ANTHONY J. WENN

Senior Associate

Navigating AI Innovation: USPTO's Updated Subject Matter Eligibility Guidance

August 1, 2024

In a significant move to foster innovation in emerging technologies, the U.S. Patent and Trademark Office (USPTO) has issued an updated guidance on patent subject matter eligibility, specifically targeting advancements in artificial intelligence (AI). This latest update, effective from July 17, 2024, aims to provide greater clarity and consistency in evaluating the subject matter eligibility of AI-related inventions under patent law (35 § U.S.C. 101).

Key Highlights of the Guidance Update:

Kathi Vidal, Under Secretary of Commerce for Intellectual Property and Director of the USPTO, emphasized the agency’s dedication to nurturing innovation in AI and other critical technologies, stating “The USPTO remains committed to fostering and protecting innovation in critical and emerging technologies, including AI.”

To aid in the application of this guidance across a diverse array of technologies, the USPTO has introduced three new examples, including patentable subject matter. These examples provide hypothetical claims analyses, addressing specific questions such as whether a claim constitutes an abstract idea or if it integrates an abstract idea into a practical application.

UPSTO Example 47: Anomaly Detection

Background: The invention involves the use of an artificial neural network (ANN) to detect anomalies, offering improvements over traditional methods by leveraging machine learning techniques. The ANN can be implemented in various forms, including software, hardware, or a combination of both.

Claims:

Claim 1 describes an application-specific integrated circuit (ASIC) for an artificial neural network (ANN). The ASIC includes a plurality of neurons organized in an array, with each neuron comprising a register, a microprocessor, and at least one input. It also includes a plurality of synaptic circuits, each with a memory for storing a synaptic weight, where neurons are connected via these synaptic circuits. This claim is eligible because it falls within a statutory category (machine) and does not recite any judicial exceptions. The claim specifies a physical structure (ASIC) with concrete elements (neurons, synaptic circuits) and operations, making it a tangible, non-abstract invention.

Claim 2 describes a method of using an artificial neural network (ANN) comprising several steps: receiving continuous training data, discretizing the continuous training data to generate input data, training the ANN using input data and selected training algorithms (backpropagation and gradient descent), detecting anomalies in a data set using the trained ANN, analyzing the detected anomalies to generate anomaly data, and outputting the anomaly data from the trained ANN. This claim is ineligible because it recites a judicial exception (abstract idea) and does not integrate the exception into a practical application. The steps involve mental processes and mathematical calculations, which are considered abstract. The claim as a whole lacks additional elements or a specific technological improvement that would render the abstract idea practical and concrete.

Claim 3 describes a method using an ANN to detect malicious network packets. The method includes training the ANN with input data using specific algorithms, detecting anomalies in network traffic, determining if detected anomalies are associated with malicious packets, detecting the source address of malicious packets in real-time, dropping malicious packets in real-time, and blocking future traffic from the source address. This claim is eligible because it integrates the abstract idea into a practical application by providing a specific technological improvement (network security). The steps involve specific actions taken to enhance network security, such as real-time detection and blocking of malicious packets, which make the method practical and useful.

USPTO Example 48: Speech Separation

Background: The invention addresses the separation of speech signals from mixed audio inputs using a deep neural network (DNN). It enhances traditional techniques, especially in environments with multiple speakers.

Claims:

Claim 1 describes a method comprising receiving a mixed speech signal containing speech from multiple sources, converting the mixed speech signal into a spectrogram using a short-time Fourier transform (STFT), and using a DNN to determine embedding vectors from the spectrogram. This claim is ineligible because it recites a judicial exception (abstract idea) without integrating it into a practical application. The steps involve mathematical operations (STFT, DNN embedding vector calculation), which are abstract. The claim lacks additional elements that transform it into a practical application or provide a specific technological improvement.

Claim 2 extends Claim 1 by including additional steps: partitioning embedding vectors into clusters, applying binary masks to the clusters to create masked clusters, synthesizing speech waveforms from the masked clusters, combining the speech waveforms to generate a mixed speech signal excluding a target source, and transmitting the mixed speech signal for storage. This claim is eligible because it integrates the abstract idea into a practical application by improving speech separation technology. The steps involve concrete actions that enhance the functionality of speech separation, such as generating and transmitting a mixed speech signal, making the claim practical and useful.

Claim 3 describes a non-transitory computer-readable storage medium storing instructions that cause a processor to perform operations: receiving a mixed speech signal, using a DNN to convert the signal to embeddings, clustering the embeddings, applying binary masks to obtain masked clusters, converting masked clusters to separate speech signals, and extracting spectral features from a target source to generate a transcript. This claim is eligible because it integrates the abstract idea into a practical application by enhancing speech-to-text transcription. The steps involve specific actions that improve the process of converting speech to text, making the claim practical and useful.

USPTO Example 49: Fibrosis Treatment

Background: The invention personalizes medical treatment for glaucoma patients at risk of post-implantation inflammation (PI) using a polygenic risk score (PRS) model and a new anti-fibrotic drug, Compound X.

Claims:

Claim 1 describes a method comprising collecting and genotyping a sample from a glaucoma patient to provide a genotype dataset, identifying the patient as at high risk of PI based on a weighted PRS generated by an ezAI model, and administering an appropriate treatment to the high-risk patient after microstent implant surgery. This claim is ineligible because it recites a judicial exception (abstract idea) and does not integrate it into a practical application. The step of identifying the patient involves a mental process and mathematical calculation, which are abstract. The claim lacks additional elements that provide a specific technological improvement or a concrete application.

Claim 2 specifies that the appropriate treatment in Claim 1 is Compound X eye drops. This claim is eligible because it integrates the abstract idea into a practical application by administering a specific treatment (Compound X) to a specific patient population (glaucoma patients at high risk of PI). The administration of Compound X eye drops is a concrete step that provides a tangible benefit, transforming the abstract idea into a practical medical treatment.

Practice Pointers

When drafting or litigating AI-related patent claims to ensure they are patentable, it is essential to navigate the complexities of patent eligibility criteria carefully. To increase the likelihood of success, claims should emphasize specific, practical applications of the AI technology rather than abstract concepts. For instance, detailing how an AI system improves network security or enhances speech-to-text transcription integrates the abstract idea into a practical, technological application, as seen in the eligible claims from the examples.

It is crucial to avoid claims that only describe mathematical operations, mental processes, or abstract data manipulations without tying them to a specific, tangible application. Claims that merely present abstract ideas without demonstrating how they solve a technical problem or improve existing technology are likely to be deemed ineligible. Including detailed steps on how the AI method provides a concrete technological benefit can significantly strengthen the claim's eligibility.

Highlighting how the AI invention offers technical improvements over existing methods is another effective strategy. For example, specifying how an AI model reduces false positives in anomaly detection or increases accuracy in speech separation can demonstrate the inventive step and practical application necessary for patent eligibility. This approach was successfully employed in claims that integrated AI technology into specific, beneficial applications.

Incorporating claims that describe the physical or structural aspects of the AI system can also anchor the claim in tangible technology, making it more likely to be considered patentable. Describing specific hardware components, like ASICs used in an AI system, provides a concrete basis for the claim, as seen in the eligible claims of Example 47.

When the claim involves abstract ideas or judicial exceptions, it is important to ensure these are part of a broader, practical application that provides a tangible benefit. For instance, using an AI model to administer a specific medical treatment based on genetic analysis demonstrates the practical application of the abstract idea, as illustrated in Example 49. This approach shows how the AI technology can be applied in a real-world context to provide a concrete benefit.

Avoiding generic implementation claims is also essential. Claims should not merely instruct to apply an abstract idea on a generic computer but should describe how the AI technology specifically and uniquely improves the functioning of the computer or the technological field it is applied to. By focusing on these strategies, inventors can draft AI-related patent claims that clearly demonstrate practical applications, technological improvements, and tangible benefits, thereby enhancing their chances of being considered patentable by the USPTO.

Public Participation and Further Information:

The USPTO invites public comments on the guidance update and the new examples through September 16, 2024. Interested parties can find the full text of the guidance update and the examples on the USPTO’s website. Detailed instructions for submitting comments are available in the Federal Register Notice.

Citations:

2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, 89 Fed. Reg. 45,177 (July 17, 2024), available at https://www.federalregister.gov/documents/2024/07/17/2024-15377/2024-guidance-update-on-patent-subject-matter-eligibility-including-on-artificial-intelligence (last accessed July 18, 2024).

July 2024 Subject Matter Eligibility Examples, United States Patent and Trademark Office, July 2024, available at https://www.uspto.gov/sites/default/files/documents/2024-AI-SMEUpdateExamples47-49.pdf (last accessed July 18, 2024).

This website uses cookies to enhance your browsing experience and provide you with personalized services. By continuing to use this site, you consent to the use of cookies. See our Terms of Engagement to learn more.
ACCEPT