TPN Europe 2019: USPTO and EPO aligning on AI patentability
The challenge of obtaining patent protection for artificial intelligence (AI) inventions from IP offices across the globe dominated discussion at Technology Patent Network Europe, hosted by WIPR, in London this morning, June 20.
According to Mike Jennings, partner and patent attorney at AA Thornton in London, the European Patent Office (EPO) has provided “very clear” guidance in this respect, particularly for the first of the two-step patentability test for AI applications.
As Jennings explained, an otherwise unpatentable mathematical method can form the basis of a patentable invention in Europe if applied to a specific technical purpose.
Some of the examples provided by the EPO include the use of a mathematical method to control a specific technical system or process, such as an x-ray apparatus, or the separation of sources in speech signals.
Non-technical applications, such as word processing, linguistics and marketing, remain “too abstract” for the EPO in terms of patent eligibility, Jennings said.
While Jennings noted that the advice from the EPO had been “stable” over the past number of years, he pointed to the US as an example of where there may be significant developments for rights owners to take note of.
He cited the reform drive led by incumbent director of the US Patent and Trademark Office (USPTO), Andrei Iancu, and the efforts of Senators Chris Coons and Thom Thillis to reform section 101 of the US Code, which governs patent eligibility.
According to Jennings, the USPTO has “taken a bit of a leap” in publishing guidance for applicants on which AI-related inventions or applications of AI it thinks are patentable. This was noteworthy, Jennings said, as it came in the face of some “quite negative decisions” issued by US courts which limit the patentability of AI-related inventions.
The USPTO’s guidance, Jennings said, was striking in how similar it was to the examples of patent-eligible AI inventions published by the EPO.
Perhaps this should not be surprising, he argued, given the close level of co-operation between the constituent offices of IP5, the world’s five biggest patent offices, on dealing with the emergence of AI.
He questioned, however, whether US courts would follow the guidance published by the USPTO with respect to which uses of AI are patentable and which are not, or whether they would “go down a different path”.
As he noted, however, there is already a debate about legislative changes to section 101. According to Jennings, the conversation around section 101 indicated a “recognition that under the current US practice, there has been a failure and inadequate protection for AI”.
In his view, the “sketchy” drafts of reforms published so far have appeared similar to the general approach taken by European authorities.
Jennings welcomed the debate around potential legislative reform, noting that it indicated an “enthusiasm” for more harmonisation between the US and Europe than there has ever been in his career as a patent attorney.
The importance of data
Howard Read, senior associate at Appleyard Lees in Manchester, dealt with the challenge of getting AI-related patents past the EPO.
In Read’s view, patent owners cannot afford to ignore the increasing importance of data in preparing patent applications related to emerging technologies such as AI.
He cited a “reproducibility checklist” presented by Canadian computer scientist Joelle Pineau at the Neural Information Processing Systems Conference (NeurIPS) 2018 in Montréal, Canada. The checklist was designed to review academic papers in the field of machine learning.
Read, however, argued that many of the elements in the checklist could be of use in preparing AI-related patent applications.
The checklist includes, for any figures or tables which present results or data, a requirement to clearly explain what the data is and how it was collected. It also requires a link to a downloadable version of the data source or simulation environment, and an explanation of what data was not obtained.
While these are not yet stringent requirements required by the EPO, they were not purely academic, Read noted. Rather, he said, it is “effectively what is being required by the Japanese Patent Office” (JPO).
Under JPO rules, Read added, it is necessary to “explicitly identify correlation” between the outlined data sets and the description of a patent claim, or otherwise present test results which demonstrate that the technical effect has been achieved.
The increasing importance of data in AI-related patents, Read said, meant that it was more important than ever to involve data and computer scientists in patent applications.
This was particularly true, he said, for the many “hybrid inventions” which involved engineers and natural scientists as well as data and computer scientists.
To ensure the success of a patent application, Read said, “the whole team needs to contribute”.
Did you enjoy reading this story? Sign up to our free daily newsletters and get stories like this sent straight to your inbox.
Today's top stories:
AI as a judge and patent-destroying tool: panel discussion
USITC to investigate Amazon, Microsoft, and Samsung
