Demand for AI patents is growing exponentially as the pace of innovation accelerates.
In October, Saudi Arabia became the
first country in the world to grant citizenship to a robot. “Sophia” was
created by Hanson Robotics in Hong Kong and featured at the Future Investment
Initiative event in Riyadh that month, where her citizenship was awarded. The
following month, Sophia was named the United Nations Development
Programme’s inaugural Innovation Champion, and the first non-human to be given
a United Nations title. She can follow faces, sustain eye contact, recognize
individuals and answer questions. By January she had learned to walk.
Sophia may be on the cutting edge of artificial
intelligence, but the development of Sophia, and other developments in
artificial intelligence and machine learning, also point to the challenges and
realities of patenting AI.
“In our current system, there’s no way to recognize a
non-person [such as Sophia] as an author or inventor, but in the future there
could be,” says Stephen Martin, a partner, patent and trademark agent at Ridout
& Maybee LLP in Toronto. “Star Trek-like ethical and legal questions could
come into play . . . in 10 to 15 years.”
The greatest challenge for the future, he says, “is
how we’ll start to move from fit-to-task solutions,” such as voice recognition
and image analysis, “to more general-
purpose AI systems” that simulate human activity, such as an automated chat-bot
that will dynamically pose and respond to questions from a user and “which are
themselves starting to create inventions. . . .
“The more you get away from very specific applications
. . . it gets much more challenging because you’re trying to patent something
more akin to human mental activity,” says Martin.
The challenges of patenting
AI is a “hot technology” and
significant breakthroughs in it are underway, says Isis Caulder, a partner,
patent and trademark agent at Bereskin & Parr LLP in Toronto. A number of
high-profile technology companies such as IBM, Google, Amazon, Microsoft,
Samsung and AT&T are developing novel AI algorithms at a breakneck speed,
she notes, and many other technology companies are harnessing AI specifically
to make their existing products better and more useable and intuitive: for
example, self-driving cars, household robots and image recognition.
“What is interesting from a patenting point of view is
that AI is generating new technical problems to be solved, and these technical
solutions are fertile ground for patenting,” she says. The U.S. Patent and
Trademark Office is issuing “an exponentially growing number of patents,”
mostly to high-profile technology companies like those mentioned above, that
are developing AI-related algorithms.
As with any other invention, an AI or software-based
invention must be new, useful and non-obvious in order to be patentable, says
Caulder. “The fourth rule is subject matter: Is it too abstract or is it
something concrete enough to be patentable? It must be applied enough and
crystallized enough to be patentable subject matter.”
The importance of showing the technicality of the
invention in the patent application is key, IP lawyers acknowledge. The more
technical the problem and solution that’s being addressed by the patent
application appears to be, the greater the likelihood that the Canadian
Intellectual Property Office will conclude that the AI-related invention should
be eligible for a patent, says Matthew Zischka, partner in Smart &
Biggar/Fetherstonhaugh’s Toronto office.
“The applicant has to convince the [patent] office
that the computer is necessary or substantial for solving the particular
problem that’s being addressed by the invention,” Zischka says. “The patent
office is [now] looking more carefully at applications to ensure that
computer-implemented inventions aren’t mere schemes, formulas or business
methods, but that they amount to more technical advances in the computer
Depending on the nature of the machine-learning
algorithm, the patent office might determine that it amounts to a mere
mathematical formula rather than an invention, he says. Alternatively, the
office could conclude that the computer-implemented invention is a
computer-implemented “business scheme” and, therefore, not patentable. At the
same time, the patent office is open to granting patents for machine-learning
inventions, he says.
In 2014, the United States Supreme Court released its
decision in Alice Corp. v. CLS Bank International, which held that a computer
implementation of an abstract idea, which is not itself eligible for a patent,
does not by itself transform that idea into something that is patent-eligible.
“In the U.S. in particular now, if an examiner or a
court can characterize an invention as being an application of known
mathematical algorithms or mathematical operations, you may not be able to get
a patent,” says Roch Ripley, partner and head of the Vancouver Intellectual
Property Department of Gowling WLG in Vancouver.
“And if you get a patent, the courts have been
invalidating . . . software patents generally, at much higher rates. That’s a
Europe used to be a jurisdiction in which it was more
difficult to get a software patent because the courts “applied a
technical-effect test,” Ripley says. “Did your invention result in a technical
effect?” Now, he says, the European test is easier to satisfy than the American
Under the subject matter framework laid out by the
U.S. Supreme Court in Alice, courts have been denied patent
protection and invalidated patents, Caulder agrees, for software-implemented
inventions on the basis that they are directed to a law of nature, natural
phenomenon or an abstract idea and that the claims do not recite “significantly
more.” However, she adds, a number of Federal Circuit cases following Alice have
provided an avenue to address concerns over patent rejections.
Zischka calls the Canadian patent office approach “a
little different.” In the Alice decision, he says, the U.S.
Supreme Court applied a four-step test that was difficult to meet. But Canada’s
approach is to look at the claim that’s presented in the patent application “as
a whole. . . . Does the solution require tangible computer elements? If it
does, they generally consider that the claim is patent-eligible.”
Ridout & Maybee’s Martin cites
infringement detectability as a significant factor in patent protection. Data
services are now central to our economy, and AI is beginning to move in on
that, he says. “Data sensors are becoming omnipresent. Most patent agents have
to be aware of this technology” as it develops in every industry.
The best patents are the ones that will be infringed
upon by competitors and are detectable, Martin points out.
There are administrative procedures that have been
introduced since 2013 called post-grant review and inter partes
review that give alleged infringers avenues to invalidate patents, Ripley says.
If a case for patent infringement goes to court, defendants can also try to
invalidate a patent using a summary procedure on the basis that it’s directed
at an ineligible subject matter: an abstract idea. Procedures available to make
a summary judgment change “the economic consideration,” he notes.
“When we draft [patent applications] now,” says
Ripley, “we’re careful to include a lot of technical detail, because if you can
really establish that the invention results in technical benefits, then you’ve
got a much better chance of surviving any invalidity challenge.”
prospective patent holders
The patent application presented
before the Canadian patent office can’t read like a sales document but like a
technical document providing some sort of solution to a problem that’s technical
in nature, Zischka says.
“We impress on [clients] ‘we won’t start working on a
patent application for you unless there’s real technology there,’” says
Caulder. “You want to see if what you’re building is creating a useful,
concrete and tangible result. What is the result of the machine you’ve built?
AI or machine learning involves a training model, developing and then running
it and showing results.”
Canada is a hub of AI activity, but the valuable
patents and intellectual property will come from the use of AI to solve a
specific commercial need, Martin says. He also advises his clients to imagine
how they would feel if a competitor applied to patent an invention similar to
one they were developing. Hopeful patent holders should “get things filed as
soon as possible, before technology moves on and you get leapfrogged.”
“I try to really dig into the details,” says Ripley. “You
want to be careful not to just abstract the invention at a black-box level if
you can avoid it. The more low-level discussions you can include, I think, the
better for the purposes of validity.”
The overall message of patenting AI, he adds, is that
“it’s gotten more difficult over the last few years, and you need to take more
care in drafting a solid application. That’s how I see it.”