transmitted signal to obtain a transcript representation of the transmitted signal.
Independent claim 1 was found ineligible as being directed to mathematical concepts (abstract ideas).[13] The claim limitations recited, and relied primarily on, verbatim mathematical equations and general mathematical operations (e.g., Fourier Transforms) to receive a mixed speech signal and determine embedding vectors corresponding to the mixed speech signal.
Dependent claim 2 was found eligible, and relied on those same mathematical equations, and categorization and processing thereof.[14] However, dependent claim 2 recited details about how the AI model separated speech signals from the mixed speech signal, processed and combined them, and generated a new mixed speech signal, which amounted to significantly more than performing the mathematical operations and performing an abstract idea.
Dependent claim 3 was found eligible and included limitations relating to mathematical operations and computational hardware.[15] However, claim 3 recited details about extraction of features of a target signal (e.g., an individual speaker) from the improved mixed speech signal (e.g., of a plurality of speakers, including the speaker), and generating a transcript of the target signal.
In summary, Example 48 indicates that mathematically based actions amounting to more than recitation of formulas and equations used in processing data (e.g., generating something new based on the mathematical operations) may render a claim eligible in view of the third step of the subject matter eligibility test.
Example 49
In example, 49 a post-surgical method of treatment is claimed. Independent claim 1 specified receiving a sample from a patient, processing the sample using a machine learning model, and administering a generic treatment. Dependent claim 2 specified the treatment.
Independent claim 1 was found ineligible and included abstract ideas in the form of mental processes by genotyping (categorizing) the patient’s sample into a particular dataset using an AI model.[16] Additionally, the generic recitation of applying a treatment was found to not be meaningful, as it didn’t provide for a particular application of the abstract idea of genotyping the sample.
Dependent claim 2 was found eligible and included the singular limitation of defining the treatment as a particular treatment (“Eyedrops X”).[17] Defining the treatment as a particular treatment in view of the patient’s genotype was indicated as amounting to significantly more than an abstract idea of mental processes associated with processing and categorizing the data using AI.
In summary, Example 49 indicates that claiming granular details of actions may render a claim eligible under the third step of the subject matter eligibility test.
Example 49 may be the most perplexing from a practitioner standpoint. Claiming a particular limitation (especially one seemingly included under the umbrella of a broader independent claim limitation) is traditionally used as a tactic to address novelty and obviousness considerations (e.g., overcome and/or avoid prior-art based rejections). For example, “administering treatments” in general may be known, but “administering treatment X” may not be. Example 49 indicates that the same limitations may be used to create fallback positions for both prior art purposes and for subject matter eligibility purposes.
What impact does the 2024 Guidance Update have on Subject Matter Eligibility?
The 2024 Guidance Update is not binding on federal courts, like the Court of Appeals for the Federal Circuit, which frequently hears patent cases, or on administrative courts like the Patent Trial and Appeal Board, which usually hears patent prosecution appeals. Put another way, “[the 2024] guidance does not constitute substantive rulemaking and does not have the force and effect of law.”[18] Rather, the Guidance “will assist USPTO personnel and stakeholders in evaluating the subject matter eligibility of claims in patent applications and patents involving inventions related to AI technology (AI inventions).”[19]
“[The 2024] guidance update, together with the guidance provided by the Manual of Patent Examining Procedure (MPEP), is to be used by USPTO personnel when applying subject matter eligibility law.”[20] “[I]f any earlier guidance from the USPTO, including any section of the current MPEP, is inconsistent with guidance set forth in this notice, USPTO personal are to follow this guidance [and that] this guidance update will be incorporated into the MPEP in due course.”[21]
Despite the Guidance not being binding, a report found that the 2019 revisions to Subject Matter Eligibility Guidance results in a 25% decrease in technologies, including AI, receiving a first office action with a rejection for ineligible subject matter, and that uncertainty about determinations of subject matter eligibility for relevant technologies decreased by 44% compared to the previous year.[22]
Conclusion
The 2024 Guidance, particularly Examples 47 and 48, provide useful indicators for practitioners to rely on while drafting applications and responding to office actions.
If you have questions about patent prosecution, litigation, and/or portfolio management, please consider reaching out to a local patent practitioner. Kentucky residents have access to a host of unique resources and assistance in their state.
[1] 35 U.S.C. § 101.
[2] 35 U.S.C. § 102.
[3] 35 U.S.C. § 103.
[4] 2024 Guidance Update on Pat. Subject Matter Eligibility, Including on Artificial Intelligence, 89 Fed. Reg. 58,128, 53,138 (Jul. 17, 2024) (“2024 Guidance”). The 2024 Guidance elaborates that “AI-assisted inventions are not categorically unpatentable. Patent protection may be sought for AI-assisted inventions where one or more persons made a significant contribution to the claimed invention.
[5] 2024 Guidance, 89 Fed. Reg. at 58,131.
[6] Id. at 58,134.
[7] Id. at 58,128.
[8] Id. as 58,138.
[9] July 2024 Subject Matter Eligibility Examples, U.S. Pat. and Trademark Off., 2 (Jul. 2024) https://www.uspto.gov/sites/default/files/documents/2024-AI-SMEUpdateExamples47-49.pdf.
[10] Id. at 5.
[11] Id. at 6.
[12] Id. at 10.
[13] Id. at 18.
[14] Id. at 21.
[15] Id. at 25.
[16] Id. at 31.
[17] Id. at 34.
[18] 2024 Guidance, 89 Fed. Reg. at 58,131.
[19]Id. at 58,128.
[20] Id.
[21]Id. at 58,135.
[22]Id. at 58,130 (citing Off. of the Chief Econ., Adjusting to Alice, USPTO Patent Examination Outcomes after Alice Corp. v. CLS Bank International, U.S. Pat. and Trademark Off. (April 2020) https://uspto.gov/sites/default/files/documents/OCE-DH_AdjustingtoAlice.pdf.