Subject Matter Eligibility for AI Patents Under Section 101

Section 101 of the U.S. Patent Act defines the boundaries of what constitutes patentable subject matter. This is a critical consideration for any invention, including those involving artificial intelligence (AI). As AI technologies continue to advance and integrate into various fields, they present unique challenges for establishing patent eligibility under Section 101. The requirements under Section 101 are largely derived from judicial interpretations, reflecting a complex legal landscape shaped by key court decisions. This section explores the intricacies of determining patent eligibility for AI inventions and offers insights into the legal standards and strategies for navigating these complex requirements.

This page is divided into these parts:

Understanding Section 101

Section 101 of the U.S. Patent Act states that a patent may be obtained for any new and useful process, machine, manufacture, or composition of matter. Despite its broad language, the Supreme Court has long held that there are implicit exceptions to patentable subject matter: laws of nature, natural phenomena, and abstract ideas. These non-statutory exceptions are not explicitly mentioned in the statute but have been established through judicial interpretation to ensure that fundamental scientific principles remain free for all to use.

Under the Section 101 statute itself, an invention must fall within one of the four listed statutory categories: process, machine, manufacture, or composition of matter. While most inventions fit into these categories, some do not. For instance, an invention claimed as merely software code or pure data is not eligible for patent protection since these items do not fall under one of these four categories. Similarly, an application that only claimed a ‘trained neural network’ would not be eligible for patent protection under Section 101, but an application that claimed a method (process) of training a neural network, or a computer (machine) that utilizes the trained neural network, would be eligible. These types of statutory category issues are relatively easy to overcome by simply changing the focus of the claims. 

The real complexity in Section 101 eligibility arises from the non-statutory exceptions: laws of nature, natural phenomena, and abstract ideas. The Supreme Court has emphasized that these exceptions are crucial because patents on these basic tools of science and technology could impede innovation rather than promote it. For instance, one cannot patent the laws of physics or a mathematical algorithm in its abstract form. However, a novel application of these principles, such as a new and useful structure or process derived from them, can be patentable.

The Supreme Court’s decisions in Mayo Collaborative Services v. Prometheus Laboratories, Inc. and Alice Corp. v. CLS Bank International established a two-part framework, often referred to as the Alice test, for determining patent eligibility under Section 101. The first step is to determine whether the claims are directed to a patent-ineligible concept, such as an abstract idea or natural phenomenon. If so, the second step is to examine whether the claim elements, individually or as an ordered combination, contain an “inventive concept” sufficient to transform the ineligible concept into a patent-eligible application. This test has been challenging to apply consistently, leading to considerable uncertainty and debate over what constitutes patent-eligible subject matter.

Section 101 Recommendations for AI Inventions

Analyzing Section 101 under the Alice test can get extremely complicated. Bitlaw provides numerous resources and explanations to help in understanding this issue. A listing of these resources is found on Bitlaw's Section 101 Index. This index includes links to Bitlaw Guidance pages that cover the following topics: What is a Section 101 Rejection, Applying Step One of the Alice test, Applying Step Two of the Alice test, and How to argue against a Section 101 Rejection. Additionally, Bitlaw offers a database of Section 101 Cases, providing detailed summaries and outcomes of relevant court decisions. Additional resources include a Bitlaw summary of how the Manual of Patent Examining Procedure (MPEP) instructs examiners on Section 101.

There are some practical suggestions that can be drawn upon from all of these resources when trying to obtain patent protection on an AI invention. The following are a few suggestions for inventors and patent attorneys working in this area.

Define Technical Problem and Solution: When drafting a patent application for an AI invention, it is crucial to clearly define the technical problem that the invention addresses. Start by providing a detailed description of the issue at hand, emphasizing its technical nature. Follow this by explaining how the AI invention provides a technical solution to this problem. This approach not only aligns the invention with patentable subject matter but also demonstrates that the invention solves a specific, technical problem rather than being an abstract idea.

Highlight Technical Improvements: Emphasizing how the AI improves the functioning of a specific technology or system can be essential to overcoming Section 101. In the patent application itself, it is best to articulate the technical improvement provided by the AI, such as increased efficiency, accuracy, or performance. For instance, if the AI algorithm enhances data processing speed or improves predictive accuracy, these improvements should be highlighted in the specification. Ideally, one should compare the state of the technology before the invention with the improvements provided by the invention. Detailing these enhancements helps to demonstrate that the invention is not merely an abstract idea but a tangible improvement over existing technologies. The claims should cover those aspects of the invention that provide this technical improvement.

Avoid Data Processing Claims: When drafting claims for an AI invention, it is important to avoid those that merely involve receiving data, processing it, and outputting more data. An example of a claim to avoid is one that first accepts financial data as training data, then trains a neural network with that data, and then uses the trained neural network to score a transaction or an opportunity. Such claims are often viewed as abstract ideas and are likely to face rejections under Section 101. Instead, try to identify some specific technical innovations involved in how the data is pre-processed, or how the AI system is trained, or how the output data is used to improve some other technology.

Demonstrate Practical Application: Ensuring that the claims demonstrate a practical application of the AI technology is another effective strategy. This appears to be particularly useful in obtaining AI patents from the European Patent Office, but it can also be useful at the USPTO. Describe how the AI is applied in a concrete, real-world scenario to achieve a tangible result. For example, if the AI is used in a medical diagnostic system, explain how it improves diagnostic accuracy and efficiency in clinical settings. Highlighting the practical applications helps argue that the invention is not merely an abstract idea but has practical utility.

Identify any Unique Technical Details: Including specific technical details and steps in the claims is crucial to avoid claims that are considered to be directed to abstract ideas. Detailed descriptions of how the AI operates, the specific algorithms used, and how it interacts with other components or systems can help establish the invention's technical nature. For instance, if the AI uses a novel neural network architecture or a unique training method, these details should be explicitly stated in the claims. This level of specificity not only clarifies the scope of the invention but also demonstrates its technical depth, which can be crucial for overcoming Section 101 rejections.

Integrate AI with Hardware: If it makes sense in the context of the invention, integrating the AI with specific hardware components in the claims can be helpful. This can help in demonstrating that the AI is not merely a software algorithm but part of a patent-eligible machine or process. For example, if the AI is embedded in a medical device or integrated into an industrial machine, this hardware integration should be clearly detailed in the claims. Describing the interaction between the AI and the hardware components can support the argument that the invention is a concrete and tangible innovation, enhancing its eligibility under Section 101. Merely reciting a generic computing device that is used to train or operate a neural network, however, may not be helpful.

Example 39

The USPTO has outlined its understanding of Section 101 in the Manual of Patent Examining Procedure, specifically in sections 2104-2106. The process used by the Patent Office is rather complicated, but this is not surprising considering the difficulty in understanding the various court decisions that define this area of law. Bitlaw has summarized the Patent Office's approach on its page on the MPEP and Section 101. A major part of this approach is determining whether a claim being analyzed is directed to an "abstract idea." The MPEP has defined a limited number of groupings of abstract ideas, and only these items can be considered as abstract ideas when an examiner wants to reject an application under Section 101. These groupings are:

  • Mathematical concepts – mathematical relationships, mathematical formulas or equations, and mathematical calculations;
  • Certain methods of organizing human activity, namely:
    1. Fundamental economic principles or practices (including hedging, insurance, and mitigating risk)
    2. Commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; and business relations)
    3. Managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions); and
  • Mental processes – concepts performed in the human mind (including an observation, evaluation, judgment, opinion).

In addition to the MPEP, the Patent Office has provided many examples of fact scenarios and how they should be interpreted under the MPEP's Section 101 analysis. The only example directly relevant to artificial intelligence is Example 39. This example relates to an AI system designed to improve the ability to identify human faces in digital images. Prior methods that used neural networks for facial detection struggled with robustness, particularly in handling shifts, distortions, and variations in facial images. The invention in this example improves this technology by using an expanded training set developed through mathematical transformations on facial images and by employing an iterative training algorithm to minimize false positives. The result of this approach is a more reliable face detection model.

The claim that was analyzed in Example 39 was:

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.

Thus, the steps in this example related to collecting data, modifying the data, creating two training sets, and then training the neural network on the training sets. In analyzing this claim, Example 39 concluded that the claim was eligible under Section 101 because the claim "does not recite any of the judicial exceptions enumerated" in the MPEP. In particular, the PTO noted that these steps do not recite any mathematical relationships, formulas, or calculations (while the claims may be ‘based on mathematical concepts, the mathematical concepts are not recited in the claims’), do not recite a mental process (‘because the steps are not practically performed in the human mind’), and do not recite any method of organizing human activity, such as a fundamental economic concept or managing interactions between people. Thus, the claim is eligible because it does not recite a judicial exception.

When this example came out, most patent practitioners assumed that this would lead to most claims relating to the training of a neural network being patent eligible. After all, if none of the steps of collecting data, modifying the data, creating a training set, and then training a neural network on the training set recite an abstract idea as enumerated in the MPEP, then any claim comprising of only these steps should likewise not recite an abstract idea.

Current Understanding of Example 39 at the PTAB

Unfortunately, the Patent Office has not read its own example so logically. Instead, examiners are frequently reading Example 39 in a much more limited fashion. To better understand the current understanding of the patent office, numerous decisions of the Patent Trial and Appeal Board were examined that analyzed examiner rejections under Section 101 where the patent applicant argued that the invention should be eligible for protection based on Example 39. Numerous decisions were analyzed, and no decision after 2020 found Example 39 to be persuasive in determining that AI training claims were patent eligible. Instead, some common themes were discovered, which are listed below along with language taken from the PTAB decisions:

  1. Example 39 is Limited to Image Analysis:
    • "To the extent Appellant contends that training a network is sufficient for patent eligibility in view of Example 39 (see Appeal Br. 4), we disagree. As noted above, Example 39 involves, among other things, transforming image data into modified image data which cannot practically be performed in the human mind."
    • "In Example 39, the data used to train the model are “digital facial images” to which transformations are applied including “mirroring, rotating, smoothing, or contrast reduction to create a modified set of facial images” (id.). On the other hand, “Appellant's only inputs and outputs of the models are numerical in nature”"
    • "Thus, Example 39 addressed technological difficulties related to analyzing graphic images to identify and analyze facial images within them. Appellant has neither identified nor demonstrated that the present claims provide such image analysis. Instead, the Specification says the claims are directed to using information handling systems to estimate demand for healthcare resources"
    • "The USPTO discussion of Example 39 does not specify what particular claim language compels the determination that the steps of the claim are not practically performed in the human mind. We observe that the claims recite, beyond simply a first training set and a second training set, further limitations such as “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.” No such similar recitation is present in Appellant's claims."
  2. Example 39 is Limited to Data Preparation or Two Training Stages:
    • "Here, the training data is not transformed or curated in any particular way in order to train the machine learning model. Nor is the model trained in stages to improve its accuracy as in Example 39.
    • "Nor does claim 1 here involve steps of training a neural network with digital images or any other data through stages or otherwise."
  3. Any Financial Component (or other Abstract Idea) Renders Example 39 Moot:
    • "Put simply, claim 1, unlike the hypothetical claim in Example 39, recites a commercial interaction, including advertising, marketing, or sales activities or behaviors, which is a certain method of organizing human activity and, thus, an abstract idea."
    • "That is, unlike Example 39, which did not include any management of personal behavior or relationships or interactions between people, claim 84 helps manage the relationship between a health care provider and patient by improving the treatment protocols determined for a new patient."
    • "We acknowledge that claim 39 is similar to the present claim 1 in that both claims involve machine learning. ... [T]he commentary indicates that the claim “does not recite any mathematical relationships, formulas or calculations” and the steps cannot “practically [be] performed in the human mind.” Id. Here, as discussed above, claim 1 recites mathematical calculations and can be performed in the human mind."
  4. These Elements Are Just Generic Functions:
    • "The functions performed in the claims -- receiving training data, applying training data, accessing data, and outputting data -- are [] routine, conventional, and well-known functions of a generic computer."

Additional Recommendations Based on Example 39

Although this analysis of Example 39 indicates that it is much less useful than originally anticipated, there are some useful learnings that can be gleaned from this analysis. In particular, these recommendations can be added to the list of recommendations above.

Describe Unique Data and Data Handling: Providing detailed descriptions of how the AI is trained and how it handles data is essential. Explain the specific algorithms, models, and techniques used to develop the training data, and how these contribute to solving the technical problem. For instance, if the AI uses a novel training dataset or a unique method of data pre-processing, these details should be included in the patent application and form part of the claims. Example 39, for instance, trained the neural networks on facial images that had undergone relatively simple transformations. Yet, several PTAB decisions emphasized that the pre-processing of the training data was one of the key elements that allowed Example 39 to be eligible under Section 101.

Claim Unusual Steps in the Training Process: Providing a detailed account of the training process for AI systems can also be very helpful. Example 39 claimed a two-step training process: first training the neural network on a broad set of transformed facial images, and then retraining the neural network on a second set based on detected errors. This iterative approach not only improved the model’s robustness but also addressed false positives. Several PTAB decisions emphasized this two-step process as one of the primary reasons that Example 39 was patent eligible. If possible, new AI patent applications should highlight any unique elements in the training process in the specification, and those elements should be included in the claims to help overcome any potential Section 101 issues.

USPTO AI-Specific Section 101 Guidance

President Biden issued an Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence on October 30, 2023. In this Order, the President required that the USPTO develop guidelines to help interpret when AI-assisted inventions were eligible for patent protection. In response, the Patent Office issued its Inventorship Guidance for AI-Assisted Inventions on February 13, 2024. The order also required that, by July 26, 2024, the USPTO shall issue "additional guidance to USPTO patent examiners and applicants to address other considerations at the intersection of AI and IP, which could include, as the USPTO Director deems necessary, updated guidance on patent eligibility to address innovation in AI and critical and emerging technologies." Hopefully, in the near future the USPTO will issue the required guidelines and provide additional insight into when and how AI-related inventions can be considered patent eligible under Section 101.

Artificial Intelligence (AI) Patent Attorney

Please see Dan Tysver's bio and contact information if you need any AI-related legal assistance. Dan is a Minnesota-based attorney providing AI advice on intellectual property and litigation issues to clients across the country.