USPTO Issues 2024 Guidance on Patent Eligibility for AI Inventions

by Dennis Crouch

Earlier this week I was reviewing some of the USPTO’s eligibility examples, noting that they were all quite old.  As if on cue, the Office has released a new set of updated guidelines – focusing on Artificial Intelligence related inventions and including three new examples.  In Bilski, the Supreme Court explained that the best way to understand whether a particular claimed invention is directed to an “abstract idea” is to look back on old examples for guidance.  The USPTO has found that a good way to administer this approach is to provide examples of situations that pass or fail the test.  Here, they introduce three new examples 47, 48, and 49.  And, while the Alice/Mayo test for analyzing subject matter eligibility has not changed, the new guidance is helpful as AI technology rapidly develops.   The USPTO continues to be open to issuing patents on AI inventions, including the use of AI. However, there must be a technical solution to a technical problem.

Although the guidance is effective July 17, 2024, the USPTO is open to comments via the regulations.gov portal.  The agency has provided two updated flow charts for its analysis that are included below. Although the USPTO guidance is not binding law,  it is the guidebook that examiners will be trained upon and required to use.  As such, any patent attorney operating practicing in the AI area should dig through these examples and the particular cases chosen by the Office as representative.

One key to the USPTO approach is that, for the USPTO, an abstract idea will fall into one of three categories: mathematical concepts, mental processes and “certain” methods of organizing human activity. Here these “certain methods” are almost entirely limited to “fundamental economic principles or practices, commercial or legal interactions, and managing personal behavior or relationships or interactions between people.”

There are a few newish cases on-point that the USPTO believes are important for examiners. The first two cases side with the patentee, while the remaining six find the claims abstract.

• Not Math: XY, LLC v. Trans Ova Genetics, 968 F.3d 1323 (Fed. Cir. 2020).  The claimed 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 was not directed to “the abstract idea of using a ‘mathematical equation that permits rotating multidimensional data’” even though they may have involved mathematical concepts.
• Not Mental Process: ADASA Inc. v. Avery Dennison Corp., 55 F.4th 900 (Fed. Cir. 2022). The claimed “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”), was found to be patent eligible because it cannot practically be performed in the human mind.
• Mental Process: Trinity Info Media, LLC v. Covalent, Inc., 72 F.4th 1355 (Fed. Cir. 2023). The claimed 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 to be an abstract idea 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”).
• Mental Process: In re Killian, 45 F.4th 1373 (Fed. Cir. 2022). The claimed method of 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) is what the USPTO would call an abstract mental task.
• Mental Process: PersonalWeb Techs. LLC v. Google LLC, 8 F.4th 1310 (Fed. Cir. 2021). The claims reciting “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 to be an abstract idea because they cover “a medley of mental processes that, taken together, amount only to a multistep mental process,” such that the steps can be practically performed in the human mind.
• Organizing Humans: Weisner v. Google LLC, 51 F.4th 1073 (Fed. Cir. 2022). The claimed method of “collect[ing] information on a user’s movements and location history [and] electronically record[ing] that data” (i.e., “creating a digital travel log”), is ineligible as a form of “managing personal behavior or
  relationships or interactions between people.”
• Organizing Humans: Elec. Commc’n Techs., LLC v. ShoppersChoice.com, LLC, 958 F.3d 1178 (Fed. Cir. 2020).  Claimed method of “monitoring the location of a mobile thing and notifying a party in advance of arrival of that mobile thing”  is an example of a fundamental economic principle or practice.
• Organizing Humans: Bozeman Fin. LLC v. Fed. Reserve Bank of Atlanta, 955 F.3d 971 (Fed. Cir. 2020). Here, the claimed to method 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, is collectively a fundamental economic principle or practice.

The Guidance also spends significant time on what the USPTO calls Step 2B –  asking “whether the claimed invention as a whole integrates the recited judicial exception into a practical application.”    The memo reiterates that the examiner is not required to provide evidence in most situations:

As explained in MPEP 2106.07(a), subsection III, “[a]t Step 2A Prong Two or Step 2B, there is no requirement for evidence to support a finding that the exception is not integrated into a practical application or that the additional elements do not amount to significantly more than the exception unless the examiner asserts that additional limitations are well-understood, routine, conventional activities in Step 2B.

Example 47: AI for Anomaly Detection Using Neural Networks

Example 47 focuses on claims that recite an application-specific integrated circuit (ASIC) implementation of an artificial neural network for anomaly detection.

Claim 1 recites an ASIC comprising neurons organized in an array, with each neuron having a register, processing element and input. The neurons are connected via synaptic circuits that include memory for storing synaptic weights.  Here, this is not a software implementation, but rather a design of a new integrated circuit.   With that in mind, the guidance concludes that this claim does not recite an abstract idea because it is focused on a specific hardware implementation of a neural network. The claim is patent eligible.

In contrast, claim 2 more broadly recites a method of detecting network intrusion using a neural network model. The USPTO guidance concludes that this claim recites an abstract idea (a mental process of identifying suspicious activity) but is still patent eligible because it integrates the abstract idea into a practical application – an improvement in network security technology.

This example highlights that AI claims focused on specific hardware implementations or that incorporate the AI model into an improved technological process are more likely to be patent eligible.

Example 48: AI for Analyzing Speech Signals

Example 48 focuses on claims reciting AI-based methods of analyzing speech signals to separate desired speech from background noise or extraneous speech.

Claim 1 broadly recites identifying speech portions in an audio input using an AI speech model, and the USPTO guidance concludes that it is directed to the abstract idea of a mental process without an integration into a practical application or additional elements amounting to significantly more.

However, claims 2 and 3, which incorporate the speech analysis AI model into an improved speech signal processing method and computer system are found to be patent eligible.  The USPTO guidance concludes that here that the AI model is not simply applied to the field of speech recognition in a general way, but is used in a specific way (generating a speech mask) to achieve a specific practical application (improved speech recognition in noisy environments). According to the guidance, this specificity and practical integration make the claims patent eligible, even if the AI model itself is considered an abstract idea.

I expect that these two claims are the most controversial, a defense attorney could put together a strong brief arguing that they are ineligible.  In an actual case, the proof would likely be found in the specification — was the claimed “speech mask” simply a linguistic tool generically used to include within the specification, or was it technically described and form the basis of a technological solution?

Example 49: AI for Personalized Medicine

The final example in the guidance considers an AI model designed to recommend treatment options personalized to a particular patient based on their physiological characteristics and the treatments’ predicted therapeutic effects.

Claim 1, particularly recites using the AI model to recommend a treatment for a patient based on a dataset of patient information and treatment outcomes. The guidance concludes that this claim is directed to an abstract idea of a mathematical calculation without significantly more.

However, claim 2, which recites using a more specific AI model based on a neural network and training data to select a treatment from a specific list of options is found to integrate the abstract idea into the practical application of a specific treatment selection process. The claimed AI model and training provides a “particular way to achieve a desired outcome” rather than just claiming the idea of a solution.

Looking at more detail, claim 2 requires a neural network model trained using a plurality of training datasets, each training dataset comprising physiological parameters and at least one indicator of the disease associated with a respective patient, and wherein each training dataset is associated with a respective treatment selected from the group consisting of treatment A, treatment B, and treatment C.  The USPTO concluded that, while using an AI model to recommend treatments may be an abstract idea (a mathematical calculation), claim 2 integrates this idea into a practical application of an improved treatment selection process.

A few key thoughts: 

A couple of weeks ago I set up an automated AI analysis of Supreme Court cases, creating a summary of each.  The guidance makes clear that sort of invention – at least as described at this high level – will not be patent eligible because it is simply applying AI tools in a to automate what was otherwise a mental process.

The guidance focuses primarily on specific hardware implementations or technological integrations of the AI models. In the situation of integration, the preferred approach continues to be: provide a technical solution to a technical problem.

= = =