USPTO Examples for AI Invention Claims Must Play Out in Practice

The US Patent and Trademark Office’s July 17 patent eligibility update targets patent examiners and stakeholders when evaluating the eligibility of AI-related patent claims under 35 U.S.C. § 101. The latest guidance incorporates feedback from stakeholders and recent Federal Circuit decisions. Whether it is a useful guide to drafting eligible claims will likely be in the eye of the beholder.

The update focuses on the two critical prongs of the now well-known Alice/Mayo test, namely determining whether a claim recites an abstract idea, then evaluating whether the claim integrates the judicial exception into a “practical” application.

Three new artificial intelligence-related examples illustrate the USPTO’s latest application of its eligibility analysis, ignoring whether AI was used in developing the claimed invention. They highlight what the USPTO believes to be an eligible versus ineligible AI claim.

Example 47

“Artificial Neural Network for Anomaly Detection,” Example 47, analyzes claims that recite the use of an artificial neural network to detect anomalies. The example presents three claims. Claim 1 is directed to an application-specific integrated circuit for an ANN. It is found to be eligible, as it is directed to a physical device with specific hardware components rather than a mere abstract idea.

Claim 2, however, recites a method claim for discretizing continuous training data via an ANN by receiving training data, processing it through the neural network, and detecting and outputting anomalies. While the claim is found to recite a mental process, it’s eligible because it integrates the abstract idea into a practical application in a “particular machine” (the ANN).

Claim 3 is similar to Claim 2 but adds more details about the ANN and is thus eligible because it is said to integrate the abstract idea into a practical application, reflecting an improvement in the functioning of a computer or an improvement to other technology or technical field.

Example 47 illustrates how AI-related claims can be eligible when they involve specific hardware implementations or when they apply abstract ideas using particular machines or techniques that improve computer functionality or other technology.

Example 48

“AI-Based Speech Signal Processing,” Example 48, analyzes claims that recite AI-based methods of analyzing speech signals and separating desired speech from extraneous or background speech by receiving a speech signal, extracting features, applying a trained machine learning model to separate the speech signal into components, and outputting a desired speech component.

The example cites the separation of speech components as a “mental process” not integrated into a practical application because the use of a “generic” machine learning model to perform the abstract idea amounts to mere instructions to apply an exception using a generic computer component. Thus, the claim is ineligible.

Claim 2 builds on Claim 1 by adding specific details about the structure and training of the machine learning model, including the use of a particular neural network architecture and training technique. This claim is found to be eligible because it integrates the abstract idea into a practical application, thus improving speech signal processing technology.

Claim 3 is similar to Claim 2 but focuses on use of the trained model rather than its training process. This claim is also found to be eligible as integrating the abstract idea into a practical application. The example states that Claim 3 recites a specific implementation of the trained model that improves the technology of speech signal processing.

Example 48 illustrates how adding specific technical details about the structure, training, or implementation of AI models can transform an otherwise ineligible claim into an eligible one by demonstrating a practical application that improves technology. But the reasoning for Claim 1’s “mental process” ground of ineligibility is likely open to debate among technologists, with some wondering how performing feature extraction on speech signals to train a machine learning model can be performed in the human mind.

Example 49

“AI-Assisted Personalized Medical Treatment,” Example 49, analyzes claims reciting an AI model designed to assist in personalizing medical treatment based on the individual characteristics of a particular patient. Claim 1 is a method for generating and administering a treatment plan by receiving patient data, processing it through a machine learning model, and generating and administering a treatment plan based on the model’s output.

This claim is likewise found to recite a “mental process” (generating a treatment plan) not integrated into a practical application because the use of a generic machine learning model amounts to mere instructions to apply the “mental process” exception using a “generic” computer component. Claim 2 builds on Claim 1 by adding details about the treatment involving a particular compound, and is found to be eligible because it integrates the abstract idea into a practical application via the addition of a specific treatment modality.

Example 49 demonstrates how AI-assisted medical treatment methods might be eligible only when they involve specific technical implementations of AI models in practical applications that have “meaningful limits”—e.g., the particular treatment via Claim 2’s reciting a specific compound.

Thus, while the new examples might help to frame one’s claims to AI inventions, they are far from a definitive roadmap to drafting an eligible AI-related claim. As the examples illustrate, claims that describe a specific application of AI to a particular technological field or a specific way to achieve a desired outcome are more likely to be considered eligible.

The USPTO wants to see claims that on their face improve existing technological processes. While such improvements can and should be explained in the patent’s specification, reciting the technical details that lead to this improvement in the claims can be critical in demonstrating that the claim isn’t merely an abstract idea, but a practical application that enhances technology.

The proof of the latest examples’ efficacy will be whether the USPTO actually allows claims that mimic those in the examples.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.

Author Information

Nicholas Martin is shareholder at Greenberg Traurig.

James DeCarlo is shareholder at Greenberg Traurig.

Write for Us: Author Guidelines