AIPPI World Congress 2019: AI around the world

While the number of patent applications relating to artificial intelligence (AI) has soared in recent years, there is still some confusion around the patentability of the technology, as lawyers discussed at the AIPPI World Congress.

In China there is no official definition for AI but rather a very broad one, which complicates patent application, Lili Wu, a partner at Han Kun Law in China, explained yesterday, September 17.

“You need to be able to show an improvement on a method of data collection, or a method of implementation of data in order to get a patent,” she said.

In China, most patent filings in relation to AI are by internet and smartphone companies. Other popular filers are universities, especially state-owned institutes.

“In China, when we talk about AI patentable subject matter, there are two main hurdles,” Wu said.

First, if the patent claims are “pure algorithm” then the patent will not be registered.

The second hurdle is defining the invention, which must pass a three-part technical test. This means the invention must identify a technical problem, must have a technical means and must have a technical effect.

“There needs to be a connection among these elements. The examiners need to see that the technical effect is linked to the function of all of the invention’s features,” she said.

Wu said patents are most commonly granted when the invention is used in industry, such as in manufacturing, or if the invention is used to process external data like video and audio.

It is a similar picture in the US. Kate Gaudry, a partner at Kilpatrick Townsend, said that for an AI invention to be patented, the innovator needs to be able to identify the problem they are trying to solve, and the solution to that problem.

“But both of these must meet a certain standard of innovation, as you are likely to face prior art objections if you try to register a patent that applies AI to solve quite a general problem,” Gaudry said.

Similarly, pure algorithms will not be accepted in the US. Currently, in the US, machine learning is on the border of patent-eligible subject matter.

But, Gaudry said this may change once more AI patent cases are tried before courts.

Vanessa Bailey, senior IP policy counsel at Intel, speaking on her own behalf, said the US government should help build on innovation by committing to a national AI strategy.

Currently in the US there approximately 15 legislative bills on AI, but none of these address IP she said, which means there isn’t enough clarification on important aspects of the technology.

“The hardest issue for me to get my head around is who is the person with ‘the ordinary skill in the art’ when it comes to AI? Is it the skilled inventor?” she asked.

“Or, if you want to say the inventor is the AI, then is the skilled person another AI? This is something we need to figure out.”

John Brunner, a patent attorney at Carpmeals & Ransford in the UK, said that from an EU perspective, the European Patent Office has been relatively consistent in its approach to AI patents.

“Similarly to China, there is no definition of an AI patent. But the standard requirements used in the assessment of patentability also apply to AI applications,” he said.

First, the invention must have a technical character.

“AI inventions defined as computational algorithms without reference to a technical device will be excluded from patentability as a mathematical method,” he said.

But, inventors can overcome this relatively easily by including a technical device in their invention.

Having passed this stage, examiners will then assess whether the inventive step of the patent is based on a technical feature.

“This is because the AI-based features of the claim need to provide a technical purpose in line with the requirements to solve a technical problem,” he said.

He pointed out that at the the EUIPO does not like “nebulous terms devoid of technical meaning”.

In other words, adding language that broadly relates to technology but doesn’t clearly define a technical invention will not fulfil the requirements of having “technical character”, and may even lead to the application failing on lack of clarity, he said.

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:

Marques 2019: navigating social change

AIPPI World Congress 2019: Brexit and IP

Foundation hits back at Columbia University over encyclopedia ownership