On September 22, 2022, the United States Patent and Trademark Office (USPTO) reminded patent practitioners of current case law and sections of the Manual of Patent Examining Procedure (MPEP) as patent practitioners continue to work in the field of artificial intelligence (AI). technology room. A summary of these reminders (and links to more information) can be found here.
MPEP Sections You Should Know – Especially for AI Inventions
A few key areas for applications targeting AI inventions are (1) eligibility of patent subject matter and (2) enabling disclosure. The highlighted sections contain MPEP 2106, MPEP 2181and MPEP 2173.05(g).
MPEP 2106 provides general guidance on the suitability of subject matter, including the definition that the claimed invention must belong to one of the four statutory categories, the claimed invention must also be considered patentable subject matter (e.g. the claim must not seek a court exception, unless the claim as a whole contains additional limitations that go significantly beyond the exception). The section also includes a flow chart that explains how examiners must analyze claims to determine whether they are directed to patentable subject matter (under Step 1, Step 2A Part One, Step 2A Part Two, and Step 2B), and also provides a handful including helpful examples of valid and invalid claims. For example, specifically for AI, MPEP 2106.03 describes products that have no physical or tangible form, such as information (often referred to as “data per se”) or a computer program per se (often referred to as “software per se”). se”) when claimed as a product without structural naming, not directed to any of the legal categories and therefore not naming patentable subject matter.
MPEP 2181 provides general guidance on checking the limitations of means-plus-function (35 USC 112(f)) to avoid an indefinite claim. In these cases, the means plus function analysis can be invoked by using the terms “means” or “step” in a claim, or “general placeholder” terms are used in place of “means” or “step”. To overcome this interpretation, applicant may (1) provide sufficient evidence that the claim limitation specifies sufficient structure to perform the claimed function to avoid interpretation under Section 112(f); or (2) amend the claim limitation to avoid interpretation under Section 112(f) (e.g., by specifying sufficient structure to perform the claimed function).
MPEP 2173.05(g) discusses functional limitations that do not rely on 35 USC 112(f) but may still render the claims vague. In this section, the claim identifies a feature “by what it does, not by what it is” (e.g., as evidenced by its specific structure or ingredients). Unlike means-plus-function assertion language, which only applies to purely functional constraints, functional assertion often involves the recitation of a structure followed by its function. For example, the claim may recite a conical spout (the structure) that “allows[ed] several grains of popped popcorn pass at the same time” (the function). Applicant should be careful not to formulate claims that are vague to indicate functional limitations.
Key PTAB and USPTO petition decisions related to AI
At the time of the meeting, two decisions provide a good picture of today’s AI patent landscape. The decisions are: Ex parte Hannun (formerly Ex parte Linden), 2018-003323 (April 1, 2019)which applies the 2019 Patent Eligible Guidance (PEG) to a method for “enhancing speech-to-text transcription”, and Regarding appl. No. 16/524,350 (“DABUS”), according to which invention must be limited to natural persons and not to AI/machines. More details on each case are below.
in the Ex parte Hannunthe patent at issue claimed a system and method for improving a speech-to-text transcription. The PTO examiner rejected the claims because they allegedly did not address patentable subject matter. For example, the PTO Examiner claimed that the claims only related to “mathematical relationship/formula” and also “certain methods of organizing human activities” while also claiming “since humans can listen to an audio file and transcribe the audio data into text data, all of which can be done in the head.” The applicant appealed and the board agreed that the claim was directed to patentable subject-matter on several grounds. For example, as part of the analysis of Step 2A Prong One, the Board asserted that “[w]Although transcription can generally be performed by a human, the claims here are directed to a specific implementation, including the steps of normalizing an input file, generating a jitter set of audio files…” The Board also asserted that “…the claims no basic economic principles or practices, commercial or legal interactions, the management of personal behavior or relationships or interactions between people…” so they do not aim at the abstract idea of ”particular methods of organizing human activities”. In addition, the board noted that the “claims recite using predicted character probabilities to decide on a transcription of input audio material which the examiner, relying on the specification, determines is a mathematical formula. Namely, the examiner notes that the specification discloses an algorithm for obtaining the predicted character probabilities… However, the mathematical algorithm or formula is not cited in the claims.” Although the analysis could have stopped at Step 2A Prong One, the Chamber also states that “the claims of the current application contain specific features specifically designed to achieve an improved technological result” and “provide improvements in this technical area” (under step 2A prong two). The board also addressed the rejection of the PTO examiner in Step 2B. For example, the PTO examiner concluded that the claims did not contain “additional elements that amounted to much more than a judicial exception” but did not provide sufficient factual support, thereby addressing a further inefficiency of the denial.
in the Regarding appl. No. 16/524,350 (“DABUS”)the applicant attempted to claim a machine as the inventor of a patent application. For example, in the Application Data Sheet (ADS), a single inventor was named “DABUS” as a first name and “(Invention generated by artificial intelligence)” as a family name. The legal successor was listed as “Stephen L. Thaler” (legal representative of DABUS). The “Inventor Statement” states that the invention was conceived by a “creativity machine” called “DABUS” and should be named as the inventor in the ‘350 application. The USPTO issued an initial notice for filing missing portions of a nonprovisional application, noting that the ADS “does not identify each inventor by his legal name” and an $80 surcharge for late filing of the inventor’s oath or declaration he lifted. The applicant filed a petition under 37 CFR 1.181 requesting a regulatory review of the notice and setting aside the notice for unfairness and/or invalidity. The USPTO then issued a second notice filing missing portions of the nonprovisional application and denied applicant’s petition. The applicant submitted a request for review of the decision to reject the applicant’s application. The USTPO ruled that the “inventor” must be a natural person and cannot include machines (e.g., neural networks) by reviewing case law, sections of the USC, and various USPTO rules cited in the MPEP, all of which relate to refer to natural persons and pronouns.
Examiner training on professional suitability for AI
The USPTO has published various training materials for examiners to examine AI inventions. The examiner training materials are available here. In particular, these training materials cover the current guidance on patent subject suitability (e.g. PEG guidance pre-2019, PEG guidance post-2019, nature-based products and life sciences, and a selection of court decisions).
Particular attention is paid to this 2019 PEG example 39, which analyzes a patent claim directed to a “method for training a neural network for face recognition” and confirms that the claim recites patentable subject matter. For example, the claim’s preamble recited a “computer-implemented method for training a neural network for face recognition.” The claim elements included:
- collecting a set of digital facial images from a database;
- applying one or more transformations to each digital face image, including mirroring, rotating, smoothing, or contrast reduction, to create a modified set of digital face images;
- creating a first training set that includes the collected set of digital facial images, the modified set of digital facial images, and a set of digital non-facial images;
- training the neural network in a first stage using the first training set;
- creating a second training set for a second training level, comprising the first training set and digital non-facial images that are incorrectly recognized as facial images after the first training level; and
- Training the neural network in a second stage using the second training set.
Under Step 2A Prong One, the USPTO asserted that the claim does not list any of the court exceptions listed in the 2019 PEG, so the claim is directed to patentable subject matter. Specifically, the 2019 PEG asserts that the claim does not list any mathematical relationships, formulas, or calculations. While some of the limitations may be based on mathematical concepts, the claims do not cite the mathematical concepts. Further, the claim does not recite any mental process since the steps are not practically performed in the human mind. Finally, the claim does not mention a method for organizing human activity, such as a basic economic concept or the management of interactions between people. Thus, the claim is eligible because it does not cite a court exception and the analysis does not proceed to Step 2A Prong Two or Step 2B.