Add a bookmark to get started

Forest
29 March 20218 minute read

Understanding the USPTO guidance on patenting AI technologies

Innovations in artificial intelligence (AI) seeking to improve AI processes, AI architectures, and computer hardware are rapidly increasing in fields as far-ranging as commercial banking, sports analytics, cybersecurity, natural language processing, computer hardware, review authentication, and healthcare.  As companies innovate, the rate of corresponding patent applications has dramatically increased. The USPTO reports that, from 2002 to 2018, the number of AI-related applications more than doubled.1The European Patent Office estimates that since 2016 the number of international patent families in core AI technologies has increased at an average annual rate of 54.6 percent.2

Because most AI filings are embodied as software-based applications, applicants should understand the complexities of subject matter eligibility.  In this article, we take a look at one aspect of the most recent USPTO eligibility standard and how it can be addressed.

In 2019, the USPTO published two sets of eligibility guidelines which significantly helped advance software applications – the 2019 Revised Patent Subject Matter Eligibility Guidance (the SME Guidance) and the October 2019 Update: Subject Matter Eligibility (the October Update). The SME Guidance sets forth guidelines for subject matter eligibility in order to increase predictability and consistency in the patent eligibility analysis. Under the SME Guidance, the USPTO revised Step 2A of the Alice test to emphasize that the appropriate analysis should focus on whether the claim is directed to an abstract idea, not merely whether a claim recites an abstract idea.  Specifically, Step 2A is now subdivided into two prongs: (I) determining whether a claim recites a judicial exception; and (II) evaluating whether the claims, as a whole, integrate the recited judicial exception into a practical application of the alleged exception.  Step 2B remains unchanged and focuses on whether the claims include limitations that provide significantly more than the judicial exception.

While Step 2A, Prong 2 has received the most attention, as it has provided Applicants with a strong avenue for overcoming eligibility challenges, Step 2A, Prong 1 has provided Applicants pursuing patent protection on AI innovations with an equally strong option. Specifically, to assist examiners in determining whether a claim recites a judicial exception, the USPTO set forth enumerated groupings of abstract ideas under Step 2A, Prong 1.  The enumerated groupings include mathematical concepts, certain methods of organizing human activity, and mental processes.  Below, we consider the application of each category to AI innovations.

Mathematical concepts

The SME Guidance defines “mathematical concepts” to include mathematical relationships, mathematical formulas or equations, and mathematical calculations.  As those skilled in the art understand, underlying most, if not all, innovations in AI and machine learning are mathematical algorithms that power AI/machine learning engines.  While on its face the mathematical concepts grouping may appear to problematic for AI filings, the October Update clarifies what is meant by reciting a mathematical concept. 

The October Update provides that, when determining whether a claim recites a mathematical concept, examiners should consider whether the claim recites a mathematical concept or merely includes limitations that are based on or involve a mathematical concept.  Here, the USPTO has made a clear distinction between a specific recitation of a mathematical relationship, formula, equation, or calculation in a claim compared to limitations that include elements that are based on or involve mathematical concepts.  The distinction is especially beneficial for AI applications that depend on mathematical concepts.  As a result, applicants can navigate around the mathematical concept category by drafting claims in a manner that avoids direction recitation of the mathematical algorithms that underly AI models.

Mental processes

The SME Guidance defines the grouping of “mental processes” as concepts performed in the human mind.  Examples include, but are not limited to, observations, evaluations, judgments, and opinions. 

The October Update expanded on this definition by providing that a claim recites a mental process when the claim includes limitations which can be practically performed in the human mind.  In contrast, a claim does not recite a mental process when the claim includes limitations which cannot be practically performed in the human mind, such as when the human mind is not equipped to perform the claim limitations.  This nuance is key for AI innovations.

One of the benefits of AI technology is the ability to more accurately (and quickly) perform observations, evaluations, or forecasts using large sets of data.  In fact, AI can be broadly defined as a computer system that can perform tasks that normally require human intelligence. Still, as AI innovation expands, so too does this definition.  AI can now be broadly defined as systems that can learn and innovate as data evolves.  Although AI is often equated with human intelligence and human learning, the processes carried out by AI systems and the ways in which AI systems learn are significantly different from their human equivalents.  As a result, applicants can navigate around the “mental processes” grouping by arguing that the claim includes limitations that are not practically performed in the human mind. 

Certain methods of organizing human activity 

The SME Guidance defines the grouping of “certain methods of organizing human activity” as including fundamental economic principles or practices, commercial or legal interactions, and managing personal behavior, relationships, or interactions between people.  Out of the three categories discussed in this article, “certain methods of organizing human activity” will likely pose the most trouble for applicants as it permits examiners to look at the task for which the AI innovation is optimized. 

For example, “certain methods of organizing human activity” may be triggered in a claim that utilizes AI techniques in the context of financial technology, such as forecasting spending habits (fundamental economic concept).  In another example, the grouping may be triggered in a claim that utilizes AI techniques in the context of advertisement technology, such as tailoring advertisements to users (commercial interaction). 

This category is troublesome because of the low bar set by the SME Guidance for examiners to satisfy the first prong of Step 2A: examiners only need to show that a claim includes elements that recite an abstract idea.  Accordingly, should applicants receive an eligibility challenge that is based on the “certain methods of organizing human activity” grouping and the claim involves any of the above subgroupings, applicants should utilize both the second prong of Step 2A and Step 2B to overcome the challenge.

In addition to the SME Guidance and October Update, the USPTO’s 2019 Subject Matter Eligibility Examples: Abstract Ideas (SME Examples) can also assist applicants in challenging an eligibility challenge under the first prong of Step 2A.  The SME Examples include illustrative claim analyses that are to be performed under the SME Guidance.  Example 39, focused on machine learning processes, is an example of a claim which the USPTO deems patent eligible:

A computer-implemented method of training a neural network for facial detection comprising:

collecting a set of digital facial images from a database;

applying one or more transformations to each digital facial image including mirroring, rotating, smoothing, or contrast reduction to create a modified set of digital facial images;

creating a first training set comprising 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 stage of training comprising the first training set and digital non-facial images that are incorrectly detected as facial images after the first stage of training; and

training the neural network in a second stage using the second training set.

 

The USPTO explains that Example 39 “does not recite any mathematical relationships, formulas, or calculations. While some of the limitations may be based on mathematical concepts, the mathematical concepts are not recited in the claims. Further, the claim does not recite a mental process because the steps are not practically performed in the human mind.” 

 

Additionally, although not explicitly stated by the USPTO, the claim does not recite certain methods of organizing human activity.  For example, when considering the task for which the claimed neural network is optimized (facial detection), the task does not fall under any of the subgroupings (fundamental economic concept, commercial or legal interactions, and managing personal behavior, relationships, or interactions between people).  Accordingly, applicants would be prudent to rely on Example 39 in traversing an eligibility challenge.

 

Conclusion

 

Software-based innovations related to AI innovations are thriving under the most recent standard set forth by the USPTO.  As a result of the guidance, it has become more difficult for examiners to categorize artificial intelligence innovations under a subject matter grouping, opening the door for applicants to obtain patent protection for their artificial intelligence technology.



1 https://www.uspto.gov/sites/default/files/documents/OCE-DH-AI.pdf

Print