Supplementary guidance on patent eligibility for AI-related inventions by IPOS
31 Oct 2024
(Above word cloud is computer-generated from selected text of the IPOS Guidance).
The IPOS Guidance on AI Patent Eligible Subject Matter
In October 2024, the Intellectual Property Office of Singapore (IPOS) released “Singapore Guidance for the Examination of AI-related Patent Applications” (hereafter, “the IPOS Guidance”). This concise sixteen-page document details the pertinent legal criteria for patent eligible subject matter within Singapore as applied to artificial intelligence (AI) related patent claims. The IPOS Guidance includes evaluations of nine example claim sets, labelled SG1 through SG9. Each of the examples applies the legal criteria discussed in the IPOS Guidance to evaluate patent eligible subject matter of one or more claims of a given technology.
Subject matter not considered to be patentable includes “discoveries; scientific theories and mathematical methods; aesthetic creations; schemes, rules or methods for performing a mental act, playing a game or doing business; and presentation of information”. According to the IPOS Guidance, the determination of whether a patent claim defines patent eligible subject matter involves the following sequential steps:
-
Properly construing the claim;
-
Identifying the actual contribution; and
-
Asking if the actual contribution falls solely within subject matter that is not patentable.
For those seeking AI patents, perhaps the most important phrase in the IPOS Guidance is “actual contribution”. The actual contribution of a claimed invention is described in the IPOS Guidance as:
“ . . . an exercise in judgment probably involving the problem said to be solved, how the invention works, what its advantages are. What has the inventor really added to human knowledge perhaps best sums up the exercise. The formulation involves looking at substance not form – which is surely what the legislator intended.”
The IPOS Guidance further discusses:
-
Specific Problem Linked to Functionally Limited Solution “If the actual contribution of an AI-related patent claim solves a specific (as opposed to a generic) problem, then it is likely not solely a mathematical method. . . . The claimed method solves a specific problem if the claim is functionally limited to solve the problem, either explicitly or implicitly, for example, by establishing a sufficient link between the problem and the steps of the mathematical method by clearly specifying how the input and the output of the sequence of mathematical steps relate to the problem so that the mathematical method is causally linked to solve said problem. . . . An AI-related patent claim where an AI algorithm is simply used ‘to control a system’ or ‘to manage risk’, without anything more, would be regarded as being directed to a generic rather than a specific problem”.
-
Substance Not Form: “. . . claims that are directed, for example, to a program, a programmed computer or a computer-implemented method but define the same task are expected to be the same in substance and so have the same actual contribution. A separate determination of the actual contribution of each type of claim is not normally expected to be necessary”. The agnostic approach to claim preambles is exemplified in examples SG2 and SG5 of the IPOS Guidance, wherein claims for a server, a method, and a computer program product all are construed as directed toward the same actual contribution.
-
“May Solve” Unlikely to Be Sufficient: “The mere fact that the mathematical method may solve a specific problem is unlikely to be sufficient”. This requirement is impacted by the Singapore patent law construction of terms such as “for” and “configured to”, and “adapted to”, which are most commonplace within system claims. These terms are construed under Singapore patent law as meaning “suitable for”. Hence, AI patent claim sets can benefit from a combination of claim drafting styles such as means plus function and method claims in additional to system claims.
These concepts are not new for Singapore patent law. The Examination Guidelines for Patent Publication published October 2023, which were published roughly a year before the IPOS Guidance, discusses: (i) “actual contribution” in its sections 8.3 to 8.7; (ii) “substance not form” in its section 8.4; and (iii) “specific problem linked” and “may solve” in section 8.24.
Examples from the IPOS Guidance for patent eligble subject matter claims included:
-
Example SG1 (determination of fake data)
-
Example SG3 (removal of extra nodes)
-
Example SG5 (allocation of memory to neural network)
-
Example SG6 (user interaction with CAD software)
-
Example SG7 (prediction of tensile strength using models)
-
Example SG8 (prediction of ACP from measured parameters)
-
Example SG9 (computation of physical nozzle blockage fraction)
The claims of examples that were found to not be patent eligible subject matter included SG2 (stock trading strategy is solely a business method) and SG4 (adding of node and adjustment of weights of AI model).
Each of these examples stepped through the process of construing the claims, identifying the actual contribution, and asking if the actual contribution falls solely within subject matter that is not patentable. In each example, the construction of the actual contribution was critical. If the problem to be solved by the invention was construed as solely a mathematical calculation or solely a business decision, then the claims of the example would be found to not be patent eligible subject matter.
As with examples SG5 (allocation of memory to neural network), SG7 (prediction of tensile strength using models), and SG9 (computation of physical nozzle blockage fraction), claims directed toward use of AI with physical components were found to be patent eligible subject matter while maintaining very broad claim language. Example SG5, for instance, is a computer program product executed on a generic computer with steps commencing with only the words “determine” or “define” that result in the control of a memory device. Each of these two examples provide three claims, one for a computer-implemented method, one for a computer program product, and a server. As a demonstration of the “substance over form” discussed in the IPOS Guidance, all three claims are found to be patent eligible subject matter.
The USPTO Guidance on AI Patent Eligible Subject Matter
The publication of the IPOS Guidance in October 2024 was preceded by a few months with the publication by the U.S. Patent and Trademark Office (USPTO) providing specific guidance for the evaluation of patentable subject matter for AI technology patent applications (hereafter, “the USPTO Guidance”). The USPTO Guidance discusses the patentable subject matter law guidelines as they relate to AI inventions and evaluates eight sample claims against these guidelines.
The difference in complexity between the USPTO Guidance and the examples of the IPOS Guidance are notable. The nine examples of the IPOS Guidance are provided within twelve pages of text, with some of these examples covering less than a page of text. In contrast, the eight USPTO Guidance extend over 34 pages, thus averaging a length four times that of the examples of the IPOS Guidance. The discussion of legal guidelines for the IPOS Guidance covered less than four pages of text. In contrast, the discussion of legal guidelines for the USPTO Guidance covered about 30 pages of text.
In regard to guidance specific to AI inventions, the USPTO Guidance provides the following:
-
Recites v. Involves: “While it is common for claims to AI inventions to involve abstract ideas, USPTO personnel must draw a distinction between a claim that ‘recites’ an abstract idea (and thus requires further eligibility analysis) and one that merely involves, or is based on, an abstract idea . . . a recitation of the words ‘apply it’ (or an equivalent) or are no more than instructions to implement a judicial exception on a computer . . . AI inventions may provide a particular way to achieve a desired outcome when they claim, for example, a specific application of AI to a particular technological field (i.e., a particular solution to a problem” The distinction between “recite” versus “involve” in the USPTO Guidance is similar to the requirement in the IPOS Guidance that the “actual contribution” relates to an exercise in judgment involving the problem said to be solved, how the invention works, and what its advantages are.
-
Mental Process Must by Humanly Possible. “The mental processes grouping is not without limits, and as such, claim limitations that only encompass AI in a way that cannot practically be performed in the human mind do not fall within this grouping.” This USPTO Guidance somewhat conflicts with Singapore Examination Guidelines section 8.7, which states that “if the technical features recited in the claim are such that they are no more than the workings of a standard operating system, in particular, the use of a generic computer or computer system to perform a pure business method, then such an interaction would not be considered to be a material extent and it is apparent that no specific problem is solved.” It can be noted that an AI invention may be deemed a “mental processes”, yet still not be eligible patent subject matter if the claims are considered to constitute a discovery, scientific theory, mathematical method, scheme, or method of doing business.
-
AI Cannot be Inventor: “. . . current statutes (e.g.,35 U.S.C. 101 and 115) do not provide for recognizing contributions by tools such as AI systems (or other advanced systems) for inventorship purposes, . . . Patent protection may be sought for AI-assisted inventions where one or more persons made a significant contribution to the claimed invention.” Singapore provisions and Examination Guidelines have not yet addressed this concern. However, Singapore may be likely to follow the guidance of the recent Thaler case that decided that AI cannot be listed as an inventor under the UK Patents Act 1977.
The bulk of the discussion in the USPTO Guidance, both in its legal guidance and its as evaluation of the eligibility of the eight examples, follows the general guidance on patent eligible subject matter already present within sections 2103 through 2106.07 of the MPEP. Hence, the USPTO Guidance supplements, rather than alters, the existing framework for patentable subject matter evaluation. As a result, the bulk of the example patentable subject matter discussion in the USPTO Guidance is directed at the existing framework detailed in the MPEP.
According to MPEP section 2106, only inventions that fall within one of the four statutory categories—process, machine, manufacture, or composition of matter—are patent eligible subject matter. The eligibility process involves a two-step analysis from the Alice/Mayo framework to determine whether a claimed invention is patent eligible subject matter. When evaluating claims, a first step is to identify whether the claim is directed to a statutory category. If yes, the second step involves determining if the claim is directed to a judicially-excluded concept, such as an abstract idea, law of nature, or natural phenomenon. If it is, the examiner must then assess whether the claim includes an "inventive concept" that transforms it into something more than a mere application of these concepts. This step ensures that the invention is not simply an attempt to patent an abstract idea or natural law, aligning with the broader goals of the patent system. The MPEP also emphasizes that specific claims should be evaluated holistically, based on their practical application, and consider the claims in their entirety to support true innovation.
Of the eight examples detailed in the USPTO Guidance, the more notable include the following:
-
Example US47-1: This example included a very short claim but was directed at a physical semiconductor integrated circuit component and found to be patent eligible subject matter.
-
Example US47-2: This example was found to not constitute patentable subject matter as the claimed invention was directed merely at the training and use of an AI model.
-
Example US48-2: In this example, a dependent claim overcame the patentable subject matter rejection of its rejected independent claim US48-1 by providing functional limitations linked to a specific problem.
-
Example US49-1: The claim in this example computed mathematical vectors as its outputs. It was found to not constitute patentable subject matter as it was directed solely at a mathematical method.
These examples illustrate that use of AI with physical components or devices is more likely to be deemed patent eligible subject matter, as contrasted to inventions directed at training of a model or mathematical calculations. Example US48-2 is helpful to illustrate that addition of further technical details to independent claims is a primary strategy toward overcoming office action rejections. For Singapore patent applications, the format of the AI claims is not critical. AI claims directed at a server, a method, and a computer program product can all be construed by the examiner as having the same actual contribution. To reduce risks associated with adverse claim constructions, it may also be helpful also to submit claims sets directed at means plus function and method claims in additional to system claims.
Drafting AI Patent Applications to Overcome the Patent Eligible Subject Matter Hurdle
Both the AI guidance from IPOS and the USPTO underscore that patent eligible subject matter should be a primary consideration that extends beyond the drafting of the claims. The actual contribution of various embodiments of an AI invention can be embedded throughout the specification to help support potential amendments made during prosecution of the patent. Actual contribution can first be introduced within the Abstract of the patent application. More extensive technical detail can thereafter be provided within the Detailed Description and figures of the patent application.
This technical detail may include identifying the technical problem and its solution, specifying the data inputs and outputs, and listing any physically controlled electrical components used in the invention's embodiments. Further, the linkages (physical or causal) between the technical elements can be illustrated within one or more figures of the AI patent application. Sample source code and experimental results, which can often be difficult to obtain from an inventor during the early stages of the product development cycle, when patents are typically drafted, can also be included. Adding this extra level of detail into the application draft can mean the difference between securing a patent grant for an AI-related invention at a modest prosecution cost or undergoing the expense of repeated office action responses ending only in abandonment of the application. Front-loading extra efforts prior to the filing of the application is therefore the obvious choice.
For an example AI patent that sailed through prosecution see: Singapore Patent Application No. 10201908409R, entitled “Method of Optimizing AI/ML Predictions from a Classical Data Feed with a Hybrid Simulator Generated from Classical and Quantum Model Structures”. The application was filed on 2019-09-11 with a request for prosecution under Singapore’s Accelerated Initiative for Artificial Intelligence and received a notice of eligibility to grant a mere 22 calendar days later on 2019-10-03. This patent application includes sample source code and experimental results for various sample configurations, wherein the experimental results demonstrate improved performance of the invention.
- Written by Patrick W. Duncan, Senior Patent Associate at Drew & Napier LLC
This posting is not intended to constitute or to be relied upon as legal advice. You should consult Singapore legal counsel if you require legal advice regarding Singapore patent practice.