Artificial intelligence technology has been around for a long time, but has recently made advances prompting recognition as the transformative force that it truly is. While applicants have successfully patented artificial intelligence inventions for many years, the US has been more favorable than Europe for some types of AI. Here we focus on one area of difference between patent-eligibility of NLP inventions in the USPTO versus the EPO. We use board decisions for distinguishing the two jurisdictions.
As some background, machine learning is the AI technique most frequently disclosed in patents, and is included in more than a third of all AI-related patent applications.
Photo credit Aglaé Bassens et al., “Deep Learning Illustrated: A Visual, Interactive Guide to Artificial Intelligence” (August 5, 2019).
Machine learning is an AI technique ever-growing in dominance. Deep learning and neural networks are fastest growing of the lot. This is at least partially represented in patent filings, where filings of machine learning patent applications have grown annually on average by 28% from 2013 to 2016. This is notably higher than the average annual growth across all new areas of technology, which was 10% during the same time. Within this category, deep learning, used for example in speech recognition, had a 175% average annual growth rate from 2013 to 2016.
Among functional applications, computer vision is the most popular, and is mentioned in almost half of AI-related patents. The next hottest area in functional applications is natural language processing. Examples of NLP in industry include classifying documents; machine translation; search engine optimization; speech recognition; and chatbots. We focus here on patent-eligibility of NLP in this blog post.
Source: WIPO Technology Trends 2019, “Artificial Intelligence,” at 14 and 31.
In the US, machine learning applications have generally fared well for patentability purposes. Even though many of these machine learning inventions are rooted in software, and presumably vulnerable to Alice-type eligibility rejections, allowance rates have been substantially higher than other software classes. For example, comparing class 706 (Artificial Intelligence: Data Processing) to class 705 (Financial, Business Practice, Management, or Cost/Price Determination: Data Processing), we see a big difference.
One reason why these allowance rates are much lower may depend on the art unit differences. Machine learning inventions typically get assigned to the 2121 or 2122 art units whereas business method inventions get assigned to the 3620s, 3680s and 3690s. The 3600 art units are well known for applying knee-jerk Section 101 patent-ineligible rejections whereas AI art units are not as preoccupied with Section 101. Often times, Examiners in these machine learning art units see the cutting edge technology of machine learning in these applications and generally quickly grant the patents for these inventions.
But not all AI inventions are as easy to get allowed, especially depending on the jurisdiction. Take Europe, for example. The standard for patent-eligibility at the EPO is somewhat different than the US in that it requires a sufficiently technical nature (i.e., the claim must have a technical implementation or technical application). For image processing and speech recognition, this technical nature can be easily shown. But other types of machine learning tech, such as NLP, have not been so recognized as having a technical purpose.
As pointed out in this Marks & Clerk blog post, there is a historical context to difficulties in patenting some NLP technologies.
In T 52/85, the Board considered a system for automatically generating a list of expressions whose meaning was related to an input linguistic expression. The Board held that the relationship between the input and output expressions was not of a technical nature, but was instead a matter of their “abstract linguistic information content”. The Board consequently found that the claimed subject-matter was unpatentable.
In another relatively old decision, T 1177/97, the technology at issue related to machine translation. The Board again found the claimed subject-matter to be unpatentable, stating “Features or aspects of the method which reflect only peculiarities of the field of linguistics, however, must be ignored in assessing inventive step.” This statement is often quoted by examiners when applying the Comvik approach to inventions in the field of natural language processing. Although the Board in T 1177/97 also held that “information and methods related to linguistics may in principle assume technical character if they are used in a computer system and form part of a technical problem solution”, it is hard in practice to convince the EPO that a technical problem is solved by the linguistic aspects of an invention.
The US has been more favorable about the patentability of NLP technology. For example, the PTAB has recently reversed patent-eligibility rejections in a large proportion of AI applications. Results will follow. For example, one PTAB panel recently overturned a NLP-related invention as passing the two-step Alice framework. In Ex parte BAUGHMAN et al., Appeal No. 2019-000665 (PTAB Sept. 25, 2019), the PTAB overturned an Examiner’s rejection for the claims being directed to an abstract idea.
The Board, under step 2A prong 2, reasoned that the claim recites additional elements, which are outside the abstract idea, that include: “receiving, by the question answering system, a function call comprising an input question and a set of non-local context evidence in closure form.” The Board explained that the recited use of a “function call” and the use of “closure form” are particular (non-generic) software technology limitations. Specifically, the “function closure”-related software limitations recited in claim 1’s first step are integrated with the limitations that describe the abstract idea for generating answers to a question.
The Board viewed the claim holistically by stating that “[t]aken as a whole, claim 1 recites a set of steps for a particular query- and hypothesis-based processing sequence and set of rules, executed by a QA system.” Then citing McRO, the Board held that this amounts to “us[ing] the limited rules in a process specifically designed to achieve an improved technological result in conventional industry practice,” i.e., to improve the technology of QA systems.” After coming to this determination, the Board found that the claim imposes meaningful limits on the application of the recited judicial exception for generating candidate answers to a question and thus are not directed to a patent-ineligible abstract idea.
AI will continue to transform all sectors of industry and patentability standards across jurisdictions will continue to change. Patentability standards across jurisdictions should continue to evolve to balance the growing impact of AI on society. As it does so, it is important to anticipate prosecution strategy internationally with the best patent data.