This article was originally published in Landslide, in September/October 2019.
Artificial intelligence (AI)—the use of various methods and algorithms to simulate cognitive functions in machines—has its origins dating back to the 1950s. In recent years, the confluence of advancements in computational power and the abundance of rich data sets has led to a surge in the number of practical applications for AI. The time and money required to properly train, test, and deploy AI in these applications makes protecting them using patents a crucial consideration in any business plan.
Unsurprisingly, then, according to the World Intellectual Property Organization, there has been an "AI patent boom"—of the 340,000 AI-related patent applications that have been filed globally since the 1950s, over half have been published in the last seven years.1 The number of Canadian filings for AI-based inventions has increased during this time as well. While some of this increase in Canadian filings may simply be a result of Canada having a technologically advanced, G7 economy, some may also be a result of Canada's playing a leading role in the development of AI-based technologies generally.2
This article discusses issues relating to patentability of such inventions in Canada and contrasts the position of the Canadian Intellectual Property Office (CIPO) with that of the United States Patent and Trademark Office (USPTO) and the European Patent Office (EPO).
In Canada, subject matter eligibility is governed by section 2 of the Patent Act, which defines an "invention" as "any new and useful art, process, machine, manufacture or composition of matter, or any new and useful improvement in any art, process, machine, manufacture or composition of matter."3 Correspondingly, section 27(8) of the Patent Act excludes from patentability "any mere scientific principle or abstract theorem."4
The earliest Canadian court decision relating specifically to eligibility of computer-implemented inventions is the Federal Court of Appeal (FCA) decision in Schlumberger, which held that merely implementing an algorithm that itself was not patentable using a computer did not render that algorithm patentable.5 More particularly, the FCA held that the mathematical formula to be performed, as claimed, was the only inventive aspect of the invention and was a mental operation that qualified as a "mere scientific principle or abstract theorem."
In a later case, Canada (Attorney General) v. Amazon.com, Inc., relating to Amazon's one-click online ordering system, the FCA mandated that the statutory subject matter analysis performed by the Commissioner must involve a "purposive construction of the patent claims."6 The court further noted that patentable subject matter must be "something with physical existence, or something that manifests a discernible effect or change."7
Subsequent to Amazon, the CIPO issued two practice notices on March 8, 2013, which provided guidance relating to examination of computer-implemented inventions.8 As part of the eligibility analysis, the practice notices require purposive construction to be performed to determine the essential elements of a claim using a problem-solution approach. Where a claimed computer element is considered to be essential to solving a problem addressed by the invention in the patent application, then the claim is directed to statutory subject matter.9 The practice notices note that one factor to consider in determining whether a claim reciting computer elements is statutory is to determine whether the claimed elements address a "computer problem"—that is, a problem with the operation of a computer, as opposed to a problem in which the computer merely comprises part of an operating environment.
Canada's Manual of Patent Office Practice (MOPOP) provides additional guidance for drafting and prosecuting software-related inventions. For example, a method of "controlling a computer's operations to achieve a technological result" would be considered statutory. Further, a claim element that considered in isolation is common general knowledge is not necessarily precluded from being an essential element of the claimed invention.10 The MOPOP notes that where it is explicit in the patent application that the nature of the problem is a computer-related one, then examination should generally proceed on that basis, and a "significant focus in the description" regarding details of the solution can assist in identifying the problem.11
From a Canadian perspective, in order to improve the odds of avoiding or overcoming a subject matter eligibility rejection directed at an AI-based invention, the description should ideally describe the technical problem solved by the inventors. Preferably, the technical problem can be characterized as a "computer problem" (e.g., the use of a neural network to more quickly classify images compared to conventional, non-AI implementations). Moreover, providing a significant amount of technical detail in the description regarding the hardware and software used to solve the technical problem may be helpful in establishing that such hardware and software is essential, and that a claim that recites them is accordingly patent eligible.
Application of the Case Law
In Canada, once a patent examiner concludes that examination has reached an impasse, the examiner may issue a final rejection, which automatically refers an application to the Patent Appeal Board (PAB) for further consideration. The PAB reviews the application and issues a "Commissioner's Decision" in respect of patentability.
Commissioner's Decision No. 133912 relates to a patent application for applying and adapting a predictive model such as a neural network for automating detection of fraudulent activity in financial transaction processing such as credit card transactions rather than prior art mathematical models based on parameter analysis. Claim 1 of the subject application included features such as training a predictive modeling means with data (e.g., consumer profiles and past fraud-related variables).
In this decision, the PAB considered whether the computer limitations (e.g., a computer-implemented process and a computer-controlled transaction processing system) were essential features of the claimed solution. The PAB first defined the practical problem as how to adapt a predictive model (or neural network) to improve credit card fraud detection and the solution as providing a specific derived training data set that can be applied to a neural network predictive model algorithm. The PAB noted that the computerized features were "material to the operating environment of the conventional fraud transaction system, but not essential to the solution of providing a training data set that can be applied to a neural network.13 The PAB commented that although neural networks were typically executed using computers, "needing a computer for practical convenience . . . does not make the computer essential for the working of an invention.14 Moreover, where the claim does not define a solution to a computer problem "or overcome any technical problem in the operation of the computer system," then the computer is likely being used as a matter of convenience to perform calculations.15 The PAB found that the computer limitations of the claim were "primarily to perform the neural network calculations in an expeditious and efficient manner" and consequently provided only "a convenient working environment".16 Accordingly, the PAB concluded that the computer limitations were inessential, that the essential elements of the claims defined only data processing and mathematical calculations, and that the claims were ineligible.
As indicated by the above Commissioner's decision, avoiding and overcoming a subject matter eligibility rejection can be difficult for an AI-based invention, in part due to black box implementation of AI-related tools such as neural networks, which can make drafting a detailed description challenging. Notwithstanding this, the CIPO has recently issued many patents directed to inventions that practically apply AI, and techniques exist for drafting and prosecuting applications to increase the likelihood that they will be found allowable. For example, in the context of a neural network, the description may include various types and configurations of the layers of the neural network; the effect of different types of training data on network performance and training times; and the advantageous performance of a neural network-based implementation as opposed to conventional, non-AI implementations. Including information such as this in an application and emphasizing these points during prosecution can help establish eligibility.
In view of the number of Canadian patent applications that are Patent Cooperation Treaty (PCT) national entries, the drafting and prosecution of an application is not done in isolation. Rather, Canadian prosecution often occurs in a broader context of a number of international filings. Therefore, how Canadian practice compares to practice in other jurisdictions often is relevant to patent applicants. In particular, the non-Canadian applicants who file the most in Canada are American and European applicants.17
In January 2019, the USPTO published updated patent eligibility guidance (2019 PEG) applicable to computer-implemented inventions, including inventions applying AI.18 Under the 2019 PEG, an applicant has three opportunities to establish that a claim is patent eligible: (1) by establishing that a claim does not recite a judicial exception to patentability, which in the context of AI is likely to be an abstract idea; (2) by establishing that a claim that does recite a judicial exception nonetheless integrates that exception into a practical application; and (3) if a claim recites a judicial exception that is not integrated into a practical application, by establishing that the claim provides an inventive concept that is sufficient to establish eligibility.
Canadian purposive construction holistically embodies several of the same principles that the 2019 PEG does. For example, when asserting that a claim recites a judicial exception, the USPTO is required to "[i]dentify the specific limitation(s) in the claim under examination . . . that the examiner believes recites an abstract idea".19 Similarly, as discussed earlier, in Canada an examiner is required to purposively construe the specific limitations in a claim and, in order to find that a claim is ineligible, must conclude that the claim's essential elements do not recite computer equipment.
Additionally, when assessing whether a claim integrates a judicial exception into a practical application, the USPTO considers certain factors identified in Federal Circuit jurisprudence. One of those factors is whether integrating the judicial exception improves the functioning of a computer or another technology or technical field. The 2019 PEG specifically recites that the integration needs to do more than "generally link the use of the judicial exception to a particular technological environment."20 Similarly, under Canadian practice, the CIPO considers whether the applicant's purported solution is technical in nature, and whether it solves a "computer problem" as opposed to a problem in which the computer simply is part of the operating environment. If it does, the claim is likely to be found eligible.
Further, when assessing whether a claim provides an inventive concept, the 2019 PEG states that limitations that "simply append well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality," will fail to establish eligibility.21 In Canada, the CIPO is similarly likely to find analogous claim limitations as representing inessential post-solution activity and accordingly insufficient to establish eligibility.
Concurrently with the 2019 PEG, the USPTO published updated guidance, specific to computer-implemented inventions, directed at written description, enablement, and functional claiming.22 In this regard, Canadian practice is more generous than U.S. practice. Canadian law does not have an analog to 35 U.S.C. § 112(f), and accordingly even explicit "means for" language in a Canadian claim does not limit an applicant to corresponding implementations found in the description. Similarly, failure to provide a structural description in Canada of claim elements described using "means for" language does not necessarily invalidate a patent so long as the invention remains enabled. And, while Canadian practice does recognize certain issues that fall within the scope of the written description requirement (e.g., claims that are invalid for overbroadly claiming a genus from example species), it does not have as extensive a body of jurisprudence that gives life to a written description requirement distinct from the enablement requirement.
The EPO in the fall of 2018 also published updated guidance specific to examining AI inventions.23 The European guidelines note that AI and machine learning "are based on computational models and algorithms . . . such as neural networks [and] support vector machines . . . [and] are per se of an abstract mathematical nature."24 The European guidelines note, however, that specific applications of these otherwise abstract models may result in an invention having sufficient technical character to be patentable. While this framework does not directly translate to Canadian practice, generally claiming tools used to implement AI inventions such as neural networks without specifying how those tools are applied can result in eligibility issues in Canada. When the CIPO assesses an applicant's problem and claimed solution during purposive construction, a claim that does not recite a solution sufficiently may be broad enough to encompass statutory and nonstatutory solutions to the identified problem. However, the claim can often be narrowed during prosecution, and typically without the new matter issues common in European practice, so as to more particularly claim a technical solution directed solely at patent-eligible embodiments.
As is occurring in other countries with technologically advanced economies, Canada is seeing an increase in the number of patent filings directed at AI-based inventions. Examination of AI-based inventions in Canada involves determining which elements of a claimed invention are essential to solving a problem identified by the CIPO during prosecution. Where a claim's essential elements include computer equipment, the CIPO will find the claim patent eligible. Factors that weigh in favor of a claim being found statutory overlap with analogous factors the USPTO and EPO apply. For example, drafting claims that solve a technical problem such as a problem related to computer operation and supporting them by including in the description significant details regarding the invention's technical underpinnings help to maximize the likelihood that a claim will be found patent eligible. The CIPO will also consider any statements, prejudicial or favorable, contained in an application's description; accordingly, expressly offering that an invention may be performed manually is unwise, while emphasizing an invention's technical benefits may be helpful during prosecution. With properly drafted and prosecuted patent applications, applicants can expect to receive patents for meritorious AI-based inventions in Canada.