Artificial intelligence (AI) and machine learning is
allowing life sciences firms to better analyse data, produce
information and more effectively contribute to drug discovery.
But in-house counsel at these companies are increasingly aware
of the implications that could have for their patent and
Patenting laws and guidance have been developed for
patenting AI and machine learning solutions themselves. The
same cannot be said for AI-derived inventions. As life sciences
companies increasingly evolve into biotech firms, many are
beginning to consider how they might commercialise the new
tools of their trade as well as the products they help
Under current laws, the question of inventorship and
patentability is framed in terms of human creation –
particularly in the US, which only recognises individuals as
inventors and not companies or machines.
But as AI becomes more capable of creating its own
pharmaceutical inventions, and drug companies increasingly want
to use the technology for that purpose, that necessity is
raising questions among in-house lawyers about who would be an
AI-derived invention’s inventor for the purposes
of a patent application.
The head of IP in oncology at a European pharmaceutical
manufacturer tells Managing IP that innovator drug companies
are already questioning whether an inventor must be a person,
or if patent laws need to allow for a legal entity, such as a
corporation, to be the inventor of an AI-derived invention.
"Presumably, the inventor will have to be more than the
person who pushed the button to start the computer program," he
A list of possible human inventors could include the AI
software of hardware developers, the medical professionals who
provided the input for AI’s development or those
that reviewed the AI results and recognised that an invention
had been made.
"A person who identifies a problem, but does not
contribute to the solution, is not an inventor"
That uncertainty is a problem because this aspect of the
patent application could impact the quality of the patent and
even become a focus point for invalidity proceedings in the
He adds that there have likely already been patent
applications where this matter of ownership has come up, and a
person has been named as the inventor for convenience to avoid
a human-inventor issue at the patent office.
If the laws change, however, and AI becomes registrable as
the inventor of a drug, the question of inventorship could
become more complex.
"A person who identifies a problem, but does not contribute
to the solution, is not an inventor," says the IP head of
oncology. "Just as if I were to ask a PhD student to solve a
problem and she goes away and comes up with the solution, she
would be the inventor and not me. So perhaps the intelligent
computer that solves the problem – or at least the
corporate entity that owns the computer – should be
The IP and legal counsel at a global pharmaceutical company
adds that in sciences the person who posed the hypothesis is
normally determined as the inventor and that will likely be the
case in any law that is revised to incorporate the increased
use of AI in drug discovery.
"If you ask the software if there is a link between high
blood pressure and smelly feet, for example, and it goes
through your databases and finds a correlation that leads to an
invention, the real invention is posing the hypothesis and the
data then bears on that hypothesis," he adds.
Where’s the bar?
There are other barriers to patenting AI-derived inventions.
The head of IP in oncology points out that as AI becomes more
used by drug companies, patent offices may start to shift
patentability criteria because "the person of ordinary skill in
the art" might not actually be a person. Because criteria such
as claim construction and obviousness are based on the notion
of the skilled person, the bar for those tenets may end up
"If the computer is making it much easier to find new
compounds, for example, does that raise the bar for
obviousness?" the head of IP in oncology asks.
"It has happened before. Ten or 15 years ago you would be
able to get quite a broad claim to antibodies in a patent. But
technology has progressed and, in general, as soon as an
antigen is known, the tools are there for anyone of moderate
skill to manufacture antibodies.
"As such, the hurdle to show inventive step has gone up and
I wonder if the use of AI becomes routine in drug
"If we are the alleged infringer and are responsible
for the actions of a computer, as we are for the
maintenance of a waste system in our factories, if
something goes wrong one would argue that we are
Given that most AIs have the technological capacity to
infringe patent claims, some in-house counsel are also
uncertain about who would be held responsible for actions taken
by machine-learning software. Should it be the developer, for
example, another person or the AI itself?
The automotive industry is undergoing a similar quandary
with the idea of who would be responsible for a car crash when
the vehicle is driven by a piece of software and not a human
The head of IP at a global pharmaceutical manufacturer says
it is unlikely that medical companies will be able to dodge
liability by pointing fingers at their AI.
"For argument’s sake if we are the alleged
infringer and are responsible for the actions of a computer, as
we are for the maintenance of a waste system in our factories,
if something goes wrong one would argue that we are liable
– whether because we’ve broken
environmental law or infringed patent rights."
But, he adds, there may be many in the profession who would
New business models
While the right filing strategies for AI-derived inventions
are an unknown it has become much easier to patent the software
itself in recent years – and for life sciences firms
that are making their own enhancements to machine-learning
technologies as opposed to using an off-the-shelf system, that
presents lucrative licensing and commercialisation
Aaron Smethurst, director of IP policy at UCB, explains that
his company has developed a system called Bonebot that provides
vertebral fracture predictions based on CT scans, which could
have been taken for other initial purposes. He says that the
software helps the company to identify patients with potential
future fractures which can help physicians begin treatment
early and prevent serious injury.
The software helps the company to identify drug developments
that will help treat the condition, but it also have obvious
application for doctors looking to identify it early so they
can treat it more effectively.
"The interesting question for a company like ours is how do
we commercialise that invention? We have the knowledge around
the program but how do we best deliver the value of this
innovation to patients as a drug company.
"We’re not a software company and are not used
to selling such tech to hospitals, so such technology will most
require partnering and the IP strategy has to take that into
These drug companies may also choose to monetise the data
generated by these inventions as they become more sophisticated
and more effective at processing information on patients and
Ferenc Molnar, head of IP support at Roche Diabetes Care,
tells Managing IP that pharmaceutical and medical device
companies may begin to go down the route of companies such as
Google or Facebook and sell the data they generate on patients
to other firms that can use it to develop their own
He adds that in that scenario, however, companies would have
to be careful not to violate data privacy laws such as the
EU’s General Data Protection Regulation. They
could do this by anonymising the information they sell.
AI advancements could one day make the medical advancement
portrayed in sci-fi shows a reality. For that to happen, life
sciences companies will first need to navigate the glaring gaps
in patent law when it comes to patenting AI-derived inventions.
But in the long run, these increasingly tech-focused businesses
will benefit from new revenue streams that arise from these
products and royalties from the machines that helped come up