Blog Post: When it comes to obtaining a patent, how much generative AI is too much?
Is generative artificial intelligence ("AI") a mysterious gremlin, lurking in the background on the edge of sentience, as portrayed in dystopian sci-fi plotlines? Or is it a benign tool, merely analyzing inputs to predict an output? Can such a tool be used to assist in creating a patentable invention? Or will the AI gremlin soon become the inventor itself?
Demystifying Self-Learning AI:
AI is a complex array of information that "teaches" itself by comparing its predicted output to the expected output. The information is arranged as matrices and vectors which are multiplied to create a "neural network." This neural network acts as a generative pre-trained transformer ("GPT"), that transforms the given input into a predicted output by breaking the input into small tokens of information. Each token of information is associated with a vector. Vectors can be conceptualized as arrows, pointing to different areas of the matrix.
The neural network "learns" through a process called back-propagation. Back-propagation works by comparing the neural network's prediction with the expected outcome and sending this information back. Back-propagation nudges the weights and biases put on the tokens to redirect the vectors. As the predicted output is course-corrected via back-propagation, these vectors are adjusted and nudged to point to areas of the matrix which produce a more expected output.
This process of learning consists of many small adjustments layered on top of each other. Information propagates forwards and backwards until the vectors are nudged in such a way as to predict an expected output.
This is how over time, a neural network can learn to compose a story or find certain answers. It utilizes vectors, perfected by previous forward and back-propagation, to determine the "direction" the story should go or "point" to the area in the matrix in which the answer lies.
Because AI's learning process is mathematical, consisting largely of multiplying matrices and vectors which form the neural network, there is a question as to whether it is considered an abstract idea. Abstract ideas, such as mathematical concepts, are generally not patent eligible.
Patentability of AI-Assisted Inventions:
To what extent can an inventor use generative AI to assist in a patentable invention? Can machine learning be utilized to assist in the conception of an invention? Can the AI process itself be considered an invention when it produces an improved output?
The U.S. Patent and Trademark Office ("PTO") recently issued its 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence (the "Guidance Update").[1] This Guidance Update came as a response to Executive Order 14110 (the "EO"), addressing the "Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence."[2] The EO lays out eight (8) guiding principles for the development and use of AI, including promoting "responsible innovation, competition, and collaboration" so that the United States may "lead in AI and unlock the technology's potential solve some of society's most difficult challenges."[3] The PTO's Guidance Update was issued to create clarity in evaluating the subject matter eligibility of claims in patent applications involving AI inventions. The guidance also provided three examples of what types of inventions are patent eligible and which are not.
Generally, certain categories are not eligible for patent protection because they are not considered appropriate subject matter.[4] These categories include the following judicial exceptions: abstract ideas, laws of nature, and natural phenomena.[5] Abstract ideas include mathematical concepts and mental processes.[6] However, a judicial exception, such as an abstract idea, can still meet subject matter eligibility if the claim also describes the integration of the abstract idea into a practical application of that abstract idea.[7] An example of such an abstract idea that meets the subject matter eligibility requirement might be a blockchain-based voting system. Although blockchain technology is an abstract idea/mathematical concept, this example might be patent eligible if the invention describes how the abstract idea of blockchain technology is used to create the specific implementation of a transparent and secure voting system.
Because of the mathematical nature of AI, there is a question as to whether AI-assisted inventions would be considered patent ineligible abstract mathematical concepts. However, the PTO's Guidance Update indicates that an invention which uses abstract ideas to create a practical application that improves another technology will meet the subject matter eligibility requirement. In support, the PTO provides three (3) examples of AI-assisted inventions and explains whether and why the claims in each example would or would not meet the subject matter eligibility requirement.[8] These examples are:
- Anomaly detection (Example 47): claims that recite the use of an artificial neural network to identify or detect anomalies;
- Speech separation (Example 48): claims that recite AI-based methods of analyzing speech signals and separating desired speech from extraneous or background speech; and
- Fibrosis treatment (Example 49): claims that recite an AI model that is designed to assist in personalizing medical treatment to the individual characteristics of a particular patient.
Each of these three examples recites multiple claims. Subject matter eligibility is met when the claim either does not describe a judicial exception, i.e. an abstract idea, law of nature, or natural phenomenon, or when it describes a judicial exception and then integrates that exception into a practical application. When there was not an inventive concept provided above and beyond the use of AI, the PTO found the claim to be ineligible.
As a practical matter, when does a mathematical concept such as AI improve another technology?
The Guidance Update cited In re Board of Trustees of Leland Stanford Junior University and emphasized the Federal Circuit's ruling that an improvement in the judicial exception itself is not an improvement in the technology.[9] More specifically, it held that the improved process, which yielded a greater number of predictions, was not an improved technological process, but was instead, an improved (and not patent eligible) mathematical process.[10] In other words, a patent claim that merely produces a better output and does not use that output to improve an application will not meet the subject matter eligibility requirement. To be patent eligible, the claim would have to describe an improvement to a technological process.[11]
The Guidance Update does not give specific guidelines or rules for when an AI-assisted invention would be deemed to improve another technology or technical field. Thus, the question has not been answered as to whether an invention that merely improves the method for producing an output, rather than using that output for an additional application, would be patent eligible.
Humans Remain the Inventor:
Case law and recent PTO guidance demonstrate that, although AI can be used to assist in an invention, each claim of a patent application must name a human inventor.[12] AI systems cannot be the sole inventor.[13] Further, an invention must use AI to integrate the AI into a specific application to be patent eligible. Thus, although AI has the potential to greatly increase innovation and can be a powerful tool in doing so, it seems that, for now, the spark of conception of an invention remains in the realm of the humans.
[1] https://federalregister.gov/d/2024-15377 (last visited August 5, 2024).
[2] https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence (last visited August 5, 2024).
[3] Id. at Sec. 2 (b).
[4] See Manual of Patent Examining Procedure ("MPEP") 2106.
[5] See MPEP 2106.04 (I).
[6] Id. at 2106.04(a)(2), subsection I and subsection III.
[7] See id. at 2106.04(II).
[8] Guidance Update, at V; see https://www.uspto.gov/sites/default/files/documents/2024-AI-SMEUpdateExamples47-49.pdf (last visited August 5, 2024).
[9] In re Board of Trustees of Leland Stanford Junior University, 989 F.3d 1367, 1370, 1373 (Fed. Cir. 2021) (Stanford I).
[10] Id. at 1373.
[11] Id. at 1373–74.
[12] Thaler v. Vidal, 43 F.4th 1207, 1210, 1213 (Fed. Cir. 2022), cert. denied, 143 S. Ct. 1783 (2023).
[13] Id.
