Beruflich Dokumente
Kultur Dokumente
Abstract—The artificial intelligence technology has brought performance is a well-defined branch of theoretical
unprecedented challenges to today's ethical standards, legal computer science often referred to as computational
rules, social order and public management systems. In terms learning theory. Machine learning is used to make
of Patent law, the main challenge is the patent eligibility of the predictions to improve patient care, avoid credit card fraud,
artificial intelligence invention. Since inventions in the field of
artificial intelligence include methods for implementing mental
manage financial portfolios, etc.
steps using computers and devices designed to automate Deep learning is a specific approach used for building
mental steps, the implicit requirements for judging patent and training neural networks, which are considered highly
eligibility may be inconsistent with the nature of artificial promising decision-making nodes. An algorithm is
intelligence technology. However, due to the importance of considered to be deep if the input data is passed through a
artificial intelligence to social development, many countries series of non-linearities or nonlinear transformations before
have begun to revise patent examination guidelines related it becomes output. In contrast, most modern machine
patent-eligibility criteria to address the challenges of artificial learning algorithms are considered "shallow" because the
intelligence. input can only go only a few levels of subroutine calling.
Keywords-Artificial Intelligence; Computer Technology; Deep learning makes training the neural network easier and
Machine Learning; Patent Eligibility faster, and it can yield a better result that advances the field
of artificial intelligence.
B. Key Technology Overview
I. INTRODUCTION This study analyzes the IPC codes of related patents in
“AI is the new electricity. I can hardly imagine an the field of artificial intelligence in the past 15 years. The
industry which is not going to be transformed by AI.” said patent data come form the patent database Questel Orbit.
Andrew Ng, CEO, Landing AI and deep learning.AI, The frequency codes with high frequency indicate that the
California, US, speaking via video call. The emergence of corresponding technology categories are hotspots in the field
the artificial intelligence revolution has brought of artificial intelligence. Figure 1 shows the artificial
unprecedented challenges to today's ethical standards, legal intelligence related patents from 2004 to 2018 as the data
rules, social order and public management systems. It not source. Those key technology are: Computer
only conflicts with the existing legal order, but also technology(4914), It methods for
highlights the defects of the supply of legal system products, management(1607), Control (1583), Digital
and even subverts the legal cognition that we have formed. communication (975) , Measurement (830) , Telecommunic
In terms of Patent law, the main challenge is the patent ations (672) , Medical
eligibility of the artificial intelligence invention. technology (485) , Handling (434) , Furniture, games (430),
Electrical machinery, apparatus, energy (389),
II. TRENDS OF ARTIFICIAL INTELLIGENCE PATENT Transport (346) , Audio-visual technology (320), Other
special machines (273), Civil engineering (163) , Other
A. Artificial Intelligence
consumer goods (135), Thermal processes and
Artificial intelligence is the ability of machines to apparatus (134) , Machine tools (115), Analysis of
execute functions that generally are associated with human biological materials (102), Chemical engineering (102),
intelligence. Psychological activities performed by artificial Engines, pumps, turbines (79) , Optics (78) , Environmental
intelligence techniques include cognitive and analytical technology (73), Mechanical elements (61), Materials,
activities. They include problem solving, perception metallurgy (60) , Biotechnology (54), Basic communication
patterns, understanding language, answering questions, processes (49), Food chemistry (47) , Semiconductors (38),
proving theorems, and learning from experience. Textile and paper machines (32), Macro-molecular
Machine learning is a core part of artificial intelligence. chemistry, polymers (27), Basic materials
Learning without any kind of supervision requires an ability chemistry (25) , Pharmaceuticals (19), Surface technology,
to identify patterns in streams of inputs, whereas learning coating (15) , Organic fine chemistry (11) , Micro-structure
with adequate supervision involves classification and and nano-technology (10).
numerical regressions. Classification determines the
category an object belongs to and regression deals with
obtaining a set of numerical input or output examples,
thereby discovering functions enabling the generation of
suitable outputs from respective inputs. Mathematical
analysis of machine learning algorithms and their
111
IV. NEW TRENDS IN THE PATENT ELIGIBILITY judicial exception. A claim that recites a judicial exception,
LEGISLATION OF ARTIFICIAL INTELLIGENCE but is not integrated into a practical application, is directed to
the judicial exception under Step 2A and must then be
A. The United States of America evaluated under Step 2B (inventive concept) to determine
On January 7, 2019, the U.S. Patent and Trademark the subject matter eligibility of the claim.
Office published updated examination guidance, instructing The new Guidance has a very positive impact on AI
the examining corps and the PTAB how they should apply patents, which has paved the way for more legal certainty. It
35 U.S.C. 101. On the same day, the USPTO also released will lead to even more AI patents and will inspire more AI
the latest in its series of examples of how to apply the related inventions.
Article 101 to examine a application. B. The European Patent Convention
Prior to this, the most influential decision for artificial
On 1 November 2018 ,the EPO updated official
intelligence inventions may occur in the case of Alice Corp.
guidelines came into effect. This is the first time, the
v. CLS Bank International(2014) . The Alice case a decision
European Patent Office published guidelines on the patent
of the United States Supreme Court about patent eligibility. eligibility of artificial intelligence (AI) and machine learning
The case involves a patent for a computerized method of (ML) technologies. The new guidelines defines AI and ML
financial transaction systems. When only one party is as "based on computational models and algorithms for
financial, this method can reduce the “ settlement risk ” classification, clustering, regression, and dimensionality
exchange agreement to fulfill its obligations. The method reduction, such as neural networks, genetic algorithms,
proposes to use a computer system as a third-party support vector machines, k-means, kernel regression, and
intermediary to facilitate financial obligations between the discriminant analysis."
parties. The US Supreme Court ruled that the two-step test As the guidelines provide, these models and algorithms
established by Mayo governed all subject matter of patent “are per se of an abstract mathematical nature, irrespective
eligibility. In particular, for the background of abstract of whether they can be ‘trained’ based on training data”,
concepts, the Supreme Court established the two-step the guidance concerning mathematical methods – which are
framework for computerized invention patent eligibility. generally excluded from patent-eligible subject. According
The Alice decision was considered as a huge setback to the updated guidelines, an AI or ML invention should be
for the application of patent eligibility subject matter rule to applied to an actual technical problem in order to become
computerized invention. It was a broad, categorical eligible for patent protection, “If a claim is directed either
exclusion of certain inventions that were deemed “directed to a method involving the use of technical means (e.g. a
to” an abstract idea,natural phenomenon, or law of nature. computer) or to a device, its subject-matter has a technical
To clearly distinguish between patent-eligible subject character as a whole and is thus not excluded from
matter and subject matter falling within a judicial exception, patentability under Art. 52(2) and (3)”. This means that if
the USPTO revised its examination procedure with respect an AI or ML invention has a technical impact on an actual
to the Alice/Mayo test. The 2019 Revised Patent Subject problem it may be patent eligible under EU patent law.
Matter Eligibility Guidance focuses on two major changes To further clarify the difference between technical
purposes and non-technical purposes, the new guidelines
to the first step of the Alice / Mayo test.
First, the Revised Guidance expounds that abstract ideas give two examples of AI and ML: “ For example, the use
can be grouped as, e.g., mathematical concepts, certain of a neural network in a heart-monitoring apparatus for the
methods of organizing human activity, and mental processes: purpose of identifying irregular heartbeats makes a technical
a) Mathematical concepts – mathematical relationships, contribution. The classification of digital images, videos,
audio or speech signals based on low-level features (e.g.
mathematical formulas or equations, mathematical
edges or pixel attributes for images) are further typical
calculations;b) Certain methods of organizing human
technical applications of classification algorithms. ” In
activity – fundamental economic principles or practices
(including hedging, insurance, mitigating risk); commercial contrast, “ Classifying text documents solely in respect of
or legal interactions (including agreements in the form of their textual content is however not regarded to be per se a
contracts; legal obligations; advertising, marketing or sales technical purpose but a linguistic one (T 1358/09).
activities or behaviors; business relations); managing Classifying abstract data records or even
personal behavior or relationships or interactions between "telecommunication network data records" without any
people (including social activities, teaching, and following indication of a technical use being made of the resulting
rules or instructions); c) Mental processes – concepts classification is also not per se a technical purpose, even if
the classification algorithm may be considered to have
performed in the human mind14 (including an observation,
valuable mathematical properties such as robustness (T
evaluation, judgment, opinion).
Second, this guidance explains that a patent claim or 1784/06).”
patent application claim that recites a judicial exception is Whether the new Guidelines will make patenting AI and
not “ directed to ” the judicial exception if the judicial ML inventions as such easier in Europe is difficult to say yet.
At least, it provides welcome confirmation that the EPO is
exception is integrated into a practical application of the
open to allowing inventions which relate to the use of
112
computer-implemented technology. The updated guidelines REFERENCES
provide application standards for applicants' reference. [1] Kica E, Groenendijk N. The governance of European
provide criteria for AI and ML patent that applicants may intellectual property rights: Toward a differentiated
use to assess the possibility that a given application will pass community approach[J]. 2018.
muster under the EPO’s examine. [2] Higgins B. Intellectual Property Technology Law Journal[J].
Intellectual Property & Technology Law Journal, 2019, 31(3).
C. China [3] Onishi K, Yamauchi I. Intellectual Property Rights for
To stimulate the incentives for computer-related Software and Accessibility to Venture Capitalists[R].
innovations, the CNIPA ( China National Intellectual Research Institute of Economy, Trade and Industry (RIETI),
2018.
Property Administration) revised its Examination Guidelines
[4] Wilka R, Landy R, McKinney S A. How Machines Learn:
for Patents, which came into force from April 1, 2017. Where Do Companies Get Data for Machine Learning and
The new revisions related to AI include the patent What Licenses Do They Need[J]. Wash. JL Tech. & Arts,
eligibility of computer software and business method. 2017, 13: 217.
Software patents [5] Holder C, Khurana V, Harrison F, et al. Robotics and law:
For computer-implemented inventions, the Guidelines Key legal and regulatory implications of the robotics age
emphasis (Part II, Chapter 9, Section 2) that "Computer (Part I of II)[J]. Computer Law & Security Review, 2016,
32(3): 383-402.
program per se" is different from "inventions relating to
[6] Fujii H, Managi S. Trends and priority shifts in artificial
computer program". The former belongs to rules and intelligence technology invention: A global patent analysis[J].
methods for mental activities and is ineligible for patent Economic Analysis and Policy, 2018, 58: 60-69.
protection, while the latter is patentable. A software claim [7] Yanisky-Ravid S, Liu X J. When Artificial Intelligence
may be drafted in a form of "medium plus computer Systems Produce Inventions: The 3A Era and an Alternative
program process" or as an apparatus claim including a Model for Patent Law[J]. 2017.
component implemented by a computer program. [8] Hattenbach B, Glucoft J. Patents in an Era of Infinite
Under the revised Guidelines, software claims such as Monkeys and Artificial Intelligence[J]. Stan. Tech. L. Rev.,
2015, 19: 32.
"a computer program product", "a machine-readable
medium", and "an apparatus comprising a processor [9] Cockburn I M, Henderson R, Stern S. The Impact of
Artificial Intelligence on Innovation[R]. National Bureau of
configured to execute instructions on a computer-readable Economic Research, 2018.
medium to perform steps of ..." shall become patent-eligible. [10] Bostrom N, Yudkowsky E. The ethics of artificial
Business Method Inventions intelligence[J]. The Cambridge handbook of artificial
For business method-related inventions the new intelligence, 2014, 316: 334.
Guidelines state (Part II, Chapter 1, Section 4) that “If a [11] Lefstin J A, Menell P S, Taylor D O. Final Report of the
claim relating to business model includes not only business Berkeley Center for Law & Technology Section 101
rules or methods, but also a technical feature, it should not Workshop: Addressing Patent Eligibility Challenges[J].
Berkeley Technology Law Journal, 2018.
be excluded from being patentable under Article 25 of the
[12] Dirican C. The impacts of robotics, artificial intelligence on
Chinese Patent Law.” business and economics[J]. Procedia-Social and Behavioral
Previously classified as a "mental activity", business Sciences, 2015, 195: 564-573.
method inventions were excluded from patent protection. [13] Jin H R. Think Big: The Need for Patent Rights in the Era of
Even though business method per se is still not eligible for Big Data and Machine Learning[J]. NYU J. Intell. Prop. &
patent under the revised Guidelines, a window is opened. An Ent. L., 2017, 7: 78.
invention related to business method is usually implemented [14] Garza R D. Software Patents and Pretrial Dismissal Based on
by software in practice, and viewed to include technical Ineligibility[J]. Rich. JL & Tech., 2017, 24: 1.
features. [15] Hattenbach B, Snyder G. Rethinking the Mental Steps
There is no doubt that, the updated guidelines relaxed Doctrine and Other Barriers to Patentability of Artificial
Intelligence[J]. Colum. Sci. & Tech. L. Rev., 2017, 19: 313.
barriers to obtaining patents to AI inventions in China.
[16] Dane T G. Are the Federal Circuit's Recent Section 101
V. CONCLUSION Decisions a Specific Improvement in Patent Eligibility
Law[J]. Fed. Cir. BJ, 2016, 26: 331.
The emergence of the artificial intelligence revolution
has brought unprecedented challenges to today's ethical
standards, legal rules, social order and public management
systems. The implicit requirements for judging patent
eligibility may be inconsistent with the nature of artificial
intelligence technology. To address the challenges of
artificial intelligence, many countries have begun to revise
patent examination guidelines related patent-eligibility
criteria. Whether these new Guidelines will make
patenting AI inventions as such easier in these
countries is difficult to say yet. At least, a window is
opened.
113