Patenting AI – overcoming the natural language processing barrier

Artificial intelligence (AI) is transforming daily life with real-world applications such as speech recognition, smart cars, robots and health care. Patent offices worldwide continue to issue patents for AI innovations in ever-increasing numbers. Within the AI patent field, Natural Language Processing-related patent applications are the most prevalent type of AI specialization in both Canada and the U.S., according to the CIPO report on Processing Artificial Intelligence: Highlighting the Canadian Patent Landscape.

A key challenge to patenting an NLP-related invention is convincing the Patent   Office that the claimed invention is patentable subject matter.

In Canada, a software invention is considered to be patentable subject matter if computer components can be considered essential to a claimed invention using a problem/solution analysis. In the U.S., an invention needs to contain “significantly more” than just “an abstract idea,” which is determined in part based on whether the idea is “integrated into a practical application.”

A patent application in the name of Fair Isaac Corporation related to automated detection for fraudulent financial transactions was considered by Canada’s Patent Appeal Board (PAB). The PAB found that the computer primarily performed neural network calculations in an expeditious and efficient manner but that it was not material to the solution. The PAB concluded that the computer implementation was not an essential element and that the invention was not patentable subject matter.

In the U.S., a patent application in the name of eBay described a conversational agent designed to simulate a conversation with a user to gather listing information. The Patent Trademark and Appeal Board (PTAB) found that the claims were directed to an abstract idea and that the claims did not contain “significantly more” as the conversational agent did not need to be implemented by AI and, accordingly, that the invention was not patentable subject matter. After the PTAB’s decision, the    applicant added claim limitations relating to combining posting information with a user-uploaded image, which resulted in allowance.

In another U.S. PTAB case, a patent application in the name of IBM covered data processing for iterative deepening knowledge discovery using time-weighted closures based on dimensions of evidence. The PTAB found that the claims recited an abstract idea of generating candidate answers to a question but that the abstract idea was integrated into a practical application and, accordingly, that the invention was patentable subject matter. The PTAB found there were steps for a particular query- and hypothesis-based processing sequence and set of rules to improve the technology of question-answering systems.

A key lesson is that NLP-related inventions are more likely to be patentable in Canada and the U.S. if they can be characterized as more than mere improvements to algorithms and rather as: (1) computerized data gathering, processing and outputting; and/or (2) steps for a processing sequence to improve the technology of particular NLP-related systems.

Isi Caulder is a partner with Bereskin & Parr LLP and a member of the Electrical & Computer Technology practice group. She is the co-leader of the Artificial Intelligence (AI) practice group and the leader of the Cleantech practice group. Ray Kovarik is an associate with Bereskin & Parr LLP and a member of the Electrical & Computer Technology practice group.