On November 1, the European Patent Office’s (EPO) updated Guidelines for Examination went into effect. Of note, the Guidelines include a new subsection on “artificial intelligence and machine learning.” This is the latest milestone in a recent world-wide wave of interest in patenting in the field of artificial intelligence. However, the legal framework for patenting such inventions is uncertain, evolving, and not uniform across the globe. This post addresses the current state of artificial intelligence patenting in Europe and the United States in particular, and offers key takeaways that practitioners should consider when drafting and prosecuting patent applications in this field.
Background on Artificial Intelligence and Machine Learning
For context, artificial intelligence (“AI”) may be summarized as the simulation of intelligent human behavior by machines. A subcategory of AI, machine learning (“ML”), refers to ability of systems to learn from data and improve from experience automatically—in other words, without being explicitly programmed. In practice, the beneficial results delivered by AI and ML are rooted in algorithms and mathematical models. These features, however, have generally been excluded from patentability in both Europe and in the United States. While AI and ML hold promise as the next breakthrough technology, this legal precedent raises concerns about the ability to secure and maintain patents in this field.
Europe
In Europe, patentability of computer implemented inventions is determined using an intricate set of rules. To qualify for a patent, an invention must first have a “technical character,” in that it relates to a technical field and is concerned with a technical problem. However, this requirement is readily satisfied by claims that recite, for example, a computer or other device. The more critical hurdle for patentability, particularly for computer-implemented inventions, is that the claimed subject matter must also have an “inventive step” over the prior art—one that is found based only on features that “contribute” to the technical character. Under the EPC, however, computer programs and mathematical methods, considered on their own, are ordinarily excluded from qualifying as such features.
Importantly, the new EPO Guidelines state that AI and ML “are based on computational models and algorithms” that “are per se of an abstract mathematical nature.” Not surprisingly, the newly-added subsection on AI and ML falls under the broader section on “mathematical methods.” Thus, like other mathematical methods, AI and ML inventions are also excluded from patentability when claimed as such. The Guidelines also urge examiners to pay careful attention to use of the terms “neural networks,” “reasoning machine,” and “support vector machine” when examining whether the claimed subject-matter has a technical character as a whole; the EPO cautions that these AI industry buzzwords “usually refer to abstract models devoid of technical character.”
The Guidelines cite authority for the notion that a mathematical method “contributes to the technical character of a computer-implemented method only in so far as it serves a technical purpose.” The Guidelines thus indicate that, as with other inventions based on mathematical methods, an AI/ML invention could be patentable if claimed such that the underlying algorithm or computational model serves a technical purpose. The Guidelines further provide exemplary real world applications for AI and ML, whose purposes have been deemed to be technical.
As an example of a patent-eligible AI invention, the Guidelines cite “the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats.” On the other hand, “classifying abstract data records or even ‘telecommunication network data records’ without any indication of a technical use being made of the resulting classification is [] not a per se technical purpose,” regardless of any “valuable mathematical properties such as robustness.” These examples, previewed by the EPO at its May 30 Conference on Patenting AI, stem from actual Boards of Appeal decisions.
United States
In the United States, the patenting of computer-implemented inventions is also evolving and thus somewhat unpredictable. And unlike the EPO, the U.S. Patent and Trademark Office (USPTO) has not issued any examination guidelines specific to AI inventions. In practice, whether a patent is ultimately granted on a computer-implemented invention in Europe or in the U.S. often hinges on the same key considerations; but while some of these considerations are assessed as a matter of “inventive step” in Europe, they are instead assessed as a matter of patent eligibility in the U.S.
Eligibility for U.S. patents is assessed according to the Supreme Court’s two-part Alice test, and further guided by post-Alice Federal Circuit cases. Importantly, “abstract ideas” (one of the judicially-created patentable subject matter “exceptions” addressed in Alice) are not to be confused with improvements to computer-related technologies, which may be patentable if appropriately claimed. However, if a claimed process is capable of being performed without a computer, the Federal Circuit has indicated that it cannot improve computer related technology.
How this framework actually plays out for each AI-related invention will hinge on the particular claim at issue. In one recent case, the Northern District of California found that a patented claim to ML-driven predictive analytics was directed to “the abstract concept of the manipulation of mathematical functions and make[s] use of computers only as tools, rather than provid[ing] a specific improvement to computer related technology.” The Federal Circuit affirmed this decision earlier this month, but did so in a single-line ruling that provides no actual guidance.
In April, USPTO Director Andrei Iancu addressed the topic of AI patenting during testimony before the Senate Committee on the Judiciary. Director Iancu acknowledged the confusion stemming from Supreme Court decisions such as Alice, and indicated that USPTO is working on guidance aimed at providing some clarity. Going a step further, the Director also expressed the view that algorithms (including those that form the basis of AI) are the result of human ingenuity and are thus “very different” from mathematical equations that simply represent naturally occurring phenomena.
Observations and Analysis
Given the many similarities in legal frameworks for patenting in the U.S. and in Europe, many AI-related inventions are likely to face similar odds of patent allowance under either regime. For example, if an AI-related invention is articulated as a specific improvement to computer-related technology, then it is likely both statutorily eligible for a U.S. patent, and also contributes technical character as required for a European patent.
One possible distinction in AI patenting in Europe versus in the U.S. is the potential impact of limiting the recitation of a mathematical method to a specific technical purpose or application. Under the EPO Guidelines, such a limitation may contribute to the technical character for purposes of the “inventive step” assessment, and thereby possibly result in allowance. A different outcome might occur under U.S. law because of the heightened emphasis given to patent eligibility. USPTO guidance provides that “generally linking the use” of an abstract idea, law of nature, or natural phenomenon “to a particular technological environment or field of use”—for example, a claim limiting the use of a mathematical formula to the petrochemical and oil refining fields—is an example of a limitation that courts have held insufficient to qualify as an “inventive concept” under the Alice test for patent eligibility.
Director Iancu’s testimony before the Senate—distinguishing AI algorithms from mathematical equations—diverges from the EPO’s declaration of AI algorithms as per se of an abstract mathematical nature. In view of the post-Alice case law, however, the Director’s comments appear to reflect his own perspective and therefore should not be taken as actual USPTO policy. Nonetheless, the USPTO has signaled that it will issue additional guidance to help examiners categorize and practically apply the judicial exceptions to patentable subject matter based on the case law. This forthcoming guidance should prove to be a helpful step toward providing much needed uniformity and predictability for the patenting of AI-related inventions.
Takeaways for Practitioners
The patenting of AI-related inventions remains uncertain and evolving, and patent practitioners are well-advised to keep abreast of the latest developments in the U.S. and Europe, as well as other key jurisdictions, in order craft a robust patent portfolio strategy. The USPTO is hosting an upcoming conference titled “Artificial Intelligence: Intellectual Property Policy Considerations.” The event is expected to include a discussion on how AI-related inventions can be protected, and may provide an update on any forthcoming guidance that pertains to the patenting of AI-related inventions.
The following practices may be generally helpful for patenting in this area:
- Strive to spell out, explicitly in the claims, the specific technical solution that is the subject of the AI-based invention, guided by the recent examples provided by EPO;
- Make the technical purposes of the invention clear, again guided by the EPO’s new Guidelines;
- Where applicable, ensure that the specification characterizes the AI-related invention as providing an improvement to a computer-related technology (e.g., providing an efficiency or other performance metric in the computer running the AI-related invention, as compared to a computer running previously known solutions);
- Ensure that the specification details the technical solutions offered by the invention, but note that such descriptions must sufficiently correspond to the claimed language; and
- Avoid overreliance on commonly-used AI buzzwords such as “neural network,” “reasoning engine,” and “support vector machine.”