“The extra attention that the EPO and USPTO are paying toward AI and ML will likely help swing the pendulum of patentable subject matter toward a place that is in harmony with the current state of technology.”
It is safe to say that Artificial intelligence (AI) and Machine Learning (ML) are hot topics and, as with any rapidly growing technological area on the industry side, there is also a rapidly growing number of patent applications being filed.
In view of this, the European Patent Office (EPO) issued new guidance for examination for AI and ML patent applications in November 2018. Meanwhile, in January 2019, the U.S. Patent and Trademark Office (USPTO) also issued revised guidance directed to what constitutes patent eligible subject matter under 35 U.S.C. §101. Although the USPTO’s revised guidance is more generally directed to software applications, at least one of the accompanying hypothetical examples (Example 39) is directed to the AI and ML space.
Additionally, in a recent conference held at the USPTO, “Artificial Intelligence: Intellectual Property Considerations,” USPTO Director, Andrei Iancu, said:
“As director of the USPTO, one of my top priorities is making sure the United States continues its leadership when it comes to innovation, especially in the emerging technologies of the future, including AI and machine learning. To that end, we must harness our long history of innovation, born of our nation’s founding document and perpetuated by our people’s innovative spirit, and apply the same spirit to AI technologies.”
Therefore, while there may be lingering concerns that AI and ML inventions will face extra scrutiny toward patentability due to their software-centric nature, the extra attention that the EPO and USPTO are paying toward AI and ML will likely help swing the pendulum of patentable subject matter toward a place that is in harmony with the current state of technology. The below analysis reviews the recent developments by the EPO and the USPTO to provide specific guidance on the topic of AI and ML.
The EPO’s Take
The EPO’s basic take is that subject matter that “is directed to a purely abstract mathematical method and … does not require any technical means” is not suitable for patent protection. However, “[i]f a claim is directed either to a method involving the use of technical means (e.g. a computer) or to a device, its subject-matter has a technical character as a whole and is thus” eligible for patent protection.
Further, the EPO explains that AI and ML are based on computational models, which “are per se of an abstract mathematical nature, irrespective of whether they can be ‘trained’ based on training data.” Hence, the EPO instructs its examiners to “carefully” look at certain expressions within a claim (i.e., trigger words), such as “support vector machine,” “reasoning engine,” or “neural network,” because they “usually refer to abstract models devoid of technical character.”
According to the EPO, the key for patent eligible subject matter is whether such subject matter is tied to something “technical” or has a “technical purpose.” As an example, the EPO explains that “the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats” and the classification of “digital images, videos, audio or speech signals based on low-level features (e.g. edges or pixel attributes for images)” have a technical purpose or application, and would thus be eligible for patent protection. By contrast, “[c]lassifying text documents solely in respect of their textual content” does not serve a technical purpose, but rather a linguistic one. The EPO also indicates that “[c]lassifying abstract data records or even ‘telecommunication network data records’ without any indication of a technical use being made of the resulting classification is also not per se a technical purpose.” As such, these types of classifications would not be eligible for patent protection.
The USPTO’s Take
The revised guidance revises Step 2A of the Alice–Mayo framework into a two-prong approach. First, a determination has to be made whether the subject matter is directed to an “abstract idea,” by deciding whether the subject matter falls into one of three categories – (1) mathematical concepts, (2) mental processes, and (3) certain methods of organizing human activity. If the subject matter does not fall into one of these categories, then the claim is patent eligible. Otherwise, the claim is directed to an “abstract idea” and prong two must be evaluated – whether this “abstract idea” is integrated into a “practical application.”
To satisfy prong two, the claim must recite “an additional element or a combination of additional elements … [that] apply, rely on, or use the … [abstract idea] in a manner that imposes a meaningful limit.” Limitations that are indicative of integration into a “practical application” include, among others, improvements to the functioning of a computer or to another technology or technical field, use or application of a particular machine, effecting a transformation or reduction of a particular article to a different state or thing, and applying or using the abstract idea in some other meaningful way beyond generally linking it to a technological environment.
If the determination is made that there is no “practical application,” Step 2A fails, and the analysis moves to Step 2B, in which a determination of whether the claim recites “significantly more” is performed. Note that this analysis has not been modified by this latest guidance from USPTO.
As mentioned above, although the USPTO does not explicitly tailor its new guidance to the AI and ML field, it is clear that the revised guidance may, in many cases, be applied to claims within this field. This is apparent from hypothetical Example 39 accompanying the issuance of the USPTO’s revised guidance. All of the new hypothetical examples can be found here.
The claim in Example 39 is directed to a computer-implemented method of training a neural network for facial detection. This method comprises the following:
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.
In applying the aforementioned analysis, the USPTO concludes that this claim is eligible for patent protection under Step 2A, prong one, because it is not directed to an “abstract idea.” The USPTO explains that the claim does not recite mathematical relationships, formulas, or calculations (although some of the limitations may be based on mathematical concepts), the claim does not recite a mental process because “the steps are not practically performed in the human mind,” and the claim also does not recite organizing human activity such as a fundamental economic concept or managing interactions between people.
Undoubtedly, there are challenges to obtaining patents in the AI and ML space. Both the EPO and the USPTO appear to acknowledge these challenges and provide some relatively broad guidance on what may or may not be eligible for patent protection. Of course, every application is analyzed on a case-by-case basis. What is clear is that AI and ML claims before the EPO and USPTO must be more than just mathematical concepts, and must have some type of “technical purpose” or “practical application.”
To be able to achieve this requirement, it is paramount that careful consideration be given when initially drafting the specification. It is helpful to discuss a variety of technical applications with different levels of detail. Real-world examples of how the invention would actually be implemented may help solidify the practical applications as well as show the advantages or improvements over previous known methodologies.
The claims can only be as good as the specification will allow. Thus, the more description in the specification, the higher the chance of having enough backup material to amend the claims in such a way as to render them patent eligible. Furthermore, it is important to consider word choices in the claims as these can lead to additional scrutiny. For example, in view of the USPTO’s guidance and the various accompanying examples, avoiding mathematical equations in the claims may be good practice, although claim features may still be based on mathematical concepts. Including equations in the claims may trigger extra scrutiny from examiners on both sides of the Atlantic as well as unduly narrow the claims. However, it is important to note that reciting a mathematical equation in a claim does not automatically render it patent ineligible.
Taking a look back at the aforementioned Example 39, we know that the USPTO finds the claim to be eligible for patent protection as it is not directed to an “abstract idea.” However, what would the EPO’s opinion of this claim be in view of the new guidance? Well, for one, the EPO would look at this claim more “carefully” as it contains expressions such as “neural network.” However, similar to the example of the neural network being used in a heart-monitoring apparatus, Example 39 uses a neural network for facial detection, which suggests a technical purpose.
Although Example 39 is directed more to the training of the neural network than the use of such network in facial detection, “[w]here a classification method serves a technical purpose, the steps of generating the training set and training the classifier may also contribute to the technical character of the invention if they support achieving that technical purpose.” Thus, it is possible the EPO may find Example 39 eligible for patent protection if the recited training of the neural network supports achieving the technical purpose. Of course, this claim may receive extra scrutiny at the EPO as it is directed to concepts the guidelines were specifically issued to address. Thus, Example 39 may have an easier time before a USPTO examiner than an EPO examiner.
All in all, it remains to be seen how the guidelines will be applied to AI and ML inventions, both in Europe and the United States.