Article
27 of the Agreement on Trade-Related Aspects of Intellectual Property Rights
requires its member states to grant patents only to inventions which are new,
involve an inventive step and are capable of industrial application. The
requirement for inventive step—also known as non-obviousness—provides that an
inventor cannot be granted a patent unless the claimed invention makes a
significant advancement over existing technology.[1] The non-obviousness requirement
is implemented by WTO members in their national legislations.
Non-obviousness
is an important lever in the patent system. Primarily, it ensures that, one the
one hand, patents are not granted to inventions that merely make trivial
improvements to known technologies, thereby hindering technological progress.
On the other hand, it also ensures that inventions that require substantial
effort to make are not denied the deserved protection.[2] So critical is the
non-obviousness requirement that it has been termed “the fundamental gatekeeper
to patenting,”[3]
and “the ultimate condition for patentability.”[4]
Presently, non-obviousness is
assessed through the human lens, because creativity and innovation has, for a
long time, been the preserve of natural persons.[5] However, the use of
artificial intelligence (AI) in research and development challenges
non-obviousness on various fronts. For instance, AI applications such as machine
learning and deep learning are already generating outputs that compete with
manmade products in terms of inventiveness.[6] Scientists and researchers
are increasingly relying on AI to make diagnoses and generate necessary
medicines. For instance, Watson, an AI owned by IBM, has already identified new
drug targets as well as new indications for known drugs.[7] As I subsequently
demonstrate, these developments fundamentally alter the non-obviousness
requirement. Unless patent law evolves to address these changes, we risk
granting patents to what I term ‘obvious inventions.’ Although this article
analyses arising issues from a U.S. perspective, the derived lessons are
largely applicable to other jurisdictions.
II. Non-obviousness
in the U.S.
The non-obviousness doctrine is embodied in 35 U.S.C § 103 which stipulates that a patent cannot be granted “if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains.”
Two elements are pertinent to
assessing non-obviousness: prior art and a person having ordinary in the art
(PHOSITA). Prior art is a term that defines a broad range of evidence which
indicate that the claimed invention is already known at the time of filing.
Anything can be prior art—existing technologies, public records, publications
in newspapers and journals etc. Inventors are presumed to have constructive
knowledge of prior art.[8] Consequently, they cannot
take away knowledge that is already in the public domain. Additionally, prior
art must be analogous i.e. it must pertain to the problem with which the invention
is concerned.[9]
A PHOSITA is a hypothetical person
used when assessing non-obviousness. A PHOSITA is “...a person of ordinary
creativity, not automation,”[10] and is “presumed to have
known the relevant art at the time of the invention.”[11] The level of ordinary
skill in the art is critical in defining a PHOSITA. The Federal Circuit has
laid down several factors to determine the level of ordinary skill in the art,
including (i) types of problems encountered in the art; (iii) prior art
solution to those problems; (iii) rapidity with which innovations are made;
(iv) sophistication of the technology; and (v) educational level and any
specialities of active workers in the field.[12]
In Graham v John Deere,[13]
the Supreme Court established the approach for assessing non-obviousness. The
first step analyses the three factors provided in the statute: (i) determining
the scope and content of the prior art; (ii) ascertaining the differences
between the claimed invention and prior art; (iii) and resolving the level of
ordinary skill in the art. The second step involves incorporating evidence that
the Supreme Court termed secondary considerations, such as commercial success
of the patented product and the failure of other parties to come up with the
claimed invention.
Prior to 2007, the US Trademark and Patent Office applied a test known as the Teaching-Suggestion-Motivation (TSM) test. Pursuant to this test, an invention is obvious if a PHOSITA can find a teaching, suggestion or motivation to combine particular prior art references to remedy a technical problem.[14] However, this test was considered rigid and mechanical.[15] The Supreme Court overturned the TSM test in KSR Int’l v Teleflex Inc.[16] Consequently, the rigid requirements for teaching, motivation or suggestion were replaced by a flexible list of non-exhaustive factors that courts consider in determining whether a PHOSITA would have made the claimed invention. This is the prevailing test for non-obviousness.
III. AI and Non-obviousness
First, AI alters the notion of PHOSITA.
Scientist and researchers are already using AI tools to develop novel
inventions.[18]
AI tools have capacity to compute and combine data from different fields. This
is contrary to the current conception of PHOSITA who is required to possess
knowledge only in the field of technology which the claimed invention is
concerned with. By processing large quantities of unrelated data, AI tools can
generate technical solutions which, though obvious to the machine, might be not
be obvious to human inventors, given the limited ability of humans to correlate
such vast volume of data.[19]
Where an AI makes substantial contribution
to an invention by, for example, computing immense data which constitutes the
pith of the invention, it is inappropriate for the ordinary skill of the
PHOSITA to be still judged in light of human capabilities. Although AI is
increasingly employed in inventive processes, curiously, patent law does not
require applicants to disclose the use of AI in generating inventions.[20] In such cases, although a
human may be named as the inventor, failure to consider the role in the AI in
the inventive process, and the level of sophistication of AI in defining the
ordinary skill in the art, lowers the level of non-obviousness. Consequently, we
risk granting patents to inventions that may be more obvious.
Second, it is probable that AI will become
the standard inventors. Some scholars have noted that an age in which
“computers will overtake human inventors as the primary source of new discoveries
is foreseeable.”[21]
For instance, in the DABUS applications which have been considered in major
patent jurisdictions, including the U.S., the AI was fed data relating to the
field for training, and it independently devised and generated a potential
solution to an unspecified technical problem.[22]
Where an AI autonomously generates an invention, it is doubtful whether a human
should still be the PHOSITA. As in the first scenario, doing so exceedingly
lowers the ordinary skill in the relevant art.
Lastly, AI also impacts prior art. While a PHOSITA cannot possess, understand and recall all the prior art in his field of inquiry,[23] an AI can possess, understand and remember all the prior art contained in a database.[24] BenevolentBio, a pharmaceutical start-up, has developed an AI platform which can be fed data which constitute prior art, such as scientific publications, pharmaceutical patents, hospital records, and clinical trials.[25] Given their computational prowess, AI tools can easily identify that which humans may not easily identify and would, therefore, not consider part of prior art. Moreover, a PHOSITA is not expected to be familiar with, or consider combining, prior art from different technological realms, a capability that is well within the reach of AI.
IV. Recommendations and Conclusion
From
the foregoing analysis, it is evident that the use of AI in inventive process
affects the non-obviousness requirement in fundamental ways as to enable AI to
generate inventions which are obvious to the machine but nonobvious to human
inventors. This section formulates necessary recommendations.
First, in instances where a human
inventor uses an AI technique to generate an invention, but such use does not
substantially contribute to the invention, it is proper that the PHOSITA
remains a human. However, where an AI makes significant contribution to an
invention by, for example, mining, crossing and testing enormous data sets
which become so critical to the invention, it is improper to construe a PHOSITA
entirely as a natural person. As a first step, patent law should require
applicants to disclose if they have used AI in inventing the claimed invention.
Second, once such disclosure is made, it
may be beneficial to apply a differentiated
approach to account for the skill that the AI has contributed in making the
invention. For instance, inventions which result from more contribution of the
skill of the AI should be subjected to a higher standard of ordinary skill in
the art. This approach is proper since a human is still named as the inventor,
but a deserved attribution and discounting is made to account for the skill of
the AI in generating the invention. Consequently, we avoid the highly contentious
question of whether an AI can be named as a joint inventor. However, this
approach may discourage inventors from using AI tools, given that being
subjected to a higher standard of nonobviousness may also reduce their chances
of obtaining patents.
Third, as AI tools become standard
researchers, the PHOSITA can be a skilled person aided by machine, as has been
previously suggested.[26]
Fourth, in the event that AI start to routinely and autonomously generate inventions
that are patentable, it is, once again, inappropriate to pose a natural person
as the PHOSITA. This article agrees with the opinion of some scholars that in
this instance, the PHOSITA should be replaced by a “machine of ordinary skill
in the art.”[27]
However, although this approach would appropriately elevate the non-obviousness
standard, it would challenge other conditions for patentability, such as
novelty and inventorship.
Fifth, AI tools are able to analyse huge
amounts of data of what would constitute prior art rather easily. Further,
while it might be obvious for an AI to combine prior art from different fields to
solve a technical problem, this might not occur to humans. Thus, the scope of
prior art should be expanded to compensate for the extensive capabilities of
AI.
Last is the assessment of the differences
between the claimed invention and prior art. This exercise should be relatively
easy to conduct once the scope of prior art is expanded—the differences will be
much less, and obviousness much higher where an AI is utilised.
Implementing these recommendations would likely
ensure that the non-obviousness doctrine effectively sieves obvious inventions
from penetrating the patent ecosystem.
Endnote
[1]Gregory Mandell, ‘The Non-Obvious
Problem: How the Indeterminate Non-Obviousness Standard Produces Excessive
Patent Grants’ (2008) University of California, Davis Law Review Vol 42:57 p.
62.
[2]Ibid.
[3]Dmitry Karshtedt, ‘Nonobviousness:
Before and After’ (2021) Iowa Law Review Vol 106:1609 p. 1611, referencing John
R. Thomas, ‘Formalism at the Federal Circuit’ (2003) AM University Law Review.
[4]Ibid.
[5]Daniele Fabris, ‘From the PHOSITA
to MOSITA: ‘Will “Secondary Considerations” Save Pharmaceutical Patents from
Artificial Intelligence? (2020) IIC 51:685-708, p. 690.
[6]Ibid
[7]Ryan Abbott, ‘Everything is
Obvious’ (2019) UCLA Law Review Vol 2, p. 22.
[8] Mast, Foos & Co v Stover
Manufacturing 177 U.S. 485, 493 (1900).
[9]Abbott (n 7) 21.
[10]KSR Int’l v Teleflex Inc., 550 U.S.
398, 421(2007).
[11]USPTO Patent Examination
Guidelines.
[12]In re GPAC Inc. 57 F.3d 1573, 1579
(Fed Cir.1995).
[13]Graham v John Deere Co. of Kansas
City, 383 U.S. 1 (1996).
[14]Fabris (n 5) 689.
[15]For a comprehensive critique of the
TSM test, see Mandell (n 1) and Karshtedt (n 3).
[16]Supra (n 10).
[17]Borroughs Wellsome Co v. Barr
Labs., Inc., 40 F.3d 1223, 1227-28 (Fed. Cir. 1994).
[18]Supra (n 5).
[19]Fanris (n 5) 679.
[20]Lexi Heon, ‘Artificially Obvious
but Genuinely New: How Artificial Intelligence Alters the Patent Obviousness
Analysis’ (2022) Senton Hall Law Review p. 378.
[21]Ryan Abbott, ‘I think therefore I
invent: Creative computers and the future of patent law (2016) Boston College
Law Review 57: 1079-1126.
[22]Douglas Goldhush, ‘DABUS Denied:
Only Natural Persons can be Named as Inventors on US Patents https://www.iptechblog.com/2020/04/dabus-denied-only-natural-persons-can-be-named-as-inventors-on-us-patents/#:~:text=The%20USPTO%20took%20the%20position,in%20other%20non%2Dpatent%20contexts.
[23]In re Wood 599 F2d. 1032, 1037 (Fed
Cir. 1979).
[24]Heon (n 20).
[25]Fabris (n 3) 696.
[26]Abbott (n 7) 35.
[27]Fabris (n 3) 692.
Fidelice is
an IP lawyer, specialising in patent law. She holds an LL.M. in Intellectual
Property & Competition Law from the Munich Intellectual Property Law
Center. She is a Thomas Edison Innovation Law & Policy Fellow at the Center
for Intellectual Property x Innovation Law, George Mason University. Fidelice
is also a Tutorial Fellow at Maseno University, School of Law, and is qualified
to practice before the Kenyan Bar. Her email address for purposes of
correspondences is: fideljustice8@gmail.com