ABSTRACT
Artificial
Intelligence (AI), in recent times, has been a hotly debated topic of interest
worldwide. Various stakeholders, including the technologists, researchers,
lawmakers, regulatory authorities, administration or the public in general are anxiously
awaiting solutions to multifaceted challenges grappling AI industry today.
Importantly, notions pertaining to legalities underpinning the evolution,
emergence and execution of AI enabled technologies have gained widespread
relevance. Amongst them, proprietary issues of inventorship, ownership, product
liability etc. remain a key question. This primarily holds true for researchers
interested in conceptualizing a new piece of intellectual work, establishing
its ownership and eventually commercializing the same. Intellectual Property
(IP) laws have been discretely administering the patent rules for ‘natural
persons’. However, determination of true proprietorship over technologies exclusively
borne of a ‘non-natural person’ is setting a strong undercurrent within patent
community today.
Many
questions pertaining to proprietary interest in AI technologies have been
baffling. Is the creator of AI machine also responsible for future
uncertainties? How a machine is made
accountable? Is it possible to confer a distinct identity to machine borne
work? What may be the implications on existing patent regime? Will the
unrealistic assessment of machine work by humans question the standards of
patentability? If strict liabilities or penalties imposed on original creator
of such machine stifle AI innovations in any way? The present study summarizes
the assessment of factors influencing AI patent regime, potential challenges
encountered by governing authorities during examination, and probable solutions
for its reparation.
I.
INTRODUCTION
AI is not a single technology, but a
combination of many. When any of AI techniques, for example machine learning,
fuzzy logic, probabilistic reasoning, neural network, SVMs, is utilized for
creating a possibly novel and non-obvious solution to a real world problem, a
patentable invention is borne. For the purposes of this study, such work is
termed as AI enabled inventions (AEIs). Here, the deployment of human faculties
towards creative contribution acknowledges a ‘natural person/individual’ as an
inventor. Accordingly, prevailing laws on patentability that attribute exclusive
rights to true and first inventor remains logically applicable, and any
deviation in this context is rather unfounded. On the contrary, as the role of
AI systems become predominant in invention making process to the extent of
excluding human interference in any substantial way, elementary concern of true
inventorship arises. Imagine a situation in which an original creator of AI
system is long gone, and the machine over the years has gained tremendous
capability of performing a task more efficiently by self-learning and without
being explicitly programmed. Agreeably, a natural person is not an inventor
here. The focus, thence, will be to scout for the true inventor and analyze in
granular detail what happens later. Such works exclusively borne out of
machine’s doing will be referred as Machine borne inventions (MBIs) in this
study. The scope of discussion shall be limited to nuanced analysis of
inventorship and ownership concepts, patentable elements and patentability
standards, and probable patent regulations that can be designed to cater to
such AEIs and MBIs. Further, the attempt is to address pertinent questions
raised by various Patent Offices (primarily US Patent & Trademark Office)
regarding evaluation of existing guidelines for AI patent applications.
II.
INVENTOR OF AI INVENTION
In the present context, it is important to
identify peculiar patentable elements for both AEIs & MBIs and bifurcate
incoming inventions in either of these branches. The reason being simply that
subject matter involving human intervention demands a differential treatment from
the ones that are plainly and solely machine borne, when viewed from a legal
standpoint. It will be way off base if someone is rewarded for innovation one
never indulged in.
Let’s find a probable solution ahead. An
initial assessment whether an invention falls into one of the categories-
either AEIs or MBIs is necessitated at first. In cases where the human
involvement has contributed towards an invention making process (in whole or in
part) as defined by reference to the language of claims, inventorship can be
accredited to the individual fulfilling the inventorship criteria as per the
general patent rule. Consequently, an individual claims all benefits and
entitlements accompanying an inventor under present patent framework. In
contrast, where the act of invention is entirely a machine effort- an
autodidact one, a new taxonomy defining the true creator would be more
desirable. Since the MBIs are beyond initial conception and predictability of
original inventor, titlehood directed towards machines that systematically and
innovatively solves real world problems seems to be a more lawful measure.
Speaking plainly, there should be an opportunity to confer legal personhood, perhaps
inventorship, upon such electronic personality. Demarcating inventorship
between individual and electronic personality will not only explicitly
establish authenticity of true originator of relevant work, but also eliminate undue
need to admit improper titles.
However, status quo of inventorship is settled
based on disclosure made in a patent document. As a basic rule, true inventor
has to disclose in full detail the particulars of his invention along with
advantages and practical utility accompanying it. Where a human is considered
as an inventor, the disclosure shall entail all probable and foreseeable
capabilities and benefits derivable from its use. After all, this would be one
of the primary reasons for acknowledging one as an inventor. On the other hand,
all that remaining unlisted, unforeseeable and unpredictable may be envisaged
as a part of machine endeavor. Under such circumstances, the electronic
personality could be assumed to have all patent rights and obligations as a
human.
III.
INTRODUCING CONCEPT OF OWNERSHIP FOR ELECTRONIC
PERSONALITY
Along being an inventor, the machine is
assigned with other associated denominations, including that of ownership. The
concept of ownership for MBIs have potential implications in realms of patent
law. Agreeably, AI machines may not hold any intangible rights per se; they
cannot be trusted with any accountability as well. Still, it will be too vague
to undermine the significance of conferring distinct inventorship and ownership
rights to natural and electronic personality, just for the reason that it is
too far-reaching giving meaning to such concept. This problem, if not remedied
today, will lead to disorderliness of tomorrow. Following paragraphs propose a workable
course of action in designating ownership titles for such MBIs.
At the outset, the jurisprudence of current
patent regime can be tweaked to establish causal link between the owner and
inventor. From this perspective, ownership can be accorded to individuals undertaking
the task of producing and presenting the application on behalf of machine before
the patent office. This can, unequivocally, be an original inventor, assignee,
manufacturer, seller or anyone who is interested in furthering patent rights
granted upon the beneficiary in lieu of certain disclosure. Since patent scheme
is a quid pro quo arrangement, along with the rights flow legal risks associated
with claimed ownership. The person interested in deriving profits from
ownership of such machine shall also be ready to bear penalties and liabilities
underlying such legal risks. The assessment has to be thorough and ownership
fully documented. Though the whole scheme may appear to follow existing trends,
yet the concern regarding willingness of an inventor to own a patent laced with
that kind of liabilities and uncertainties will continue to exist. For many, a
Trade Secret may be a more viable option to muster profits with no strings
attached. Some logical answers to these perplexing thoughts may be found in
subsequent paragraphs.
IV.
RE-INTRODUCING PRODUCT PATENT REGIME IN MBIS
The above scenario may be divided into two sub
scenarios i) where product and process patent is granted, and ii) where only
product patents are granted. Before delving deeper into tradeoffs of above two
scenarios, let us first understand which parts of AI are eligible for patent
protection-
a)
Data
selection and processing to be used as a raw material for machine; or
b)
Training
process that machine undergoes to deliver appropriate results; or
c)
Mathematical
Models or algorithms that are tied to these machines to make them suitable for
specific real world application
d)
A
special purpose AI machine capable of executing algorithms or models in a
technical and non-routine manner to produce a practical effect.
Apart from these
elements, the ingenuity in filing of AI related patents may reside in a-
-
Computer
scientist who designs and executes machine learning algorithms on machines
-
A
data scientist who innovatively selects, prunes and labels training data used
to train a model for AI; a selection that does not merely constitute a mental
act, but an autonomous system that can perform this task in a specific way
-
An
engineer who is capable of identifying and integrating the output of machine
into real world application
-
Or a
combination thereof
AI based inventions make use of models and
networks that evolve as a part of training process and are probabilistic in
nature. This processing of using complex mathematical models to train the
system are safely enclosed within its hidden layers. Further, the fundamental
building blocks of AI i.e. mathematical models are not eligible for patent
protection; though once they have technical character or exhibit technical
improvement, they are fit to be se. However, it is almost virtually impossible for
an individual to discover what is routine information and processing for a
typical self-taught machine. If the solution is integrated with real world
application-expressly the outcome or technical effect of such technical
solution, it is likely to qualify as a patentable subject matter. But instead, the
end result that proposes a solution to particular problem of real world can be
discerned regardless of the training process it undergoes, which more or less
remains imperceptible, the solution still stands a chance of seeking patent
protection.
Therefore, by reinstituting the faith in
product patent regime for MBIs, the owners can be stimulated to file for
product patents and claim benefits. Infact, product patent will offer a higher
level of protection as there not be any other competitor for same product. In
parallel, relentless patent filing to match exceedingly high rate of innovation
in MBIs can also be overtly monitored and checked. Considering the accelerated growth in MBIs, a
possible scenario may exists, where small number of owners patent too many
inventions too quickly, almost monopolizing the entire set up. To evade this
speculated imbalance, it will be favorable to restrict the term of grant for
product patent in MBIs, say for example, 7 years. This reasonable period of
protection should neither seem burdensome for owners carrying weight of legal
uncertainties, nor permit selected few to control markets for a prolonged 20-year
term, especially considering the pace with which this domain is progressing.
V.
PATENTABILITY STANDARDS FOR MBIS
As the concept of “only product patents” for
MBIs gains further pace, it is exposed to another intriguing level of
complexity- deciding patentability standards for MBIs in addition to
adjudicating machine’s capability when assessed by human efforts. If the answer
is not a strict no, it is also not an absolute yes. But patentability demands
the otherwise. It has to be an absolute yes for a full proof examination. Candidly,
it will be beyond the mental faculties of humans, in general, to fairly and
squarely scrutinize capabilities of MBIs performing extraordinarily complex
computations within its series of network layers. With the complexities of
mathematical models unfolding in those hidden layers, the novelty and
obviousness of its unique output can be fairly assessed only by an equivalently
capable electronic personality having skill in art, suitably coined as ‘EPHOSITA’.
As PHOSITA (person having ordinary skill in
art) is for AEIs, EPHOSITA (electronic personality having ordinary skill in
art) may be for MBIs. The enormous volumes of AI inventions gaining popularity
over these years and the sprint with which they are aced will, sooner rather
than later, necessitate the need for developing a machine probing machine. Inducting
EPHOSITA will also reduce the burden of investigating magnitude of patent
applications that will be rampantly filed by spirited innovators in swelling
domain of AI. Additionally, PHOSITA and EPHOSITA are not likely to have their
roles interchangeably played either. Neither EPHOSITA be inspecting
applications filed by individuals for the grounds of obviousness will be too
occasional and germane; nor PHOSITA be roped in for determining capabilities of
MBIs, the premise being considerably different in each case.
However, with no EPHOSITA existing today, way
forward is to bring one. Introducing PHOSITA will be posed with several severe
challenges- making EPHOSITA functional; deciding standards of equivalence; most
importantly examining the status of patent applications filed in the interval
between filing date and date of inducting EPHOSITA (a transitional phase). The road
ahead towards building a functional and equivalently capable PHOSITA does not
seem to be bothersome for much talented pool of experts including software
developers, thinkers, technologists or lawmakers. Collaboratively, they can
develop and deliver an electronic personality that can skillfully master the
art of assessing novelty and obviousness features of MBIs. A discreet set of
rules and procedures shall be, however, required for examination of
applications filed in between the transitional phase.
Well, one approach can be to reinstate filing
of mailbox application. The solution is simple and well known to all. Simply,
it implies that until EPHOSITA becomes operational, MBIs related applications
can be filed as a mailbox application to gain exclusive marketing rights (EMRs)
for a determined period of time. Such applications can be kept pending in the
mailbox or blackbox which can be opened after a predetermined time, say for
example until 2021-whenever EPHOSITA is fit to be launched. During this
transitional phase, EMRs can be provided to all those applicants whose
applications are lying in mailbox. On and from the date of the publication of
the application for patent and until the date of grant of patent in respect of
such patents, the applicant may have like privileges and rights as if the
patent has been granted on date of publication of application. This deferred
examination and provisional protection will be equally advantageous for both
the parties. While the applicants can secure a priority date for themselves,
the examination can be appropriately performed as and when an equivalent EPHOSITA
is operative within the Patent Office.
As has been demonstrated throughout this study,
the current legal and regulatory framework may not be suitable for addressing
inventorship, ownership and patentable rights of MBIs simply because MBIs are
purely machine created, beyond human comprehension, and require a distinct
entity to equitably decide for its patentability. Hence, while the present
legal regime be reserved for AEIs or human facilitated inventions, a new
framework may be devised for MBIs with inclusions of EPHOSITA and provisional
EMRs for a fixed tenure.
VI.
CONCLUDING REMARKS
The foregoing discussion establishes the
growing need for a new mechanism of making MBIs meaningfully patentable with
inventorship and ownership rights respectively attributable to its original creator
and to one capable of bearing ownership risks. Also that the newly crafted
creations cannot be unfittingly casted in molds of existing patent order.
Although the proposed solution of introducing EPHOSITA, reinstituting product
patent regime, provisioning EMRs may appear burdensome and too demanding for
now, but the quicker we respond to these technological advancements, safer we
would be treading in these unchartered waters. The product patent regime will prevent
flooding of patent rights for MBIs with unfenced boundaries as the process of
realizing the final outcome will always remain a subject of contention.
Meanwhile during the arduous course of creating EPHOSITA, transient provisions
of EMRs will secure priority dates for applicants besides granting the owners
the right to sell and distribute their inventions until substantive
examination. To conclude, new directives are necessitated for comprehensively
addressing numerous patent law issues for MBIs involving AI activity both at
present and in foreseeable futures of technological development.
REFERENCES
1.
Dr.
Noam Shemtov, A study of inventorship in inventions involving AI activity, https://www.epo.org/news-issues/issues/ict/artificial-intelligence.html#study
2.
Tracey L. Thompson, USPTO, Federal
Register, August 27, 2019, https://www.govinfo.gov/content/pkg/FR-2019-08-27/pdf/2019-18443.pdf
3.
World Economic Forum, Artificial
Intelligence Collides With Patent Law, April 2018, http://www3.weforum.org/docs/WEF_48540_WP_End_of_Innovation_Protecting_Patent_Law.pdf
4.
Timothy
Chou and Philip H. Albert, When AI Creates AI -- Who Owns the Rights?,
September 9, 2019, https://www.dwt.com/blogs/artificial-intelligence-law-advisor/2019/09/uspto-comments-on-ai-patent-applications#page=1
No comments:
Post a Comment