11
UDK: 347.771:004.8
DOI: 10.51940/2021.1.11-61
Joseph Straus*
Will Artificial Intelligence Change Some 
Patent Law Paradigms?
“Science, like nothing else among the institutions of 
mankind, grows like a weed every year. Art is subject to 
arbitrary fashion, religion is inwardly focused and driv-
en only to sustain itself, law shuttles between freeing us 
and enslaving us.”
K.B. Mullis (Nobel Laureate for Chemistry 1993)
1. Introduction
The growth of scientific and technical literature reporting on developments and 
achievements in artificial intelligence (AI), in its broadest sense understood as “a collec-
tion of technologies that combine data, algorithms and computer power”,1 is impressive. 
Since the mid of the last century, some 1.6 million AI-related scientific publications have 
been published, more than half of them since 2001, and some 340.000 AI-related patent 
applications filed.2 As statistics reveal, AI technology has already outgrown the stage of 
theoretical research. The ratio of scientific papers to inventions has decreased from 8:1 
in 2010 to 3:1 in 2016.3 Thanks to enormous investments in and practical applications 
of AI-related products and processes across all industries, AI technology has become an 
important technological and economic factor.4
* Dr.jur., Dres. jur. h.c., Professor of Law (University of Munich and University of Ljubljana), Emeritus 
Director at the Max-Planck Institute for Innovation and Competition, Munich; j.straus@ip.mpg.de
1 European Commission (2020), p. 2.
2 WIPO (2019a), p. 13.
3 Ibid., p. 14 and pp. 39–40.
4 Cf. J. Straus (2021).
2
ZZR 2021.indd   11 10. 12. 2021   14:04:45
12
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
2. AI-Technology and its Tools
Closer to the original definition of AI as “the science and engineering of making 
intelligent machines, especially intelligent programs”, “invented” by American computer 
scientist John McCarthy in 1956,5 than the broad definition of the EU Commission’s 
White Paper comes that of the Merriam-Webster Dictionary, where AI is defined as 
“the capability of a machine to imitate intelligent human behavior [and intuition]”.6 
Computer scientists found out how to make machines intelligent by developing a meth-
od of data analysis that automates analytical model building, which uses algorithms7 
that iteratively learn from data, and which allows computers to find hidden insights 
without being explicitly programmed where to look, known as machine learning (ML).8 
Moreover, computer scientists have further developed the deep learning method, a form 
of machine learning that enables computers to learn from experience and understand the 
world in terms of a hierarchy of concepts. Thus, the computer learns from experience 
without further interacting with a human operator.9 To optimise solutions sought, re-
searchers have developed a family of algorithms inspired by biological evolution, such as 
reproduction, mutation, recombination, crossover and selection, known as evolutionary 
algorithms (EAs), which form a subset of evolutionary computation techniques (ETs). 
The most widely known type of ETs is a search technique called genetic algorithm (GA), 
a category of stochastic search to solving complex problems, which starts with an initial 
set of random solutions (the population), each representing a solution to the problem at 
issue (so-called chromosome). After the chromosomes evolved through many successive 
iterations (called generations), the algorithms converge to the best chromosome, repre-
5 G. Gurkaynak, I. Yilmaz and G. Haksever (2016), p. 753.
6 Artificial Intelligence, Merriam-Webster.
7 M. Marrafia and A. Paternoster (2017), p. 929, 930, define an algorithm as “an effective procedure 
to solve a given problem, that is, a finite sequence of elementary and totally explicit (= well defined 
and not ambiguous) instructions.” (Quoted from D. Kim, ‘AI-Generated Inventions’: Time to Get the 
Record Straight? GRUR International, 69 (5), 2020, p. 450). In computational methods, algorithms 
and codes function together. They “indicate how to organize and describe a series of actions to 
achieve a desired result: the algorithm constitutes the stage of designing and evaluating the strategy 
on which to build single actions, while coding reflects the operational phase that leads to the execu-
tion of those actions on a particular computing device, such as a PC.” (P. Ferragina and F. Luccio, 
(2018), p. 3, quoted from D. Kim, 2020, p. 450).
8 I. Bratko, (1994), pp. 305–312. However, this does not mean, that ML works without any in-
structions. ML leverages mathematical and statistical methods. ML process has two components: 
One is “purely mathematical whereby computational operations are guided by formulas, equations, 
functions, etc. constituting a part of an algorithm; the other component is a statistical matter whose 
variables are most correlated with the outcome.” (D. Kim, ibid., p. 451 with references to pertinent 
informatics literature).
9 K.G. Kim (2016), pp. 351–354.
ZZR 2021.indd   12 10. 12. 2021   14:04:45
13
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
senting the optimum or sub-optimal solution to the problem.10 However, as Daria Kim 
observes, also a computer-controlled by a genetic algorithm does not something ‘out of 
the blue’ because the application of stochastic randomised local search requires a set of 
prerequisites that ultimately determine how computation is executed.11
The development of an ML algorithm involves: obtaining high-quality data for train-
ing the algorithm, accessed or acquired from scientific libraries, etc., pre-processing data, 
including cleaning data by removing outliers or reducing dimensions, training an ML 
algorithm on the data, and obtaining a final trained algorithm (i.e. the model), which 
gives output data (solutions) when shown new input data.12
To provide the necessary training for ML algorithms, complex data sets of interest, 
i.e. the “training data”, are necessary. ML methods can use two different approaches 
to train algorithms. Either the data set used for training algorithms (the input data) is 
“labelled”, meaning that it has both relevant features and the target results that the pro-
grammer is interested in, i.e. the correct response. In this, so-called supervised learning, 
the algorithm learns the relationships between the data and the labels to make predic-
tions on new, previously unseen data. By comparing the prediction errors of the mod-
els by inputting the validation set data into each model, the algorithm measures their 
accuracy.13 In case the training data for algorithms is unlabelled, the ML algorithm is 
“unsupervised” and develops a model that extracts common elements of interest from 
the data set, teaching itself the set of features that makes the targeted subject. “In essence, 
unsupervised machine learning uses data sets that do not have specific labels fed into the 
algorithm for the purpose of identifying common trends embedded in that data set”.14
A key AI tool is the “artificial neural networks” (ANN), an “information processing 
paradigm that is inspired by the way biological systems, such as the brain, process infor-
mation”,15 for which patent applications have been filed since the late 1980s, and pat-
ents granted since the beginning of 1990s.16 Shimon Ullman describes ANN as “highly 
10 Cf. ‘Evolutionary Algorithm’, Wikipedia, <https://en.wikipedia.org/wiki/Evolutionary_algorithm> 
(last accessed 17 August 2020). See for more M. Gen and L. Lin (2009), pp. 489–490 with further 
references and D. Kim, 2020, p. 451 with further references.
11 D. Kim, 2020, p. 452.
12 Cf. H.R. Jin, 2018, p. 89; and A. Greenberg, 2020, pp. 329–337.
13 H.R. JIN, 2018.
14 Ibid.
15 C. Stergiou and D. Siganos, 2014.
16 For instance, on 10 July 1990, the US Patent and Trademark Office (USPTO) granted US Patent 
4,941,122 for ‘Neural Network Image Processing System’; on 6 October 1992 US Patent 5,153,923 
for ‘High Order Information Processing Method by Means of a Neural Network and Minimum 
Searching Method Therefor’; on 23 March 1993 US Patent 5,197,114 for ‘Computer Neural 
Network Regulatory Process Control System and Method’, which allowed for the elimination of 
a human operator from real time control of the process; and, the last to refer to, on 22 December 
ZZR 2021.indd   13 10. 12. 2021   14:04:45
14
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
reductionist approach to model cortical circuitry” and observes that “in its basic current 
form, known as a “deep network” (or deep net) architecture, this brain-inspired model is 
built from successive layers of neuron-like elements, connected by adjustable weights17, 
called “synapses” after their biological counterparts”.18 ANNs, like humans, learn by 
example. ANNs have
“The ability to derive meaning from complicated or imprecise data can be used 
to extract patterns and detect trends that are too complex to be noticed by either 
humans or other computer techniques. A trained neural network can be thought of 
as an ‘expert’ in the category of information it has been given to analyze.”19
How powerful these ANN machine learning algorithms are, is best demonstrated 
by the AlphaZero program, using a deep neural network, that recently convincingly 
defeated the world champion programs in the games of chess and shogi (Japanese chess), 
as well as Go. The AlphaZero algorithm starts from random play and with no domain 
knowledge given, except the game rules. Unlike the state-of-the-art programs, which are 
based on powerful engines that search many millions of positions,20 leveraging domain 
expertise and sophisticated domain adaptations, AlphaZero learns the necessary move 
probabilities and value estimates entirely from self-play and uses them to guide its search 
in future games.21
However, as Greenberg emphasises, although “ML algorithms learn from the data 
and come up with a final model spontaneously, the developer’s ingenuity still plays a ma-
jor role”. She points to the fact that the development process “including choosing which 
ML method(s) to employ for a given problem, how to curate the training data, which 
algorithm parameters to select, and how to test the model for accuracy”, still requires 
many human decisions.22 Thus, “The role of human intervention remains fundamental 
1998 US Patent 5,852,815 for ‘Neural Network Based Prototyping System and Method’, which 
does not require algorithmic based software in order to train or operate.
17 Weight is the parameter within a neural network that transforms input data within the network’s hid-
den layers, <https://deepai.org/machine-learning-glossary-and-terms/weight-artificial-neural-net 
work> (last accessed 3 August 2020).
18 Sh. Ullman, 2019, pp. 692–693.
19 C. Stergiou and D. Siganos, 2014.
20 In case of the AlphaGo model, defeated by AlphaZero, the model was built by reinforcement learn-
ing from a database consisting of over 13 million moves of world class Go players (cf. D. Silver et 
al., 2016, p. 484; see also H.R. Jin, 2018, p. 80).
21 D. Silver et al., 2019, pp. 1140–1144.
22 A. Greenberg, 2020, pp. 334–335. Along the same lines with detailed arguments D. Kim, 2020, p. 
455.
ZZR 2021.indd   14 10. 12. 2021   14:04:45
15
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
to the programming of AI devices, the selection of input data and the application of the 
results obtained.”23, 24
3. Does AI Autonomously Generate Products and Processes?
AI-related reports in scientific magazines, however, suggest seemingly the opposite 
and entrap many legal scholars to buy into the idea that AI itself, meaning autonomously, 
generates patentable inventions.25 The publications of Jo Marchant, “Powerful Antibiotics 
Discovered Using AI”,26 Robert Service, “AIs Direct Search for Materials Breakthroughs 
– Decision Making Algorithms Transform How Robots Evaluate and Synthesize Solar 
Cells and More”, 27 and Derek Lowe, “AI Designs Organic Syntheses”,28may illustrate this.
In the case of the discovered new antibiotic halicin, reported by Marchant, the origi-
nal research report in the “Cell” Magazine shows that the group around Regina Barzilay 
and James J. Collins, first trained a deep neural network model to predict growth inhi-
bition of Escherichia Coli using a collection of 2.335 molecules. Then they applied the 
resulting model to several discrete libraries, comprising more than 107 million mole-
cules, to identify potential lead compounds with activity against E. Coli. After ranking 
the compounds according to the model’s predicted score, they lastly selected a list of 
candidates based on a pre-specified prediction score threshold, chemical structure and 
ability.29 As described in detail, each of those steps required making complex decisions 
by the group.30
23 Committee of Legal Affairs of the European Parliament, Report on “Intellectual Property Rights for 
the Development of Artificial Intelligence” of 2 October 2020, Explanatory Statement, pp. 12–13 
(Doc. PE650.527v02-00, A9 - 0176/2020).
24 The reported AI achievements still belong in the so-called category of “Artificial Narrow 
Intelligence”. Although the AI development is approaching its next stage, the so-called “Artificial 
General Intelligence”, meaning that computers will be “as smart as humans in every aspect and 
capable of performing all intellectual tasks humans can”, the point in time when this will become 
reality, is an open question. Much more so the question, when the serious concern causing so-called 
“Artificial Superintelligence” will concur the world. Potentially the AI superintelligence will be 
“much smarter than the best human brains in practically every field, including scientific creativity, 
general wisdom and social skills.” See for more J. Straus, 2021, with further references.
25 A current example is the publication of E. Bonadio, L. McDonagh and P. Dinev, 2021, pp. 48–66.
26 Nature News, 20 February 2020, <https://www.nature.com/articles/d41586-020-00018-3> (last 
accessed 5 August 2020).
27 365 Science, 2019, pp. 1295–1296.
28 555 Nature, 2018, pp. 592–593.
29 J.M. Stokes et al., 2020, pp. 689–690.
30 Ibid., pp. 690–702. N. Benaich, 2020, p. 21), assessed the decisive role of researchers in this case 
as follows: “AI only uncovered the new antibiotic with the help of human researchers, who ensured 
ZZR 2021.indd   15 10. 12. 2021   14:04:45
16
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
As regards the article “AIs direct search for materials”, the breakthrough for the thin-
film materials, which, according to Robert Service, a group of Canadian and US re-
searchers led by C.P. Berlinguette has achieved by outsourcing the effort to a single-armed 
robot overseen by an AI algorithm, the final outcome of the process cannot be attribut-
ed to the employed AI alone. Even though the robot controlled by AI mixed different 
solutions, cast them in films, performed heat treatments and other processing steps, 
tested the films’ conductivity, evaluated their microstructure, and logged the results, and 
the AI interpreted each experiment and determined what to synthesise next31, along this 
entire process, many decisive decisions of the group had to be taken. As the publication 
of C.P. Berlinguette and his group32 reveals, they designed the complex Ada (the robot) 
self-driving laboratory specifically “to target organic hole and electron transport layers 
that are ubiquitous in advanced solar cells, as well as optoelectronics applications such 
as organic lasers and light-emitting diodes”. They, among others, selected the material 
whose hole mobility should be autonomously optimised. They organised the production 
of surrogate hole mobility data for optical and conductivity properties characterisation 
and, for instance, selected the Phoenics global Bayesian optimisation algorithm.33
Derek Lowe article’s title “AI Designs Organic Syntheses” follows the headline 
“Software that devises effective schemes for synthetic chemistry has depended on the 
input of rules from researchers. A system is now reported in which an artificial-intelligence 
program learns the rules for itself. See Article p.604”34. However, already the abstract of 
that article of Marvin H.S. Segler, Mike Preuss & Mark P. Waller, “Planning Chemical 
Syntheses With Deep Neural Networks and Symbolic AI” at p. 604,35 to which Lowe refers, 
indicates that much more is at stake than just an AI program. The authors begin, “Here 
we use Monte Carlo tree search and symbolic artificial intelligence (AI) to discover ret-
rosynthetic routes. We combined Monte Carlo tree search with an expansion policy 
network that guides the search and a filter network to pre-select the most promising 
retrosynthetic steps. These deep neural networks were trained on essentially all reactions 
ever published in organic chemistry”.36 What this means specifically, one learns from the 
article. First, that the authors selected Monte Carlo tree search (MCTS), because it is a 
general search technique for sequential decision problems with large branching factors 
without strong heuristics. It uses rollouts to determine position values. Thereby roll-
the computer avoided killing people by helping it distinguish between bad bacteria and species that 
are beneficial to healthy living cells.”
31 R.F. Service, 2019, p. 1295.
32 B.P. Macleod et al.
33 Ibid.
34 555 Nature, 2018, 592 (emphasis added).
35 M.H.S. Segler, M. Preuss and M.P. Waller, 2018, p. 604.
36 Ibid.
ZZR 2021.indd   16 10. 12. 2021   14:04:45
17
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
outs are Monte Carlo simulations, in which random search steps are performed without 
branching until a solution is found or a maximum depth reached. One learns further that 
the authors combined three different neural networks with MCTS to perform chemical 
synthesis planning (3N-MCTS). The first neural network (the expansion policy) guides 
the search in promising directions by proposing a restricted number of automatically 
extracted transformations. A second neural network then predicts whether the proposed 
reactions are actually feasible (in scope). Finally, a third neural network, responsible to 
estimate the position value, samples transformations during the rollout phase. To train 
the three neural networks, the group first extracted transformation rules from 12.4 mil-
lion single-step reactions from the Reaxys36 chemistry database, then determined the 
extraction rules for selections to train the neural networks. The group, among others, 
also trained a deep neural network as a binary classifier on successful and failed reactions 
to predict whether the reactions corresponding to the transformations selected by the 
policy network are actually feasible. Because researchers seldom report failed reactions, 
the group itself, based on a process it developed, generated 100 million negative reac-
tions from reactions published before 2015 for training and 10 million published in and 
after 2015 for testing. The last to mention here, the expansion procedure and the rollout 
procedure the group incorporated in the respective phases of an MCTS algorithm as 
a pipeline to form 3N-MCTS. As a result, the four MCTS phases are then iterated to 
build the search tree.37 The group has also evaluated the performance of 3N-MCTS by 
comparing its algorithm to the state-of-the-art search method, the best first search (BFS) 
with the hand-coded SMILES heuristic cost function.38
For all the three reported cases, the above modest ‘technical analysis’ has revealed that 
the design of the respective overall procedure, meaning the instructions on the derivation 
of the input-output relations, were provided by humans. This seems to be in line with the 
present stage of problem-solving complex computational methods employing AI tools,39 
as assessed by experts in AI and robotics. They “caution that characteristics such as ‘au-
tonomous’, ‘unpredictable’ and ‘self-learning’ are ‘based on an overvaluation of the actual 
capabilities of even the most advanced robots, a superficial understanding of unpredict-
ability and self-learning capacities [...], a robot perception distorted by Science-Fiction 
and a few recent sensational press announcements.”40
37 Ibid., pp. 604–605.
38 Ibid., p. 606.
39 For the phases of computational problem solving, cf. D. Kim, 2020, Table at p. 449.
40 ‘Open Letter to the European Commission. Artificial Intelligence and Robotics’, para. 2 (emphasis 
added) (quoted from D. Kim, 2020, p. 444). See also infra n. 7 (iii).
ZZR 2021.indd   17 10. 12. 2021   14:04:45
18
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
4. AI-Related Inventions in Patent Granting Practice
4.1. WIPO Statistics
With reference to the WIPO Statistics41 at the outset, this paper recalled that since the 
mid of the last century, some 340.000 AI-related patent applications have been filed. As 
the WIPO statistics referred to specifically reveal, the machine learning technique with 89 
percent of patent families, which is included in more than a third of all identified inven-
tions, dominates the field. ML is followed by neural networks, machine learning (general), 
supervised learning, probabilistic graphical models, support vector machines, bio-inspired 
approaches, classification and regression trees, deep learning, rule learning, unsupervised 
learning and another twelve AI technique sub-categories.42 Thereby IBM, with 8.290 in-
ventions, has the largest portfolio of AI patent applications, followed by Microsoft with 
5.930 applications. The other three companies in the top five filers are Toshiba with 5.223, 
Samsung with 5.102 and NEC with 4.406 applications. Of the top 30 patent applicants 
by numbers of patent families are twelve from Japan, six from the USA, five from China 
and four from Europe (Siemens ranking No. 11, Bosch No. 21, Philips No. 24 and Nokia 
No. 25), and three from Korea.43 Remarkably, among the top 500 patent applicants are 
167 universities and public research organizations. Of these, 110 are Chinese (the top 10 
are all from China!),44 20 are from the USA, 19 from Korea, 4 from Japan and 4 from 
Europe (highest ranking the German Fraunhofer Institute, ranking No. 159).45
The WIPO Statistics does not reveal any numbers on actually granted patents on AI-
related inventions. However, it contains interesting data on litigation and opposition cas-
es. According to these statistics, 1.264 AI patent families were involved in litigation cases 
and 4.231 in opposition cases from 1975 to 2017.46 Among the top litigation plaintiffs 
by several litigated patent families are Nuance Communications, American Vehicular 
Sciences and Automotive Technologies International.47 At the top of the list of defend-
ants is Microsoft, prominently followed by Apple, Alphabet, Samsung, Amazon, Sony 
and BMW.48 As regards the available data for oppositions in the EPO, the top three op-
ponents are Siemens, Daimler and Giesecke+Devrient from Germany, and the top three 
defendants are Samsung, LG Corporation and Hyundai, all from Korea.49 The WIPO 
41 WIPO 2019.
42 Ibid., Figure 3.5 at p. 42.
43 WIPO 2019, p. 15.
44 Ibid., Figure 4.1 at p. 60.
45 Ibid., p. 16.
46 Ibid., p. 111.
47 Ibid., Figure 6.5 at p. 110 and text at pp. 112–113.
48 Ibid., Figure 6.6, at p. 114 and text at pp. 113–115.
49 Ibid., Figures 6.8 and 6.9, at pp. 114–116, and text at pp. 115–117.
ZZR 2021.indd   18 10. 12. 2021   14:04:45
19
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
Statistics suggests that AI-related inventions are treated as patentable subject matter and 
are granted patents if they fulfil the statutory patentability requirements.
4.2. European Patent Office
Under Article 52 (1) of the European Patent Convention (EPC), European patents 
shall be granted for any inventions in all fields of technology, provided that they are new, 
involve an inventive step and are susceptible of industrial application. However, Article 
52 EPC in its paragraph 2 sets forth that, inter alia, mathematical methods (a), rules 
and methods for performing mental acts and computer programs (c) are not regarded as 
inventions and excluded from patentability, provided, as clarified in its paragraph 3, that 
they are claimed as such.
The European Patent Office (EPO) considers AI as an enabling technology of the 
Fourth industrial revolution.50 Annual patent applications in the field of AI in the EPO 
have grown tenfold in the last decade, to more than 800 filings in 2020.51 The Guidelines 
for Examination in the European Patent Office52 clarify that AI and ML “are based on 
computational models and algorithms for classification, clustering, regression and di-
mensionality reduction, such as neural networks, genetic algorithms, support vector ma-
chines, k-means, kernel regression and discriminant analysis”. Per se, such models were 
of an abstract mathematical nature even if they can be ‘trained’ based on training data. 
Thus, they, like mathematical methods, are patentable if they contribute to the technical 
character of an invention. For the assessment of their contribution, it is decisive whether 
they, in the context of the invention, serve a technical purpose.53 The EPO Guidelines 
stress that AI and ML find application in many fields of technology. As examples for 
the use of algorithms making a technical contribution, they point to the use of a neural 
network in a heart-monitoring apparatus to identify irregular heartbeats and the clas-
sification of digital images, videos, audio or speech signals based on low-level features 
(e.g. edges or pixel attributes for images). Moreover, the Guidelines explain that, where a 
classification method serves a technical purpose, “the steps of generating the training set 
50 Cf. EPO, Patents and the Fourth Industrial Revolution: The Global Technology Trends Enabling the 
Data-Driven Economy, Munich, December 2020, Table 2.2 at p. 20.
51 EPO Patent Index 2020, p. 4.
52 November 2019 edition.
53 EPO Examination Guidelines Part G-II, 3.6 2020, approvingly point out that “mathematical 
methods as such are excluded from patentability unless they are used for a technical purpose in 
the context of technical inventions, […]”. The Committee also underlines “the role of the patent 
protection framework in incentivizing AI inventions and promoting their dissemination, as well as 
the need to create opportunities for European companies and start-ups to foster the development 
and uptake of AI in Europe, […]” (ibid., No. 12, p. 8).
ZZR 2021.indd   19 10. 12. 2021   14:04:45
20
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
and training the classifier may also contribute to the technical character of the invention 
if they support achieving that technical purpose.”54
4.3. Japan Patent Office
Similar to the handling of AI-related inventions in the practice of the EPO is the 
practice of the Japan Patent Office (JPO). Article 29 (1) of the Japanese Patent Act (JPA) 
sets forth that any person who has made an invention which is industrially applicable 
may obtain a patent therefore. Thereby Article 2 (1) JPA defines “‘invention’ as the 
highly advanced creation of technical ideas utilizing the laws of nature”, and Article 2 
(4) JPA states that
“A ‘computer program, etc.’ in this Act means a computer program (a set of in-
structions given to an electronic computer which are combined in order to pro-
duce a specific result,...) and any other information that is to be processed by an 
electronic computer equivalent to a computer program.”
The JPO Examination Guidelines clarify that: “Those [inventions] utilizing the laws 
of nature as a whole and being considered as a ‘creation of a technical idea utilizing the 
laws of nature’ constitute a statutory ‘invention’ without being examined from the per-
spective of computer software, even though they utilize computer software”. This applies 
to inventions that either concretely perform control of an apparatus (e.g. rice cooker, 
the washing machine, engine, hard disk drive, chemical reaction apparatus, nucleic acid 
amplifier), or concern the processing of that control, or concretely perform information 
based on the technical properties such as physical, chemical, biological or electric prop-
erties of an object (e.g. rotation rate of engine, rolling temperature, the relation between 
gene sequence and expression of a trait in a living body, physical or chemical relation of 
bound substances).55 The Guidelines further clarify the notion “equivalent to programs” 
used in Article 2 (4) JPA, as “not direct instructions to computers and thus cannot be 
called programs, but have similar properties to programs in terms of prescribing comput-
er processing. For example, ‘data structure’, a logical structure of data that is expressed 
by correlations between data elements.56 The JPO Examination Handbook clarifies that, 
if “it is clear in consideration of the description, drawings and the common general 
knowledge at the time of filing, that the claimed invention is a ‘program’ even though 
the claimed subject matter is any word other than the ‘program’ (for example, ‘module’, 
‘library’, ‘neural network’, ‘support vector machine’ or ‘model’)”.57
54 Ibid., Part G-II, 3.3.1. For the EPO practice, see also Y. Ménière and H. Pihlajamaa, 2019, pp. 
332–336.
55 Part III, Chapter 1 No. 2.2 (1) of JPO Examination Guidelines. Ibid, No. 2.2 (1).
56 Ibid., No. 2.2 (1) (Note).
57 Annex B, Chapter 1, No. 1.2.1.2 Points to Note, <https://www.jpo.go.jp/e/system/laws/rule/guide-
line/patent/handbook_shinsa/document/index/app_b1_e.pdf> (last accessed 18 August 2020).
ZZR 2021.indd   20 10. 12. 2021   14:04:45
21
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
The JPO distinguishes between two main categories of AI-related inventions. The 
first category constitutes AI core inventions, characterised by mathematical or statistical 
information processing technology that forms the basis of AI, such as various machine 
learning methods including neural networks, deep learning, support vector machines, 
reinforcement learning, in addition to knowledge-based models and fuzzy logic, etc. The 
JPO assigns these inventions the International Patent Classification (IPC) code G06N. 
The second category designated as AI-applied inventions is applying AI core inventions 
to various technical fields. Such fields are image processing, speech processing, natural 
language processing, device control/robotics, various diagnoses/detection/prediction/
optimization systems, etc.58 The JPO ‘Recent Trends in AI-related Inventions Report’ 
published in July 2020 reveals that the number of domestic applications for AI-related 
inventions in 2018 reached a total of 4.728 applications, of which 1.527 related to the 
so-called AI-core inventions.59 In 2016, the JPO granted 803 AI-related patents and, si-
multaneously, refused 184 applications for AI-related inventions, meaning that the grant 
rate was about 80%.60 The published statistics of JPO also contains interesting informa-
tion on the application trends in various fields of AI technology.
Statistics published by the JPO Report show that the USA with 5.954 applications 
and China with 6.858 applications for AI core inventions in 2017 are the major appli-
cation destinations in the world.61 These impressive data implicitly indicate that in both 
countries, the US and China, AI-related inventions are, in principle, eligible for patent 
protection.
4.4. State Intellectual Property Office of PR China
Article 2.2 of the Patent Law of the Peoples Republic of China (CPA) of 1984, in its 
version in force, defines the term “‘invention’ as any new technical solution relating to 
a product, a process or an improvement thereof.” Thereby the Examination Guidelines 
of the State Intellectual Property Office (SIPO)62 define “‘a technical solution’ as an 
aggregation of technical means applying to laws of nature to solve a technical prob-
58 JPO Patent Examination Department, 2020, Box at p. 3; see also Comments to Patenting Artificial 
Intelligence Inventions submitted by the Examination Standards Office, Administrative Affairs 
Division, Japan Patent Office of September 2019 to the United States Patent and Trademark Office 
(USPTO), Request for Comments on Patenting Artificial Intelligence Inventions of 21 August 2019, 
Federal Register/Vol. 84, No. 166 of 27 August 2019/Notices 44889.
59 Ibid., Figure 3 at p. 4.
60 Ibid., Figure 6 at p. 7.
61 Figure 14 at p. 18, followed by Korea with 1.228, Japan 803 and EPO with 648 applications (ibid).
62 Ordinance No. 55 of January 2010.
ZZR 2021.indd   21 10. 12. 2021   14:04:45
22
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
lem”.63 Similar to Article 52 (2) EPC, Article 25.1 and 2 CPA excludes from patentabil-
ity scientific discoveries and rules and methods for mental activities. Thereby the SIPO 
Examination Guidelines clarify that a patent shall not be granted if a claim is defined 
by rules and methods for mental activities in the whole contents. In substance, it con-
cerns only rules and methods for mental activities.64 Among the many examples for such 
rules and methods, the Guidelines indicate ‘computer programs per se’. However, the 
Guidelines also state that “if a claim in its whole contents contains not only matter of 
rule or method for mental activities but also technical features, then the claim, viewed 
as a whole, is not a rule or method for mental activities, and shall not be excluded from 
patentability under Article 25”.65 Under the title ‘Some Provisions on Examination of 
Invention Applications Relating to Computer Programs’, Chapter 9 of Part II of the 
SIPO Examination Guidelines offers important details related to the patentability of 
computer programs and the computer implemented inventions. First, ‘computer pro-
grams per se’ are defined as meaning “a coded instruction sequence which can be exe-
cuted by a device capable of information processing, e.g., a computer, so that certain 
results can be obtained, or a symbolized instruction sequence, or a symbolized statement 
sequence, which can be transformed automatically into a coded instruction sequence. 
Computer programs per se include source programs and object programs”. Moreover, 
the Guidelines clarify that the term the invention relating to computer programs refers to 
solutions for solving the problems of the invention which are wholly or partly based on 
the process of computer programs and control or process external or internal objects of a 
computer by the computer executing the programs according to the process mentioned 
above.66 A key statement in the Guidelines is the clarification that
“If the solution of an invention application relating to computer programs involves 
the execution of computer programs to solve technical problems, and reflects tech-
nical means in conformity with the laws of nature by computers running programs 
to control and process external or internal objects, and thus technical effects in 
conformity with the laws of nature are obtained, the solution is a technical solution 
as provided for in Article 2.2 and is the subject matter of patent protection.”67
As such patentable inventions, the Guidelines offer inter alia the following examples: 
‘A method for controlling a die forming process of rubber’, ‘A method for enlarging stor-
age capacity of mobile computing devices’, ‘A method of removing image noise’, or ‘A 
63 Part II, Chapter 1.2, adding that “usually technical means are embodied as technical features”. 
Further that “a solution that does not adopt technical means to solve a technical problem and there-
by does not achieve any technical effect in compliance with the laws of nature does not constitute 
a subject matter” under Article 2.2 CPA (ibid.)
64 Part II, Chapter 1, Section 4.2 (1).
65 Ibid., Section 4.2 (2).
66 Part II, Chapter 9.1.
67 Ibid., under Section 2 (2).
ZZR 2021.indd   22 10. 12. 2021   14:04:45
23
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
method of measuring liquid viscosity using computer programs’.68 A further approxima-
tion to the European and Japanese handling of patentability of computer-implemented 
inventions, i.e. also AI-related inventions, China has reached with the SIPO Ordinance 
74 of February 2017, revising the 2010 Examination Guidelines, which entered into 
force on April 1, 2017.69 Section 5.2 of Chapter 9 of Part II of the Guidelines sets forth 
that claims relating to computer programs may be drafted as a process or product claims, 
for instance (not anymore meaning), the apparatus for executing the process. No matter 
what kind of claim is drafted. “If it is drafted as an apparatus claim, the various compo-
nent parts and the connections among them shall be specified, the said component parts 
may include not only hardware but also programs”.70 In the past, such claims were not 
accepted in SIPO practice.71
4.5. United States Patent and Trademark Office
The Patent Act of the USA, the other “major application destination in the world” for 
AI-related inventions, excels by several particularities. Firstly, § 101 of 35 U.S.C. (1952) 
defines the subject matter eligible for patent protection. It provides:
“Whoever invents or discovers any new and useful process, machine, manufacture, 
or composition of matter, or any new and useful improvement thereof, may obtain 
a patent therefore, subject to the conditions and requirements of this title”.
The US Supreme Court case law provided, however, three specific exceptions to § 101 
broad patent-eligibility principles: “laws of nature, physical phenomena, and abstract 
ideas”.72 Based on these exceptions, the Supreme Court, at least since the 1930s, has 
rejected claims related to, e.g. a scientific truth, or the mathematical expression of it,73 an 
algorithm itself,74 or a mathematical formula.75 Although the patent-eligibility issue in a 
number of judicial disputes addressing the patentability of computer programs, business 
methods, etc., had to be resolved, the Supreme Court only in two more recent decisions 
set rules for the patent eligibility assessment of interest. In 2012, in Mayo v. Prometheus, 
where the patentability of a method for optimizing the therapeutic efficiency of a drug 
68 Ibid., under Section 3 (2), Examples 4-7.
69 Cf. Amendments to Chapter 9, Part II (c) Amendments to Section 5.2, and the Appendix: 
Comparison table of Part II, Chapter 9 between the revised Guidelines of 2017 and those of 2010.
70 Ibid.
71 Cf. for more St. Yang.
72 Cf. US Supreme Court in Diamond v. Chakrabarty, 447 U.S. 303, 309 (1980), with further refer-
ences.
73 Cf. Mackay Radio & Telegraph Co. v. Radio Corp. of Am., 306 U.S. 86, 94 (1939).
74 Cf. Benson, 409 U.S. 63, 71-72 (1972).
75 Cf. Parker v. Flook, 437 U.S. 584, 594 (1978), Diamond v. Diehr, 450 U.S. 175, 191 (1981), and 
Bilski v. Kappos, 561 U.S. 593, 611 (2010).
ZZR 2021.indd   23 10. 12. 2021   14:04:45
24
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
was at issue, the Court observed that abstract ideas, laws of nature, and natural phenom-
ena “are the basic tools of scientific and technological work” and that “monopolizing 
these tools by granting patent rights may impede innovation rather than promote it.”76 
The Court also set forth a framework for distinguishing patents that claim laws of nature, 
natural phenomena, and abstract ideas from those that claim patent-eligible applications 
of those concepts. The Court held that in applying the § 101 exception a distinction 
must be made between patents that claim “‘building blocks’ of human ingenuity and 
those that integrate the building blocks into something more, thereby ‘transform[ing] 
them’ into a patent-eligible invention”.77
In 2014, in Alice v. CLS Bank in which the Supreme Court had to decide wheth-
er patent claims designated to facilitate the exchange of financial obligations between 
two parties by using a computer system as a third-party intermediary,78 reconfirmed the 
principles established in Mayo, and distilled from Mayo, a two steps patent-eligibility 
test. Thereunder, first, it is to determine whether the claims at issue relate to one of the 
patent-ineligible concepts, i.e., laws of nature, natural phenomena, and abstract ideas. If 
so, the Court asks, “[w]hat else is there in the claims before us”? To answer that question, 
the Court considers “the elements of each claim both individually and ‘as an ordered 
combination’ to determine whether the additional elements ‘transform the nature of the 
claim’ into a patent-eligible application”.79 Step two constitutes “a search for an ‘inventive 
concept’—i.e., an element or combination of elements that is ‘sufficient to ensure that 
the patent in practice amounts to significantly more than a patent upon the [ineligible 
concept] itself ’”.80 Based on the two steps test, the Court rejected the argument of Alice 
Corp. that its patent claimed a computer-implemented financial settlement system and 
did not fall into the patent-ineligible category of “abstract ideas”. The Court held that the 
claims were patent-ineligible on the basis that generic computer implementation does 
not transform a patent-ineligible abstract idea into a patent-eligible invention. Thus, the 
Supreme Court effectively broadened the scope of ineligible subject matter.
As observed by the Office of the Chief Economist of the USPTO, the Alice decision, 
because the ambiguity in the language of its test standard and the scope of technologies 
involving ‘abstract ideas’ made it difficult to predict how and where the standard would 
be applied.81 The Alice decision created uncertainty in the businesses and legal com-
munities, and after its implementation in the USPTO practice, also led to a significant 
76 Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, 71 (2012).
77 566 U.S. 89 (2012).
78 Alice Corp. Pty. Ltd. v. CLS Bank Int’l., 573 U.S. 208 (2014).
79 573 U.S. 217 (2014), quoting Mayo.
80 Ibid., quoting Mayo.
81 Cf. Adjusting to Alice – USPTO Patent Examination Outcomes after Alice Corp. v. CLS Bank 
International, Office of the Chief Economist IP Data Highlights, No. 3, April 2020, p. 2.
ZZR 2021.indd   24 10. 12. 2021   14:04:45
25
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
increase in the percentage of first office action § 101 rejections for patent applications in 
the so-called Alice-affected technologies.82
In its efforts to bring more clarity in the application of the Alice/Mayo two-part test, the 
only test to use to evaluate the eligibility claims, the USPTO, in 2019, revised the 2106 
Patent Subject Matter Eligibility Guidance (PEG) in its Manual of Patent Examining 
Procedure (MPEP).83 With reference to the case law, the MPEP declares the Alice/Mayo 
test as the only applicable test. The ‘machine-or-transformation-test’, i.e. a test defining a 
‘process’ patent-eligible only if: (1) it is tied to a particular machine or apparatus, or (2) 
it transforms a particular article into a different state or thing,84 although an important 
clue to eligibility, was not a separate test. Instead, that test should be considered “as a 
part of the ‘integration’ determination or ‘significantly more’ determination articulated in 
the Alice/Mayo test”.85 An MPEP Flow Chart on the ‘Subject Matter Eligibility Test for 
Products and Processes’ provides valuable guidance for examiners as well as applicants.
In a Notice of 1 July 2019,86 the USPTO also explained that under the Revised 
Guidance, abstract ideas can be grouped as, e .g . ,  mathematical concepts, certain meth-
ods of organizing human activity, and mental processes. Further, that a patent claim or 
patent application claim that recites a judicial exception is not ‘directed to’ the judicial 
exception if the judicial exception is integrated into a practical application. Lastly, that a 
claim that recites a judicial exception, but is not integrated into a practical application, 
is directed to the judicial exception and must, therefore, in a further step, be evaluated 
as regards the ‘inventive concept’ to determine the subject matter eligibility of the claim.
For a better understanding and more efficient handling of the Alice/Mayo test by 
examiners and applicants, the USPTO, in 2019, published more than 40 hypothetical 
examples intended to illustrate the claim analysis under the 2019 PEG.87 Among them 
are the following two: The first concerns a “Method for Training a Neural Network for 
Facial Detection”, claiming:
“A computer-implemented method of training a neural network for facial detection
comprising:
collecting a set of digital facial images from a database;
82 Ibid., p. 4 and Figure 2.
83 <https://www.uspto.gov/web/offices/pac/mpep/index.html> (last accessed 31 August 2020).
84 Cf. Bilski v. Kappos, 561 U.S. 605 (2010).
85 2106 MPEP [R-10.2019].
86 2019 Revised Patent Subject Matter Eligibility Guidance (PEG), <https://www.federalregister.gov/
documents/2019/01/07/2018-28282/2019-revised-patent-subject-matter-eligibility-guidance> 
(last accessed 31 August 2020).
87 <https://www.uspto.gov/sites/default/files/documents/101_examples_1to36.pdf> (last accessed 
2 September 2020) and <https://www.uspto.gov/sites/default/files/documents/101_examples_
37to42_20190107.pdf> (last accessed 2 September 2020).
ZZR 2021.indd   25 10. 12. 2021   14:04:45
26
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
applying one or more transformations to each digital facial image including mir-
roring, rotating, smoothing, or contrast reduction to create a modified set of digital 
facial images;
creating a first training set comprising the collected set of digital facial images, the 
modified set of digital facial images, and a set of digital non-facial images;
training the neural network in a first stage using the first training set;
creating a second training set for the second stage of training comprising the first 
training set and digital non-facial images that are incorrectly detected as facial im-
ages after the first stage of training; and
training the neural network in a second stage using the second training set.”
According to the analysis offered, the claim recites a series of steps and, therefore, is 
a process, i.e. falls in a statutory category. The claim is patent-eligible, because it does 
not recite any of the judicial exceptions in the 2019 PEG. Neither mathematical rela-
tionships nor concepts, formulas or calculations, nor mental processes – the steps are not 
practically performed in the human mind – are recited in the claim. Finally, the claim 
does not recite any method of organizing human activity such as a fundamental econom-
ic concept or managing interactions between people. There is no need to examine where 
the claimed method is integrated into a Practical Application, or provides an inventive 
concept, i.e. the question to be answered in the second part of the Alice/Mayo test if the 
claim recites a judicial exception.88
The second example relates to “Cryptographic Communications” and claims
“A method for establishing cryptographic communications between a first com-
puter terminal and a second computer terminal comprising:
receiving a plaintext word signal at the first computer terminal;
transforming the plaintext word signal to one or more message block word signals 
MA;
encoding each of the message block word signals MA to produce a ciphertext word 
signal CA, whereby CA=MA
e (mod n);
where CA is a number representative of an encoded form of message word
MA;
where MA corresponds to a number representative of a message and 0 ≤ MA ≤
n-1;
where n is a composite number of the form n=p*q;
where p and q are prime numbers;
where e is a number relatively prime to (p-1)*(q-1); and
88 Example 39, pp. 8–9, <https://www.uspto.gov/sites/default/files/documents/101_examples_37to 
42_20190107.pdf> (last accessed 2 September 2020).
ZZR 2021.indd   26 10. 12. 2021   14:04:46
27
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
transmitting the cypher text word signal CA to the second computer terminal over 
a communication channel.”
The USPTO analysis reveals that the claim falls in a statutory category because it 
recites a series of steps and, therefore, is a patent process. However, because the claim 
recites a mathematical formula or calculation that is used to encode each of the message 
block word signals MA to produce a cypher text word signal CA, whereby CA=MA
e (mod 
n), the claim recites a judicial exception. Consequently, questions from Part 2 of the 
Alice/Mayo test are at issue. To start with, is the judicial exception, recited in the claim, 
integrated into a Practical Application? The USPTO analysis answers this question in 
the affirmative. This is because the combination of additional elements in the claim 
(receiving the plaintext word signal at the first computer terminal, transforming the 
plaintext word signal to one or more message block word signals MA, and transmitting 
the encoded cypher text word signal CA to the second computer terminal over a com-
munication channel) integrates the exception into a practical application. The analysis 
emphasises that the combination of additional elements uses the mathematical formulas 
and calculations in a specific manner that sufficiently limits the use of the mathematical 
concepts to the practical application of transmitting the cypher text word signal to a 
computer terminal over a communication channel. Therefore, the mathematical con-
cepts are integrated into a process that secures private network communications. They 
allow a cypher text word signal to be transmitted between computers of people who do 
not know each other or who have not shared a private key between them in advance of 
the message being transmitted, where the security of the cypher relies on the difficulty 
of factoring large integers by computers. As a result, the claim is not directed to the re-
cited judicial exception and is eligible. The analysis further notes that well-understood, 
routine, conventional subject matter can integrate an abstract idea into a practical ap-
plication. Thus, even though receiving a signal at a first computer, transforming it and 
transmitting the transformed signal to a second computer described in the background 
as being conventional, the Alice/Mayo Part 2 step does not evaluate whether the addi-
tional elements are conventional to determine whether the abstract idea is integrated into 
a practical application. Because the judicial exception recited in the claim is integrated 
into a practical application, there is no need to examine whether the claim provides an 
Inventive Concept. Thus, the claim is patent-eligible.89
Statistics show that the new USPTO Examiner Guidance significantly decreased 
uncertainty in patent examination and markedly reduced the first office § 101 rejec-
tions of patent applications in Alice/Mayo-affected technologies.90 Although the US legal 
technique for assessing the patent eligibility of computer-implemented inventions, to 
89 Example 41, pp. 14–16, <https://www.uspto.gov/sites/default/files/documents/101_examples_37to 
42_20190107.pdf> (last accessed 2 September 2020).
90 Cf. Adjusting to Alice, Figure 4 at pp. 6.
ZZR 2021.indd   27 10. 12. 2021   14:04:46
28
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
which also AI-related inventions belong, differs from that applied in the EPO, JPO and 
SIPO, the result of the assessment may still be quite similar. The 2019 USPTO Examiner 
Guidance has apparently reduced the existing uncertainty and increased the probability 
that the validity of the many patents granted by the USPTO in this area will stand scru-
tiny in US courts.91
5. AI Inventions and Patent Law Paradigms
5.1. Patent Law Paradigms
Irrespective the differences which exist in patent laws of various jurisdictions, the 
rule that the patent right shall belong to the inventor,92 the subject matter patent eligi-
bility standard,93 the rules controlling the requirements which a “statutory invention” 
must fulfil to be patentable,94 and the sufficient/enabling disclosure standard,95 can be 
considered as ‘patent law paradigms’. Every advent of a major technological change has 
since ever challenged the patent law rules controlling its paradigms. To safeguard the 
proper functioning of the patent system as an instrument of economic policy providing 
incentives for and rewarding a broad range of useful human activities,96 legislators and/or 
courts have further developed and adjusted those rules to the needs generated by the new 
technology. However, without substantially altering the underlying paradigms. The ad-
vent of software in the 1960s and modern biotechnology in the 1970/1980s are the most 
recent examples. They resulted in adjusted subject matter patent eligibility standards, 
new provisions providing for the exclusion of specific categories of inventions based on 
policy or ethical considerations and the instrument of the deposit of biological material 
in a recognised depository institution as a complement of the disclosure in the written 
description.97 The fact that AI technology enables machines, which use algorithms, to 
learn iteratively from data and experience and think in concepts and eventually turns 
them into a source of new knowledge (invention?), raises several patent law relevant 
questions. For instance, whether the machine can be an inventor in patent law terms, 
whether a technical solution generated by the smart machine autonomously meets the 
91 B. Higgins, 2019, pp. 3–8, based on his analysis of the case law of the lower US courts asked for 
a clarifying of the existing decisions issued by the USPTO (at p. 7), which is now available and 
should also serve as guidance for the US.
92 Cf., e.g. Article 60 (1) EPC; §§ 101, 116, 35 U.S.C.
93 Cf., e.g. Article 52 EPC; § 101, 35 U.S.C.
94 Cf., e.g. Articles 54-57 EPC; §§ 102 and 103, 35 U.S.C.
95 Cf., e.g. Article 83 EPC; § 112, 35 U.S.C.
96 J. Straus, in: T. Takenaka (ed.), 2013, p. 19.
97 Cf. J. Straus, 2019, pp. 546–554.
ZZR 2021.indd   28 10. 12. 2021   14:04:46
29
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
subject matter patent eligibility standard, what is the yardstick for assessing inventive 
step, what constitutes the relevant prior art in such circumstances?
5.2. AI Inventions and Patent Law Paradigms—Debated by Patent Law 
Scholars
These questions have attracted much attention, especially among US patent law 
scholars. They inspired some of them not only to question the suitability of the para-
digms settled in the US patent law to be applied to AI inventions but even to challenge 
the applicability of the patent system to this type of invention in general. Although the 
analyses and the solutions proposed in the reviewed publications98 markedly differ, they 
have many characteristics in common. They are predominantly US-law-centric; they take 
it for granted that AI has already since long autonomously generated inventions, which 
they call ‘AI inventions’, they do not attempt to either specifically, i.e. with technical 
features, define an AI invention, or delimit it from AI-related inventions, understood as 
a subcategory of computer-implemented inventions. Moreover, they all contest the suit-
ability of current patent law standards for assessing the inventive step (non-obviousness) 
and determining the relevant prior art of AI inventions and are all discussing the topic in 
a broad context at a fairly abstract level using a narrow patent law relevant technical basis.
Prior to turning in some detail to the 2016 article of Professor Ryan Abbott,99 a short 
reference to the article published by Ravid and Liu and that published by Schuster100 is 
made because they represent the most radical turn away from settled patent law para-
digms.
As revealed from the following quote, Ravid and Liu place legal issues concerning AI 
inventions outside the patent law realm:
“We analyze AI systems as autonomous, creative, unpredictable, rational, and 
evolving systems, and argue that these characteristics make justifications such as 
personality theories and incentive/efficiency arguments irrelevant. We conclude 
that one cannot conclusively determine an owner for these rights within the scope 
of patent law. Therefore, the rights fall outside the scope of traditional patent 
law.”101
98 Cf., e.g., R. Abbott, 2016, pp. 1079–1126; R. Abbott, 2019, pp. 2–52; W.M. Schuster, 2018, pp. 
1945–2004; Sh.Y. Ravid and X. (J.) Liu, 2018, pp. 2215–2262; R.D. Clifford, 2018, pp. 25–39. 
For more publications of interest see D. Kim, 2020, p. 443; N.Li-T. Koay, 2020, pp. 399–404; 
T.W. Dornis, 2020, pp. 436–446.
99 R. Abbott, 2016.
100 Both referred to in n. 98 supra.
101 Sh. Y. Ravid and X. (J.) Liu, 2018, p. 2221. However, they implicitly agree with granting patents 
for computer-implemented inventions using AI (ibid).
ZZR 2021.indd   29 10. 12. 2021   14:04:46
30
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
They propose to abolish patent protection for inventions generated by AI altogether 
and suggest promoting innovations and public disclosure of AI inventions through alter-
native tools, such as, for example, first-mover advantages, social recognition of AIs, and 
the use of digital tools against copying and counterfeiting.102 Ravid and Liu argue that 
the new AI realm consists of multiple stakeholders, such as the software programmers, 
data suppliers, trainers/feedback suppliers, owners of the AI system, operators of the 
system, new employers of other players, the government, the investor and even the AI 
system as an autonomous entity,103 whose interests are at stake. They call their solution a 
‘Non-patent model within the AI multiplayer paradigm’.104
Schuster starts his paper by raising the question of whether and to whom grant pat-
ents for AI inventions.105 After analysing the US case law on the legal threshold for 
inventorship and a review of US literature addressing AI and IP law,106 Schuster turns to 
an extensive analysis of the Coase Theorem in general and a detailed ‘Coasean Analysis of 
Patent Ownership’.107 Therein Schuster addresses issues such as ‘Types of Patent Value’, 
‘Patent Valuation as a Function of Market Participation’, ‘Assignment of Patent Rights’, 
and the problematic allocation of patent rights to software companies. Although Schuster 
concedes that further research was needed to answer his initial question, he concludes 
that “Via a Coasean analysis, the rights to AI patents should be allocated to AI users (i.e., 
parties using AI to create new technologies) to maximize economic efficiency.”108 Not 
easy, to say the least, to comprehend this suggestion and place it in the framework of the 
patent system. Schuster does not explain how it could work in practical terms.
Ryan Abbott, using a more traditional approach, opened his 2016 article,109 inter 
alia, with the following statements:
“An innovation revolution is on the horizon. Artificial intelligence has been gener-
ating inventive output for decades, and now the continued and exponential growth 
in computing power is poised to take creative machines from novelties to major 
drivers of economic growth. A creative singularity in which computers overtake 
human inventors as the primary source of new discoveries is foreseeable”.110
102 Ibid., pp. 2222, 2252–2257.
103 Ibid., pp. 2231–2235.
104 Ibid., p. 2252.
105 W. M. Schuster, 2018, p. 1948. As revealed later in his paper, Schuster takes it for granted that AI 
autonomously generates inventions: “Genetic algorithms independently develop new inventions 
by mimicking biological evolution via ‘an iterative process of simulated competition and improve-
ment’.” (pp. 1955–1956).
106 Ibid., pp. 1959–1967.
107 Ibid., pp. 1967–2000.
108 Ibid., p. 2004.
109 Abbott, 2016.
110 Ibid., pp. 1079–1080.
ZZR 2021.indd   30 10. 12. 2021   14:04:46
31
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
“Computers already are generating patentable subject matter under circumstances 
in which the computer, rather than a human inventor, meets the requirements to 
qualify as an inventor (a phenomenon that this Article refers to as ‘computational 
invention’)”.111
“Soon computers will be routinely inventing, and it may only be a matter of time 
until computers are responsible for most innovation”.112
“It [the article] finds that machines have been autonomously generating patent-
able results for at least twenty years and that the pace of such invention is likely 
increasing”.113
The empirical basis for such bold statements seems narrow, and its interpretation is far 
from being clear in patent law terms. Abbott’s first example is two patents granted by the 
USPTO to Stephen L. Thaler. The first for a “Device for the Autonomous Generation of 
Useful Information”114 which Thaler called ‘Creativity Machine’ and widely publicised, 
and the second for a “Neural Network Based Prototyping System and Method”115 which 
reportedly was invented by the first patented invention, i.e. the ‘Creativity Machine’. 
Nevertheless, Thaler listed himself as an inventor, assuming that otherwise, he could not 
get the patent granted. Abbott observes that
“If Dr. Thaler’s claims are accurate, then the Patent Office has already granted, 
without knowing it has done so, a patent for an invention created by a non-human 
inventor - and as early as 1998.”116
The second Abbott’s example, titled ‘The Invention Machine’,117 relies on publica-
tions of John R. Koza.118 Koza and his team reported, on the one hand, that genetic 
programming they had developed succeeded in independently generating results that 
were subject of past patents. And on the other hand, that they got several patents issued 
by the USPTO, e.g. for ‘Methods and Apparatus for Automatic Design of Complex 
Structures Using Genetic Programming’,119 ‘Genetic Programming Problem Solver 
With Automatically Defined Stores, Loops and Recursions’,120 ‘Apparatus for Improved 
111 Ibid., p. 1080.
112 Ibid.
113 Ibid., p. 1081.
114 US Patent 5,659.666 (granted/filed on 13 October 1994).
115 US Patent 5,852.815 (granted on 22 December 1998, see also supra note 19).
116 Cf. R. Abbott, 2016, pp. 1083–1086 with numerous references.
117 Ibid., pp. 1086–1088.
118 J.R. Koza, 2010, pp. 251–284, at, e.g. 265.
119 US Patent 6,360,191 B1 (granted on 19 March 2002).
120 US Patent 6,532,453 B1 (granted on 11 March 2003).
ZZR 2021.indd   31 10. 12. 2021   14:04:46
32
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
General Purpose PID and Non-PID Controller’,121 or ‘Method and Apparatus for 
Synthesis of Controllers’.122 In other words, for inventions generated autonomously by 
AI. As a third example, Abbott mentions IBM’s AI ‘Watson’, whose ‘ Jeopardy!’ discover-
ies in food series were likely patentable.123
After a lengthy discussion about whether computers should qualify as ‘legal inven-
tors’, Abbott answers this question in the affirmative.124 He argues that this would en-
courage innovation under an inventive theory.
“Patents on computational inventions would have substantial value independent 
of the value of creative computers; allowing computers to be listed as inventors 
would reward human creative activity up-stream from the computer’s inventive 
act.”125
Abbot praises 25 January 2005, the day of the US Patent 6,847,851 B1 issue, as 
of historical significance, because with the patent issue, the genetic programming has 
passed the ‘Turing test’, i.e., the examiner did not realise “that he was looking at the in-
tellectual property of a computer”.126 As regards the ownership rights to computational 
inventions, according to Abbott, they should belong to a computer’s owner, who could 
assign them to users or developers.127
Among the most important rules controlling the requirements which a ‘statutory in-
vention’ must fulfil to be patentable is that the invention involves an inventive step, i.e., 
that “having regard to state of the art, it is not obvious to a person skilled in the art”.128 
Thus, the decisive yardstick for assessing non-obviousness is the person skilled in the art, 
a fictitious (hypothetical) figure having the knowledge and expertise in the pertinent 
art at the filing/priority date, to be determined in each individual case according to the 
field of technology. Whereas it still can be questioned whether AI already autonomously 
generates patentable inventions,129 no doubt exists that AI as a tool in many respects and 
to an ever-increasing extent facilitates researchers/inventors in their innovative activities. 
Considering the available enormous computer power as regards the speed, storage of 
121 US Patent 6,847,851 B1 (granted on 25 January 2005). Abbott reports that John Koza in a tele-
phone interview he had with him told him, that he did listed him and others as inventors in this 
patent, based on the advice of the legal counsel, although “the whole invention was created by a 
computer.” (ibid., p. 1088 and footnote 68).
122 US Patent 7,117,186 B2 (granted on 3 October 2006).
123 Abbott, 2016, pp. 1088–1091, also mentioning Watson’s involvement in drug discovery (cf. also R. 
Abbott, 2019, pp. 22–23).
124 Ibid., pp. 1092–1113.
125 Ibid., p. 1104.
126 Ibid., p. 1109.
127 Ibid., pp. 1114–1117.
128 Cf., e.g. Article 56 EPC, 35 U.S.C. § 103.
129 See supra n. 47 and the accompanying text infra 6. (iii).
ZZR 2021.indd   32 10. 12. 2021   14:04:46
33
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
data, etc., which AI algorithms use, this cannot remain without consequences for the 
standards controlling the assessment of the inventive step/non-obviousness requirement. 
Affected by the new circumstances are both the standard for determining the fictitious 
person skilled in the art and the relevant, pertinent art itself. As regards the skilled per-
son, the widespread use of AI (ML, ANN, EA, ETs and GAs)130 implies that also this fic-
titious figure, when confronted with technical solutions involving AI, will also consider 
for his assessment AI tools, available at the priority date for solving the type of problems 
at issue. Moreover, the pertinent, relevant prior art will be affected because it will also 
encompass the data used by the AI assisting the skilled person. As Ralph Clifford ob-
serves: “By becoming commonplace, AI has the power to render inventions predictable 
and obvious”.131 This is because a skilled person—the reference for the non-obvious as-
sessment—assisted by AI will, in any case, dispose of much more background knowledge 
even in unrelated prior art than has traditionally been the case.
Ryan Abbott considers going even a step further. He suggests not only that the per-
son skilled in the art should be using an inventive machine for the non-obviousness 
assessment, instead could even be itself “just an inventive machine”.132 Abbott argues, 
“Having inventive machines replace the skilled person may better correspond with real 
world conditions”.133 The implications, which the replacement of the skilled person by 
inventive machines would have, Abbott explains as follows:
“Thus, inventive machines change the skilled paradigm because once they become 
the average worker, the average worker becomes inventive. As the outputs of these 
inventive machines become routinized, however, they should no longer be inven-
tive by definition. The widespread use of these machines should raise the bar for 
obviousness, so that these machines no longer qualify as inventive but shift to 
become skilled machines – machines, which now represent the average worker and 
are no longer capable of routine invention.
[…]
130 Cf. supra under section 2.
131 R. Clifford, 2018, p. 36.
132 R. Abbott, 2019, p. 2.
133 Ibid., p. 33. This opinion is shared by D. Fabris, 2020, 685–708, who states: “[...], I believe the 
PHOSITA/PSITA should be soon replaced by what I call here a “machine of ordinary skill in the 
art” (MOSITA), which knows, through all the data it incorporates and by means of its computa-
tional capacity, all the existing prior art and all possible correlations, even in unrelated scientific 
fields. In such a scenario, since the AI undertakes R&D activities in a fully autonomous way, the 
level of skill of the human beings employing them becomes irrelevant.” (at p. 692).
ZZR 2021.indd   33 10. 12. 2021   14:04:46
34
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
Taken to its logical extreme, and given there is no limit to how sophisticated com-
puters can become, it may be that everything will one day be obvious to commonly 
used computers.”134
However, Abbott eventually realises that instead of replacing the skilled person with 
the skilled machine, it would be “less of a conceptual change, and administratively easier, 
to characterize the skilled person as an average worker facilitated by technology”.135 That 
way would more explicitly take into account the fact that human researchers’ capabilities 
are augmented with computers. Here Abbott concedes that also in inventions using AI 
significant skills may be required to formulate the precise problem to put to a machine so 
that a person might have a claim to inventorship independent of the machine, or a claim 
to joint inventorship.136 As a firm believer in inventive machines, Abbott closes his exten-
sive deliberations on non-obviousness with the following sentence: “Inventive machines 
may ultimately automate knowledge work and render human researchers redundant”.137
6. Stephen L. Thaler’s—DABUS Patent Applications
6.1. Machine—A ‘Statutory Inventor’?
Inspired possibly by Ryan Abbott’s writings, Stephen L. Thaler, who according to 
the narrative suppressed to list in the US patent applications his “creative machine” as 
an inventor,138 filed on 17 October and 7 November 2018 with the USPTO, EPO and 
the UK Intellectual Property Office (UK IPO) two patent applications without listing 
the inventor.139 Invited by the EPO and UK IPO to remedy the deficiency and designate 
the inventor as set forth in Article 81 and Rule 19 (1) EPC, and Section 13 of the UK 
1977Patents Act, Thaler indicated as an inventor a machine, DABUS (stands for Device 
for Autonomous Bootstrapping of Unified Sentience). Thaler, as an applicant, explained 
that DABUS is a type of connectionist artificial intelligence (AI) from which he had ac-
quired the right to the European patent as an employer, i.e. as a successor in title. Thaler 
declared that the invention had been made by a machine and that the machine “identi-
fied the novelty of its own idea before a natural person did”. Moreover, he argued that 
134 Ibid., p. 34. D. Fabris, 2020, pp. 698 ff, considers an intensified use of the so-called ‘secondary 
considerations’, in particular, the solution to ‘long-felt but unsolved needs’, as a possible remedy for 
overcoming the nonobviousness bar.
135 Ibid., p. 35.
136 Ibid., p. 36.
137 Ibid., p. 51.
138 See supra section5.2, n. 125 and the accompanying text.
139 EP 3 564 144 A1, filed on 17 October 2018 (Application No. 18275163.6), GB 181909.4 for “Food 
Container”, and EP 3 563 896 A1, filed on 7 November 2018 (Application No. 18275174.3), GB 
1818161.0 for “Devices and Methods for Attracting Enhanced Attention”.
ZZR 2021.indd   34 10. 12. 2021   14:04:46
35
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
the machine should be recognised as the inventor, and he, as the owner of the machine, 
was an assignee of any intellectual property rights created by this machine. In line with 
the ideas published by Abbott, Thaler also reasoned that this was in compliance with the 
purpose of the patent system, incentivising disclosure of information, commercialisation 
and development of inventions.140
In the oral hearing before the EPO Receiving Section, Thaler further argued that 
not accepting AI Systems as inventors would exclude inventions made by AI from pa-
tentability and would contradict Articles 52–57 EPC. On 27 January 2020, the EPO 
Receiving Section rejected Thaler’s applications and reasoned that indicating the name 
of a machine (DABUS) does not meet the requirements of Rule 19 (1) EPC. According 
to the EPO, names given to things may not be equated with names of natural persons. 
The latter has not only the function of identifying the person but enable it to exercise the 
rights and forms part of their personality. Things have no right, which a name would al-
low them to exercise.141 Contrary to the arguments brought forward by Thaler, the EPO 
found no support for Thaler’s arguments in the EPC travaux preparatoire.142 The EPO 
emphasised that the EPC vested with the inventor several rights vis-à-vis the applicant 
for or proprietor of a European patent (Articles 60 (1), 62 and 81, Rules 19 (3), 20 (2) 
EPC). However, AI Systems or machines have at present no rights because they lack legal 
personality comparable to natural or legal persons.143 The EPO’s decision noted that the 
EPO Boards of appeal have recognised that the inventor is a natural person. The fact 
that the Boards have not yet had an opportunity to decide whether an entity other than 
a natural person could be regarded inventor does not allow the conclusion that this was 
the case. In support of its decision, the EPO finally remarked that the understanding that 
the inventor is a natural person was shared by national courts of various countries, were 
the standard reported by the majority of the EPC Contracting States, and also followed 
by the patent offices of China, Japan, Korea and the USA.144
140 Facts summarised in the decision of the EPO of 27 January 2020, rejecting the application No. 
18275174.3 (under No. I, 1-5) (EPO Doc. PK23498), <https://www.epo.org/news-events/ 
news/2020/20200128.html> or <https://data.epo.org/publication-server/pdf-document?pn=356 
3896&ki=A1&cc=EP&pd=20191106> (last accessed 17 August 2020).
141 Ibid., Nos. 23 and 24, with references to pertinent provisions of national civil codes.
142 Ibid., No. 25.
143 Ibid., Nos. 26 and 27.
144 Ibid., Nos. 29 and 30 with further references. Thaler’s US DABUS applications were rejected in 
the USPTO in February 2020, because according to the USPTO the term ‘inventor’ in 35 U.S.C. 
§ 100 (f ) means the “individual or […] individuals who invented or discovered the subject matter 
of the invention”. The USPTO also pointed to the case law of the Court of Appeals for the Federal 
Circuit indicating that “the patent laws require that an inventor be a natural person” (cf. I. Ireland 
and J. Lohr, 2020, p. 24).
ZZR 2021.indd   35 10. 12. 2021   14:04:46
36
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
Already on 4 December 2019, the Comptroller of the UK IPO rejected the two 
Thaler’s applications and reasoned that the naming of the machine as inventor did not 
meet the requirements of the Patents Act 1977 because a person—meaning a natural 
and not merely a legal person—must be identified as the inventor.145 Thaler appealed 
the Comptroller’s decision before the England and Wales High Court (Patents Court) 
and brought forward, by and large, the same arguments as before the EPO’s Receiving 
Section. Thaler, in particular, emphasised that patent law also protects the moral rights 
of human inventors and that acknowledging machines as inventors would facilitate this 
function. It was fundamentally wrong and weakened the moral justification of patents 
that, at present, individuals are claiming inventorship of autonomous machine inven-
tions under circumstances in which they have not functioned as inventors. By contrast, 
acknowledging machines as inventors would also acknowledge the work of the machine’s 
creators.146 The Court, Justice M. Smith, rejected all of Thaler’s arguments and inter alia 
based its decision on the holdings of Lord Hoffmann in Yeda Research & Development 
Company Ltd. v. Rhone-Poulenc Rorer International Holdings ([2007] UK HL 43), who 
referred to the inventor as the natural person who came up with the inventive concept.147 In 
view of the controversially debated issue of who is to be acknowledged as inventor in AI 
related/generated inventions, it seems appropriate to quote the final remarks of Justice 
Smith, who stated the following:
“As I have noted, the question of whether the owner/controller of an artificially 
intelligent machine that ‘invents’ something can be said, him- or herself, to be 
the inventor was not a matter that was argued before me. Dr Thaler expressly 
declined to advance that submission not merely because he considered it bad in 
law but more importantly because (in moral terms) he considered that he would 
illegitimately be taking credit for an invention that was not his. Clearly, what ar-
guments are or are not framed in relation to patent applications are matters for the 
applicant. However, I would wish to clarify that I in no way regard the argument 
that the owner/controller of an artificially intelligent machine is the ‘actual deviser 
of the invention’ as an improper one. Whether the argument succeeds or not is a 
different question and not one for this appeal: but it would be wrong to regard this 
judgment as discouraging an applicant from at least advancing the contention if 
so advised.”148
145 Facts reproduced in the decision of the England and Wales High Court (Patents Court) of 21 
September 2020 – Stephen L Thaler v. the Comptroller-General of Patents, Designs and Trade Marks 
([2020] EWHC 2412 (Pat)), para. 7.
146 Ibid., para. 5.
147 Ibid., para. 45 (3).
148 Ibid., para. 52 (2).
ZZR 2021.indd   36 10. 12. 2021   14:04:46
37
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
6.2. The Claimed Food Container—DABUS Autonomous Invention?
For two reasons, some important patent law aspects of Thaler’s, e.g. EP 3 564 144 
A1 (EP ‘144) application remained unaddressed. On the one hand, the EPO, UK IPO 
and the USPTO proceedings focused exclusively on the question, whether a machine 
could enjoy the legal status of an inventor. On the other hand, no reason seemingly ex-
isted to question the compliance of the EP ‘144 description with the Rule 42 (b) EPC. 
This is because, as a matter of principle, a patent grant does neither require the inventor 
to disclose why the invention works nor that he/she specifically describes the repeated 
working of the invention if in view of state of the art this is beyond question. Only the 
facts, which Thaler has disclosed in the UK proceedings, offer the necessary basis for a 
substantive examination of those aspects. In other words, to examine whether DABUS 
autonomously, i.e. without human intervention, designed/invented the claimed food 
container, whether the invention, autonomously designed by DABUS or not, met the 
inventive step requirement, how the person skilled in the art to assess this should be 
determined, and what should be considered the relevant prior art.
Thaler neither disclosed how the problem the “machine” was supposed to solve was 
formulated and by whom, nor how it was introduced into the AI system. He also did 
not disclose what constituted the “general knowledge in the field” (which field, by whom 
provided, from where accessed?) for the training of the algorithm (which algorithm, 
one off the shelf commonly used—by whom selected, or a newly developed, if so, by 
whom?), which as Thaler claims, were not “any special data relevant to the present inven-
tion.” Whereas the disclosure of this information was (or at least seemed) irrelevant in the 
context of the application at hand, it is indeed relevant to the question of whether, where 
and to what extent human intervention was involved in the problem-solution process 
chain, which generated the claimed food container. The facts offered by Thaler do not 
seem to support his claim that DABUS autonomously invented the claimed invention.
6.3. US ‘Creative Machine’ Patent—Relevant Prior Art?
Thaler’s explicit reference to, among others, the US “Creative Machine” patent 
5,659,666 (US ‘666),149 as the model of the DABUS “Machine”, the claimed inventor 
of the food container, raises several interesting and important issues so far not addressed 
in the “DABUS” proceedings. Independent of whether DABUS autonomously or not 
“invented” the food container, the US ‘666, as a potentially relevant prior art for the 
DABUS application, requires a closer investigation.
The US ‘666 relates to a “Device for the Autonomous Generation of Useful 
Information”. To the understanding of this writer, the invention claimed and disclosed 
in US ‘666 falls primarily in the category of “AI core inventions” but also discloses 
149 See supra n. 114 and the accompanying text.
ZZR 2021.indd   37 10. 12. 2021   14:04:46
38
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
practical applications of the teaching in some selected fields. Therefore, it represents a 
“hybrid-type” of an AI invention. Its principal object is “to teach the construction and 
operation of novel means for simulating creativity.150 Its task is, as the US’666 Abstract 
reveals,
“simulating human creativity employing a neural network trained to produce in-
put-output maps within some predetermined knowledge domain, an apparatus 
for subjecting the neural network to perturbations that produce changes in the 
predetermined knowledge domain, the neural network having an optional output 
for feeding the outputs of the neural network to a second neural network that 
evaluates and selects outputs based on training within the second neural network.”
The description states,
“The present device represents a new approach and a new application of ANN’s 
in which the system synthesizes novel plans of action and original designs or cre-
ations. These systems, which we refer to as autonomous systems or ‘creativity ma-
chines’ may perform imaginative feats that extend beyond technological invention 
into the realms of aesthetics and emotions.”151
To show the difference between the claimed invention and the prior art, the descrip-
tion emphasises:
“The inventor has shown that static networks have produced some very novel out-
puts which have been detected within mathematical studies. In all known cases, 
however, they have been isolated by a human operator for their novelty. In con-
trast, the present device autonomously monitors the output of such a network and 
selects emergent concepts on the basis of some predetermined criteria established 
within the policing or patrolling neural network. Such concepts may include pro-
ducing music or musical themes for some purpose, or for designing some device such 
as a coffee mug, or producing a process planning operation, or solving a problem, 
and for many other applications some of which will be described more in detail 
hereinafter.”152
The description and the drawings disclose in detail the entire structure and operation 
of the ANN’s involved. It also offers a detailed description of some embodiments, among 
them, that of a coffee mug design. In this latter respect, the description, inter alia, states:
“Referring now to FIGs. 5-8153 there is shown an embodiment of the subject device 
that can be used in the design and production of devices such as a novel coffee mug 
150 US ‘666, column 2, lines 62–64.
151 Ibid., column 1, lines 24–30.
152 Ibid., column 2, lines 5–18 (emphasis added J.S.).
153 FIG. 5 is a diagram of one embodiment of the object device used in designing and/or producing 
a desired shape for a coffee mug, FIG. 6 is a block diagram of the means employed to produce the 
coffee mug of FIG. 5; FIG. 7 is a view showing one form of the operating members controlled by 
the subject device in the production of the desired shape for a coffee mug; FIG. 8 shows examples 
ZZR 2021.indd   38 10. 12. 2021   14:04:46
39
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
and the like. The subject coffee mug can be aesthetically pleasing and also serve a 
useful function. It is apparent, however, that the subject device is not limited to 
the production of coffee mugs and can be used to produce many other designs and 
shapes including works of art.”154
“In designing a coffee mug, various options should be assembled as to the aesthetic 
and utilitarian preferences and this information should be encoded in the AAC. 
This can be done using a computer code which generates vertically aligned stripes 
of various lengths which together stimulate the profile or potential mug design.”155
The lessons to learn from the US ‘666 patent are, firstly, that the “device”, the dual 
ANNs structure, could at best generate the claimed invention, among others a coffee 
mug of a particular shape/function, “autonomously”, if the notion of the invention is 
reduced to the final run of the “device”. In other words, ignoring the formulation of the 
problem, its input into the “device”, the selection, the provision of/access to the neces-
sary training data, encoding the information on advanced audio-coding (AAC) using an 
appropriate computer code, etc., as decisive parts of the inventive concept.
What applies to the US ‘666 patent coffee mug applies even more to the DABUS 
food container. Thaler may have used the US ‘666 “device” or other patented devices 
referred to by him in the UK proceedings, or a combination of those devices, but could 
generate the claimed food container using them for the autonomous “final run” only 
with the necessary human input.
6.4. Description Requirements in AI Related Applications
In practical terms, more important, however, is the lesson that in times of highly 
developed AI technologies with numerous AI core and AI applied inventions, which can 
be used for assisting/solving technical problems practically across all fields of technology, 
require a thorough re-evaluation of what to consider the relevant prior art. Moreover, 
in connection with this also a reassessment of search and examination strategy and the 
notion of the fictitious person skilled in the art seems necessary. The DABUS food con-
tainer applications in connection with the facts presented by Thaler in UK proceedings 
demonstrate this in an exemplary way.
The description in the DABUS EP ‘144 application departs from the disadvantages 
of prior art containers for food products. It asserts the invention is particularly suitable, 
of acceptable and unacceptable coffee mug designs (ibid., column 3, lines 31-40; see also column 
7, lines 18–48).
154 Ibid., column 7, lines 5-14.
155 Ibid., column 7, lines 49-54.
ZZR 2021.indd   39 10. 12. 2021   14:04:46
40
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
but not limited to, containers for liquids, such as beverages and other flowable prod-
ucts.156 The description then continues:
“According to an aspect of the present invention, there is provided a food or bev-
erage container comprising: a wall defining an internal chamber of the container, 
the wall having interior and exterior surfaces and being of substantially uniform 
thickness; wherein the wall has a fractal profile with corresponding convex and 
concave fractal elements on corresponding ones of the interior and exterior surfac-
es; and wherein the convex and concave fractal elements form pits and bulges in 
the profile of the wall.”157
The description of the preferred embodiments introduces as follows:
“The concept disclosed herein makes use of a fractal profile for the wall of the 
container, which has been found to provide a number of advantageous charac-
teristics when applied to a container particularly for food and beverage products. 
The skilled person will appreciate that the profile of the wall will not be of pure 
fractal form but will have a form dictated by practical considerations such as the 
minimum practical or desirable size of its fractal components. Nevertheless, the 
relationship between elements of the profile is fractal in nature. In practical em-
bodiments, the fractal container may exhibit a fractal interpretation over two or 
more size scales.”158
Because the person skilled in the art can realise the invention claimed in EP ‘144 based 
on the description and drawings, the enabling disclosure requirement is no problem and 
no additional information required. As a consequence that a patent grant is not condi-
tioned upon disclosure of the way how the inventor arrived at the invention claimed, 
the EP ‘144 contains no information which could establish a link to the US ‘666 patent 
document or to other US patents referred to by Thaler in the UK proceedings, as possible 
alternative devices.159 Also, the International Patent Classification (IPC) assigned by the 
USPTO to the US ‘666 “device” “autonomously” designing among others coffee mugs, 
i.e. beverage containers, does not suggest any relevance as potentially relevant prior art 
(published on 17 October 1997) for EP’ 144, filed on 17 October 2018. The US ‘666 
invention is classified in the IPC Cl. G 06E 1/00 (devices for processing data exclusively), 
G 06 E 3/00 (devices not provided for in-group G 06E 1/00, e.g. for processing analogue 
or hybrid data), and G 06F 15/18 (digital computers in general; data processing equip-
ment in general—in which a programme is changed according to experience gained by 
the computer itself during a complete run; Learning machines, etc.). Therein is no hint 
or suggestion that an AI learning machine such as the dual ANNs structure of US ‘666 
156 [0006] Column 2, line 3-6.
157 [0007] Column 2, lines 7-16.
158 [0019] Column 3, lines 26-47.
159 EP ‘144 does not cite a single document of this kind as reference.
ZZR 2021.indd   40 10. 12. 2021   14:04:46
41
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
specifically also discloses a method for “autonomously designing beverage containers of 
desired shape/function”. The EP ‘144 is classified in the IPC Cl. B 65 D 6/00 (containers 
having bodies formed by interconnecting or uniting two or more rigid or substantially rig-
id, components made wholly or mainly of metal, plastics, wood or substitutes thereof ), B 
65D 8/00 (containers having a curved cross-section formed by interconnecting or uniting 
two or more rigid, or substantially rigid, components made wholly or mainly of metal, 
plastics, woods or substitutes thereof ) and B 65D 21/02 (containers specially shaped or 
provided with fittings or attachments, to facilitate nesting, stacking or joining together).
The situation described reveals some deficiencies of the patent system as currently 
practised. Even to an ordinary person “in the art”, as this writer, it appears that an ade-
quate assessment of the inventive step involved in the DABUS food container invention 
should not ignore the US ‘666 patent document. Most probably also not those patent 
documents, to which Thaler referred to in the UK proceedings, but not examined here. 
An “artificial neural network discovery device” of the US ‘666 patent type, not only 
could, according to Thaler, did “design” the food container having a fractal profile. In the 
context of the inventive step requirement assessment, the decisive questions are, firstly, 
was it obvious to consider the US ‘666, an AI core invention whose use for designing, 
inter alia coffee mugs, i.e. beverage containers, had been prior art since August 1997, for 
designing beverage containers, and, secondly, was this obvious to a person skilled in the 
art? The answer to the first question seems relatively simple and should be a “yes”, the 
second question is more difficult to answer. Under the given circumstances, even a skilled 
person in the art facilitated by AI technology tools would hardly identify, e.g. the US 
‘666 patent document as prior art being worth closely considered.160
Independent of whether the “technical” assessment offered by an ordinary person 
stands scrutiny as regards the DABUS case, it should not be questioned that the US ‘666 
and the EP ‘144 patent documents should offer more guidance to the fictitious person 
skilled in the art used by the examiner/court as the yardstick when assessing the inventive 
step requirement. Considering that the US ‘666 document claims and discloses a “de-
vice” inter alia for designing beverage containers, the document should also be classified 
in the IPC Cl. B 65D.161 Moreover, applicants claiming inventions generated by the use 
of AI tools should refer in the application to the respective prior art documents, i.e., 
in the DABUS application, to the US ‘666 patent document, but also to other patents 
160 The DABUS food container application offers an opportunity for testing whether a ‘machine of 
ordinary skills’ (MOSITA), which as Fabris assumes, “knows, through all the data it incorporates 
and by means of its computational capacity, all the existing prior art and all possible correlations, 
even in unrelated scientific fields.” (D. Fabris, 2020, p. 692), would, based only on the description 
of EP ‘144, and without any further guidance, detect and consider US ‘666 patent document as 
potentially relevant prior art.
161 A subclass, which covers a broad range of containers.
ZZR 2021.indd   41 10. 12. 2021   14:04:46
42
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
referred to by Thaler in the UK proceedings. This clearly follows from the Rule 42 (b) 
EPC, under which applicants are obliged to indicate in the description:
“[T]he background art which, as far as is known to the applicant, can be regarded 
as useful to understand the invention, draw up the European search report and ex-
amine the European patent application, and preferably cite the documents reflecting 
such art.”162
Thaler, the applicant, expressly not only admitted that he was familiar with the US 
‘666 patent document and other identified US patents, but even that he used a “device” 
of the US ‘666 type for designing/inventing the claimed food container.
7. Authorities in Search for Orientation
The extensive academic debates on the challenges of patent law paradigms by AI 
yielded some fruits and induced initiatives of some IP authorities.
7.1. USPTO
On 21 August 2019, Andrei Iancu, Undersecretary of Commerce for Intellectual 
Property and Director of the USPTO published in the Federal Register a “Request for 
Comments on Patenting Artificial Intelligence Inventions”,163 i.e. inventions that utilize 
AI, as well as inventions that are developed by AI. The USPTO requested comments on 
the following questions:
1. What are the elements of an AI invention? For example: the problem to be ad-
dressed (e.g. application of AI); the structure of the database on which the AI will 
be trained and will act; the training of the algorithm on the data; the algorithm 
itself; the results of the AI inventions through an automated process; the policies/
weights to be applied to the data that affect the outcome of the result; and/or other 
elements.
2. What are the different ways that a natural person can contribute to the conception 
of an AI invention and be eligible to be a named inventor?
3. Do current patent laws and regulations regarding inventorship need to be revised to 
take into account inventions where an entity or entities other than a natural person 
contributed to the conception of an invention?
4. Should an entity or entities other than a natural person or company to which a 
natural person assigns an invention be able to own a patent on the AI invention?
5. Are there any patent eligibility considerations unique to AI inventions?
6. Are there any disclosure-related considerations unique to AI inventions?
162 Emphasis added [J.S.]. Cf. also e.g. R. Schulte and R. Moufang, 2017, § 34 marginal note 213.
163 See supra n. 58.
ZZR 2021.indd   42 10. 12. 2021   14:04:46
43
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
7. How can patent applications for AI inventions best comply with the enablement 
requirement, particularly given the degree of unpredictability of specific AI systems?
8. Does AI impact the level of a person of ordinary skill in the art? If so, how?
9. Are there any prior art considerations unique to AI inventions?
10. Are there any new forms of intellectual property protections needed for AI inven-
tions, such as data protection?
11. Are there any other issues pertinent to patenting AI inventions that we should ex-
amine?
12. Are there any relevant policies or practices from other major patent agencies that 
may help inform USPTO’s policies and practices regarding patenting of AI inven-
tions?
7.2. WIPO
On December 13, 2019, also WIPO started with what it calls ‘Conversation on 
Intellectual Property (IP) and Artificial Intelligence (AI)’.164 WIPO, like the USPTO 
Commissioner, invited a broad range of interested parties to submit comments on is-
sues, which were phrased much along the same lines as those published by the USPTO 
in August of 2019. Taking into account the comments received, which in particular 
criticised the lack of clarifying definitions for the terms used in the December 2019 
paper, WIPO in the framework of its IP/AI ‘Conversation’, on 21 May 2020, published 
a ‘Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence’.165 The 
revised version defines inter alia, the terms ‘artificial intelligence’, ‘AI-generated’ and 
‘generated autonomously by AI’.
As regards the ‘AI’, WIPO, on the one hand, defines it in a comprehensive sense as
“A discipline of computer science that is aimed at developing machines and sys-
tems that can carry out tasks considered to require human intelligence, with limit-
ed or no human intervention.”166
However, WIPO immediately clarifies that for the purpose of this paper,
“AI generally equates to ‘narrow AI’ which is techniques and applications pro-
grammed to perform individual tasks.”167
The Paper further clarifies that it uses the terms ‘AI-generated’ and ‘generated autono-
mously by AI’ interchangeably and that those terms refer to the generation of output by 
AI without human intervention. Thereby, ‘without human intervention’, meaning that 
AI can change its behaviour during operation to respond to unanticipated information 
164 Doc. WIPO/IP/AI/2/GE/20/1.
165 Doc. WIPO/IP/AI/2/GE/20/1 Rev.
166 Ibid., No. 11.
167 Ibid. According to WIPO, Machine Learning and Deep Learning are two subsets of AI.
ZZR 2021.indd   43 10. 12. 2021   14:04:46
44
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
or events. This, according to the paper, is to be distinguished “from ‘AI-assisted’ outputs 
that are generated with material human intervention and/or direction”.168
Although the rephrased “Questionnaire”, for which WIPO is seeking comments in 
the area of patents, starts with asking whether the law should “define the line between 
AI-generated and AI-assisted output”, and, if so, “how much human input should be 
considered material”, the following questions actually address predominantly issues spe-
cific to AI inventions generated autonomously, i.e. without human intervention. The 
reason is WIPO’s understanding that inventions in which AI assists inventors in the 
inventive process or constitutes a feature of an invention “might not differ radically from 
other computer-implemented inventions”.169
WIPO has subdivided its questions into five groups, called “issues”. They re-
late to “Inventorship and Authorship”, “Patentable Subject Matter and Patentability 
Guidelines”, “Inventive Step or Non-Obviousness”, “Disclosure” and “General Policy 
Considerations for the Patent System”. Each issue is subdivided into several specific 
questions, which cover a very broad range of aspects. In the case of “Inventorship and 
Authorship”, e.g. whether AI-generated-inventions at all required patent protection or 
a similar incentive system, should they require that a human being be named as the 
inventor or should the law permit an AI application to be named as the inventor, and 
what consequences should the law set forth in either case.170 As regards the patentable 
subject matter, should the law exclude from patent eligibility AI-generated inventions 
or should they be treated the same way as the other computer-implemented inventions, 
or should they require specific provisions? Should AI applications or algorithms be con-
sidered computer programs or software?171 In case of inventive step or non-obviousness, 
should the present standard of the person skilled in the art be maintained or replaced by 
an AI application trained with specific data from a designated field of prior art? Should 
AI-generated content qualify as prior art?172 As regards the disclosure, whether in the 
case of machine learning, it should suffice to disclose the initial algorithm, “where out-
comes may change depending on the input data and the algorithm adjusts the weights 
associated with neuron connections to reconcile the differences in actual and predicted 
outcome?” Further, “Would a system of deposit for AI applications or training data, 
similar to the deposit of microorganisms, be useful?”.173
7.3. Feedback to the USPTO Request for Comments
168 Ibid., No. 12.
169 Ibid., No. 16.
170 Cf. for details ibid, No. 17.
171 For more see ibid, No. 19.
172 Cf. ibid., No. 20.
173 Cf. ibid., No. 21.
ZZR 2021.indd   44 10. 12. 2021   14:04:46
45
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
The USPTO received upon its request of August 2019 altogether 99 comments, 
2 from foreign patent offices (EPO and JPO), 9 from Bar associations, 13 from trade 
associations/advocacy groups, 13 from companies, 13 from academia, 2 from law firms, 
14 from practitioners (other than firm or academia submissions), and 33 from individ-
uals.174 As the most general outcome of the comments received, the synthesis of the re-
sponses prepared by the USPTO experts reveals that according to the majority of public 
commenters, the current state of the art is limited to ‘narrow’ AI, and AGI has not yet 
arrived. Moreover, that the “current AI could neither invent nor author without human 
intervention”, and “that human beings remain integral to the operation of AI”. Finally, 
that the existing US intellectual property laws are calibrated correctly to address the evo-
lution of AI.175 In detail, the commenters expressed the following views.
The majority of replies to Question 1, ‘What are the elements of an AI invention?’, 
in line with the JPO practice,176 distinguished two AI inventions categories. On the one 
hand, inventions that embody an advance in the field of AI (e.g. a new neural network 
structure of an improved ML model or algorithm), and on the other hand, inventions 
that apply AI (to a field other than AI).177 Some commenters added as a separate category 
“Inventions that may be produced by AI itself.”178
As regards Questions 2 and 3, addressing the eligibility of a natural person as an inven-
tor in AI inventions, the feedback of the vast majority of public commenters was overall 
consistent. They asserted that the current inventorship law was equipped to handle the 
inventorship of AI technologies.179 Moreover, that there is no need for revising patent 
laws and regulations on inventorship to account for inventions in which an entity or 
entities other than a natural person contributed to the conception of an invention.180 The 
USPTO itself clarified that
“The use of an AI system as a tool by a natural person(s) does not generally pre-
clude a natural person(s) from qualifying as an inventor (or joint inventors) if the 
natural person(s) contributed to the conception of the claimed invention. That is, 
the activities by a natural person(s) that would ordinarily qualify as a contribution 
to the conception of an invention are unaffected by the fact that an AI system is 
used as a tool in the development of the invention. For example, depending on 
the specific facts of each case, activities such as designing the architecture of the 
AI system, choosing the specific data to provide to the AI system, developing the 
174 USPTO, Public Views on Artificial Intelligence and Intellectual Property Policy, Washington D.C. 
(USPTO Public Views), October 2020, Table 1 at p. i.
175 Ibid., p. ii–iii.
176 See supra section 4.3, text accompanying n. 58.
177 USPTO Public Views, p. 1 and footnote 4.
178 Ibid., p. 2 and footnote 5.
179 Ibid., p. 3 and footnote 12.
180 Ibid., p. 5 and footnote 24.
ZZR 2021.indd   45 10. 12. 2021   14:04:46
46
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
algorithm to permit the AI system to process that data, and other activities not 
expressly listed here may be adequate to qualify as a contribution to the conception 
of the invention.”181
Some commenters expressed the view that, once AI has reached the step of AGI, the 
USPTO should take up the issue again.182
In response to Question 4, the vast majority of respondents expressed the view that 
only a natural person or a company, based on assignment, as set forth in the current US 
law, were the owner of a patent or an invention.183 However, the Intellectual Property 
Owners Association (IPO) argued, “a natural person who trains the AI process that cre-
ates an AI-generated invention should be able to be an owner”.184 IBM argued that once 
machines become capable of inventions, “their inventions and the corresponding patents 
should be owned by those that own them”; and Siemens suggested that the right of the 
inventors be extended “to legal persons controlling the Al systems”.185
According to the feedback received by the USPTO to Question 5, whether there were 
patent eligibility considerations unique to AI inventions revealed the understanding that 
AI inventions do not require any different treatment than other computer-implemented 
inventions.186 In other words, a treatment in line with the USPTO current practice. As 
to the problems, which may result from the Alice/Mayo doctrine, the USPTO referred 
to one commenter who noted that “complex algorithms that underpin AI inventions 
have the ability to yield technological improvements”.187 The USPTO further added that
“claims directed to an abstract idea will still be patent-eligible if the additional 
claim elements, considered individually or as an ordered combination, amount to 
significantly more than the abstract idea so as to transform it into the patent-eligi-
ble subject matter.”188
In its introductory comment on disclosure-related considerations (Question 6), the 
USPTO emphasised that under 35 U.S.C. § 112(a), applications for AI inventions that 
include claims to computer-implemented inventions that recite functional language,
“should provide sufficient detail in the specification regarding the hardware, as 
well as software, to show that the inventor had possession of the full scope of the 
claimed invention. In particular, the specification should disclose the computer 
and the algorithm (e.g., detailed steps or procedures, formulas, diagrams, and/or 
181 Ibid., p. 5.
182 Ibid., p. 6 and footnotes 27-30.
183 Ibid., p. 7 and footnote 35.
184 Ibid., p. 7, footnote 36.
185 Ibid., p. 7, footnote 37.
186 Ibid., p. 8.
187 Ibid. (for the Alice/Mayo doctrine see supra section 4.5, text accompanying footnotes 76–85).
188 Ibid.
ZZR 2021.indd   46 10. 12. 2021   14:04:46
47
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
flowcharts) that perform the claimed function in sufficient detail such that one 
of ordinary skill can reasonably conclude that the inventor possessed the claimed 
subject matter.”189
Despite the complexity of AI inventions, according to the USPTO, the majority of 
commenters “shared the sentiment that there are no unique disclosure considerations for 
AI inventions”.190 The USPTO only referred to the comment of IBM, noting that “AI 
inventions can be difficult to fully disclose because even though the input and output 
may be known by the inventor, the logic in between is in some respects unknown”.191 
Thus, according to the USPTO, the characteristics of AI learning systems may “drive 
further discussions”.192
Less pronounced were the commenters regarding Question 7 dealing with the enable-
ment requirement of 35 U.S.C. § 112(a), which, under the settled case-law, constitutes 
a patentability requirement distinguished from that of written description. Whereas the 
first requires that the inventor possessed the claimed subject matter at the filing/priority 
date, the second is satisfied when the specification teaches one of ordinary skill in the art 
how to make and use the full scope of the claimed invention without undue experimen-
tation.193 Based on a decision of the Court of Appeals for the Federal Circuit (CAFC), In 
re Wands,194 the MPEP has developed the so-called Wands factors. They include: breadth 
of claims, nature of the invention, state of the prior art, level of one of ordinary skill, 
level of predictability in the art, amount of direction provided by the inventor, existence 
of working examples, and quantity of experimentation necessary to make or use the in-
vention based on the content of the disclosure.195 According to the MPEP,196 the amount 
of guidance or direction is inversely related to the amount of knowledge in state of the 
art, as well as the predictability in the art. The feedback of commenters to this question 
did seemingly not provide for a clear majority view. The USPTO, e.g., refers specifically 
to the IBM comment noting that some AI inventions may operate in a black box be-
cause there is an “inherent randomness in AI algorithms”.197 Moreover, to the reply of 
189 Ibid., p. 9.
190 Ibid.
191 Ibid., p. 10 and footnote 49.
192 Ibid.
193 See MPEP § 2164.01.
194 858 F.2d 731 (Fed. Cir. 1988).
195 See MPEP § 2164.01 (a).
196 § 2164.03.
197 Ibid., p. 11 and footnote 58. Ibid, p. 11 and footnote 58. As regards the ‘black box’ issue, Ménière 
and Pihlaiamma, two high EPO officers, argue that the inability of users to explain why the AI 
algorithm favours a specific decision in case of inventions based on iterative learning capability, 
does not present a challenge for meeting the disclosure requirement under Article 83 EPC. Where 
the decision-making process takes place in a ‘black box’ it were, as a rule, possible for the expert to 
ZZR 2021.indd   47 10. 12. 2021   14:04:46
48
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
the American Intellectual Property Law Association (AIPLA) stating that “most current 
AI systems behave in a predictable manner and that predictability is often the basis for 
the commercial value of practical applications of these technologies”.198 Finally, to the 
response from Genentech, suggesting that because of the greater degree of unpredicta-
bility associated with AI-based inventions, the written description requirement and the 
enablement factors from In re Wands should apply.199
As regards the potential of AI to alter the skill level of the hypothetical ordinary 
person skilled in the art and thus affect the bar for the assessment of nonobviousness 
(Question 8), many commenters, among them the EPO, JPO, Genentech, IBM and 
Siemens, asserted such a potential.200 However, numerous commenters, among them 
Novartis and Merck, suggested that the present legal framework for assessing the person 
of ordinary skill in the art were “adequate to determine the impact of AI-based tools in a 
given field” and should remain unchanged.201 The fact that the USPTO verbatim repro-
duced the following comment of an individual commenter
“Just as the existence of test tubes impacts the level of a person of ordinary skill in 
the chemical arts, and just as the existence of general-purpose computers impacts 
the level of a person of ordinary skill in the software arts (and many others), so 
[too] would AI affect the level of skill in the arts where it can be made useful”202
may be understood as a hint at the future practice of the USPTO, meaning that the 
use of pertinent AI tools will become, in appropriate cases, an inherent characteristic of 
the person skilled in the art for his/her assessment of nonobviousness.
The feedback to Question 9, as summarised by the USPTO, may convey the impres-
sion of quite diverging responses. However, the fact that “The majority of commenters 
stated that there were no prior art considerations unique to AI inventions – that current 
standards were sufficient”,203 does not necessarily mean that they entirely disagree with 
the views expressed by minorities. Some minorities’ views opined that there were such 
considerations without specifying them. Others identified them in the generation of the 
reproduce the results of an AI algorithm based on the disclosure of the algorithm and the training 
data, available to the applicant. They add that the EPO practice decides the enabling disclosure 
requirement on case-by-case basis, taking into account also the specifics of the technical field, such 
as biotechnology or chemistry, where the disclosed subject matter has a momentum (see Y. Ménière 
and H. Pihlaiamma, 2019, p. 335).
198 Ibid., p. 11 and footnote 56.
199 Ibid., p. 11 and footnote 59.
200 Ibid., p. 12 and footnote 67.
201 Ibid., p. 12 and footnote 68.
202 Pp. 12-13 and footnote 70.
203 Ibid., p. 13 and footnote 74 (among the respondents were Ericsson, Genentech and Siemens).
ZZR 2021.indd   48 10. 12. 2021   14:04:46
49
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
prior art by AI, potentially resulting in its proliferation204 or difficulty in finding prior 
art, such as source code related to AI.205 On the other hand, the minorities’ comments 
do not imply that they would not regard current standards sufficient. It all depends on 
the understanding of “prior art considerations unique to AI”. That there is a common 
understanding that using AI tools has an impact on relevant prior art transpires from the 
emphasis commonly expressed by the respondents, that training of examiner and pro-
viding examiners with additional resources for identifying and finding AI-related prior 
art were of great importance.206
The replies USPTO received to Question 10, whether a need exists for any new forms 
of IP for protecting AI inventions, e.g. data protection, reflected two opposite views. 
Some commenters viewed the current US IP framework as adequate. Others, however, 
suggested a need also to protect the data associated with AI, particularly ML.207 In its 
reply, AIPLA suggested balancing the competitive advantage of large technology compa-
nies with huge data collections by a mechanism providing access to their data reposito-
ries. Such a mechanism should, on the one hand, protect proprietary rights to the data 
but, on the other hand, enable new market entrants and others to use such data to train 
and develop their AI.208 IBM drew attention to potential IP protection gaps “for the 
trained model and its associated coefficients”.209
In view of the huge investments in human and financial resources of pharmaceutical 
companies in the already successful use of AI tools, it does not come as a surprise that 
they manifested interest in alternative IP protection forms for specific types of data. 
Genentech, for instance, suggested that “some alternative, sui generis form of protection 
might be required to ensure that bioinformatics and other practical applications of AI 
in biotechnology are protected forms of intellectual property”.210 Novartis suggested to 
consider new forms of IP, including an IP right for trained models and “an IP right for 
nonpublic data, where its generation required substantial effort and investment (similar 
to the regulatory data protection [RDP] rights available in [the] industry for proprietary 
clinical and other data submitted to FDA and other regulatory authorities)”, should the 
current system of protection prove inefficient to provide adequate incentives to suffi-
ciently effectuate the promise of AI.211
204 Ibid., p. 14 and footnote 76 (IBM argued: “AI will dramatically expand the scope of prior art avail-
able. First and foremost, AI has the capability of generating a tremendous amount of prior art.”).
205 Ibid., p. 14 and footnote 77.
206 Ibid., p. 14 and footnote 79.
207 Ibid., p. 15.
208 Ibid., p. 15 and footnote 80.
209 Ibid., p. 15 and footnote 81.
210 Ibid., p. 16 and footnote 85.
211 Ibid., p. 16 and footnote 86.
ZZR 2021.indd   49 10. 12. 2021   14:04:46
50
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
As a reaction to Question 11, related to other issues, which the USPTO should ex-
amine, the commenters once more stressed the need for examiner technical training and 
examiners guidance specific to AI.212AIPLA pointed that “AI [created] inventions may 
require substantial changes in traditional legal approaches and frameworks, including 
notions of property, ownership, and other non-IP legal principles akin to the develop-
ment of corporate law as we know it today”.213 Some individual commenters also stressed 
that the USPTO should also “consider the economic and scientific risks of granting 
patents to AI algorithm inventions and basic research that may have the potential to 
become foundational”.214
8. Summary and Outlook
The answer to the question raised in the title of this contribution, whether AI will 
change some patent law paradigms, seemingly depends entirely on whose position one 
follows. On the one hand, the main patent offices seem confident they can adequately 
handle AI core inventions, as well as AI-applied inventions,215 based on the existing le-
gal instruments. Moreover, the patent offices, without exception, accept only a natural 
person as an inventor, and a natural or legal person as a patent owner, without explicitly 
addressing the issue of inventions autonomously generated by AI. Patent authorities 
see no necessity to alter current standards determining either the prior art, the person 
skilled in the art (of ordinary skill), or those of the enabling disclosure requirement. 
Although they acknowledge that AI-generated data may/will increase the amount of data 
forming prior art to search and consider when assessing novelty and inventive step, they 
seemingly assume that the use of pertinent AI tools will become, in appropriate cases, 
an inherent characteristic of the person skilled in the art.216 The results of the USPTO 
2019 Request for Comments show that the majority of public commenters, among them 
many industry-leading players in AI, favourably assess the practice of the USPTO in the 
AI area. According to them, the current state of the art is limited to ‘narrow’ AI, and AGI 
has not yet arrived. They argue that the AI currently could not invent without human 
intervention and “that human beings remain integral to the operation of AI”. Finally, 
that the existing U.S. IP laws are calibrated correctly to address the evolution of AI,217 
in other words, the patent granting authorities and the majority of the main users of the 
patent system in the field of AI, at least for the time being and for the foreseeable future 
212 Ibid., p. 16.
213 Ibid., p. 17 and footnote 90, where also similar comments of the EPO are reproduced.
214 Ibid., p. 17, e.g. footnote 91.
215 For definitions see supra text accompanying n. 58.
216 See supra text accompanying n. 202.
217 See supra text accompanying n. 175, 179–181.
ZZR 2021.indd   50 10. 12. 2021   14:04:46
51
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
do not see AI radically changing the patent law paradigms. They view the system as flex-
ible enough to adapt to changes the AI developments generate in an evolutionary way. 
Available patent statistics seem to confirm these findings.218
Very similar results have been produced by an extensive Report commissioned by 
the EU Commission, accomplished in September 2020,219 aimed at examining the state 
of copyright and patent protection in Europe for AI-assisted outputs in general220 and 
in three priority domains: Science (in particular meteorology), media (journalism) and 
pharmaceutical research, as regards the EPC and the EPO practice. The Report reached, 
inter alia, the following conclusions:
“– The EPC is currently suitable to address the challenges posed by AI technolo-
gies in the context of AI-assisted inventions or outputs.
– While the increasing use of AI systems for inventive purposes does not require 
material changes to the core concepts of patent law, the emergence of AI may have 
practical consequences for national Intellectual Property Offices (IPOs) and the 
EPO. Also, certain rules may in specific cases be difficult to apply to AI-assisted 
outputs and, where that is the case, it may be justified to make minor adjust-
ments.”221
On the other hand, a growing number of academics and some AI (technical) experts, 
like Thaler and Koza—launching trial balloons by filing patent applications with ma-
chines labelled as inventors, argue precisely the opposite. Based on their understanding 
that AI algorithms and AI systems have since long autonomously, i.e. without human 
intervention, generated patentable inventions, they argue that AI requires a more or less 
radical change of all patent law paradigms.222 What they interpret as largely autonomous 
inventive activity of AI-applications in some areas of R&D, where they apparently rep-
resent already daily routine, Tim W. Dornis recently illustrated by an example in which 
the AI activity uses algorithms in the search for pharmaceutically active substances, e.g. 
a therapeutics with specific properties. He argued as follows:
“The person formulates the problem, finding a substance with specific characteris-
tics. The AI in search of the substance. This is carried out for instance through an 
218 See supra section 4.1 and the text accompanying n. 70. The USPTO published more than 27.000 
AI related patent applications since 2017, of which more than 16.000 within the past two years 
(https://www.uspto.gov/initiatives/artificial-intelligence) (last accessed 24 November 2020).
219 J. Allan, 2020.
220 The Report makes it clear from the outset: “While AI systems have become – and will become – 
increasingly sophisticated and autonomous, this Report nonetheless assumes that fully autonomous 
creation or invention by AI does not exist, and will not exist for the foreseeable future.” (Ibid., p. 6).
221 Ibid., p. 120.
222 See supra sections 5.2 and 6.
ZZR 2021.indd   51 10. 12. 2021   14:04:46
52
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
alignment with data banks and simulated algorithm test series as regards biochem-
ical and medicinal effect of potentially suitable substances.”223
Of course, Dornis adds, selecting subsequently among various identified active in-
gredients could constitute a human inventive contribution. However, if the result of AI 
were only a few or only even one suitable active ingredient, the AI alone had delivered 
the solution and performed the inventive activity.224
It transpires from the reference in footnote 80,225 that Dornis for his illustration 
of an autonomous AI invention relied on an article titled “KI entdeckt vielversprech-
endes Antibiotikum” (“AI discovers a very promising antibiotic”) authored by the free-
lance journalist Julia Merlot. Had Dornis consulted the original research paper “A Deep 
Learning Approach to Antibiotic Discovery” co-authored by Jonathan M. Stokes and 19 
other researchers,226 he had realised that the solution in the case he indirectly referred to 
by no means falls in the category of “inventions without inventors” (Erfindungen ohne 
Erfinder) and by no means were public domain.227 A short quotation from the “Cell” 
paper describing the discovery in connection with Figure 1 should suffice to understand 
that the AI has not autonomously discovered the antibiotic and that humans by no 
means were involved only in formulating the problem and after the actual discovery, 
selecting the suitable antibiotic. The authors, potential co-inventors, summarised their 
discovery as follows:
“Our deep neural network model works by building a molecular representation ba-
sed on a specific property, in our case the inhibition of the growth of E. coli, using 
a directed message passing approach. We first trained our neural network model 
using a collection of 2.335 diverse molecules for those that inhibited the growth 
of E. coli, augmenting the model with a set of molecular features, hyperparameter 
optimization and ensembling. Next, we applied the model to multiple chemical li-
braries, comprising >107 million molecules, to identify potential lead compounds 
with activity against E. coli. After ranking the candidates according to the model’s 
predicted score, we selected a list of promising candidates.”228
This clarification and rectification seem necessary because Dornis’ argumentation 
method seems symptomatic for at least a part of those writers who claim that AI autono-
mously generates patentable inventions without going into any technical details. How to 
223 T.W. Dornis, 2020, p. 443 (rough English translation from German, J.S.).
224 Ibid., pp. 443–444.
225 Ibid., p. 443.
226 See supra, n. 29 and the accompanying text.
227 T.W. Dornis, 2020, p. 444.
228 J.M. Stokes et al., 2020, Figure 1 and accompanying text at p. 689. Enlightening is also a comment 
made by Nathan Benaich, 2020. For the decisive human involvement in the recent successful cases 
of AI, e.g. “AIs Direct Search for Materials” and “AI Designs Organic Synthesis”, the analysis made, 
supra text accompanying n. 34–38, is recalled.
ZZR 2021.indd   52 10. 12. 2021   14:04:46
53
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
perform a serious analysis of the role of human intervention in the case of AI used in drug 
discovery or in AI-augmented target identification transpires from the Report “Trends 
and Developments in Artificial Intelligence”. The Report offers a detailed analysis of two 
real-world cases and specifically identifies the decisive steps of human intervention.229
Daria Kim has correctly observed that patent literature asserting AI systems as au-
tonomous generators of patentable inventions “refers to a handful of examples without 
providing or referencing a technical analysis, which could explain how the ‘intelligent 
systems’ were designed, and how the overall computational process leading to an inven-
tion was set up”.230
Considering the broad range of commenters from all over the world who reacted to 
the USPTO Survey, it seems safe to assume that the patent laws in force and the patent 
granting practice in the area of AI, not only in the US but also beyond, is viewed adequate 
and does not require any legislative changes. A consensus seems to exist that only a natural 
person can be an inventor231 and that machines, at least at this stage and for the foreseeable 
time, do not and will not autonomously generate patentable inventions.232 Moreover, that 
patent law provisions adequately regulate the patentability requirements. Whether one 
day “everything” will become obvious, as augured by Ryan Abbott,233 will primarily de-
pend on how the notion of patentable invention is defined. If it is narrowed down to the 
final run in the “problem-solution” chain of an inventive process, an adequately equipped 
fictitious person skilled in the art may reach such a result in the majority of cases.
229 Supra n. 220, pp. 38–43.
230 D. Kim, 2020, p. 444.
231 The complexities of an AI applied invention, however, imply that more often than in the past more 
persons may contribute to the final output of the AI system, and may claim and be entitled to own 
or co-own the invention, thus the patent. That could apply to the programmer(s) of the AI system, 
or/and to the person(s) who trained the machine (those in charge of dataset selection or creation, 
supervision of the learning process or reinforcement training). This could equally apply to the own-
er of the system who may have risked capital to produce the output, including the cost of develop-
ing the AI system, e.g. obtaining the input training data, developing the operational logic model(s) 
and the learning process, as well as turning outputs into applications. The resolution of disputes 
related to inventorship and ownership issues is a matter for national courts applying national laws 
(J. Allan, 2020, p. 105).
232 Also, experts in AI and robotics caution “that characteristics such as ‘autonomous’, ‘unpredict-
able’ and ‘self-learning’ are ‘based on an overvaluation of the actual capabilities of even the most 
advanced robots, a superficial understanding of unpredictability and self-learning capacities [...], a 
robot perception distorted by science-fiction and a few recent sensational press announcements.” 
(Open Letter to the European Commission. Artificial Intelligence and Robotics, para. 2, <http://
www.robotics-openletter.eu> (last accessed 10 August 2020) (emphasis added) (quoted from D. 
Kim, 2020, at p. 444).
233 R. Abbott, 2019.
ZZR 2021.indd   53 10. 12. 2021   14:04:46
54
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
However, the feedback to the USPTO Survey and the cautious findings in the Report 
commissioned by the EU Commission,234 strongly suggest that patent offices should 
intensify their efforts to adequately train examiners and provide them with AI tools 
necessary to perform novelty and inventive step searches effectively in an ever-increasing 
and extremely diverse available prior art. As the analysis of the DABUS application un-
dertaken in this contribution also demonstrates, this will also require new search strate-
gies for examiners, including detailed guidance for awarding the cross technologies IPC 
classifications. In addition to examination guidelines, also amendments of implementing 
regulations may be necessary to provide for stricter disclosure obligations as regards the 
description standards. Applicants like Thaler should be obliged to disclose and cite such 
patent documents as the US ‘666 because only then could examiners effectively detect 
and assess the relevant prior art. The practical consequence of such measures will not 
“raise” the non-obviousness requirement bar, as often observed, rather it will provide for 
the necessary adjustments of the capabilities of the fictitious “person skilled in the art” 
to those of researchers/inventors assisted by AI tools. In other words, make sure that the 
agreed standards remain effective. Because inventors, such as Thaler, are using in their 
search for technical solutions AI tools, which have the power to render inventions pre-
dictable and obvious,235 only restored equality of arms on the side of the “skilled person 
in the art”, the benchmark for assessing inventive step, can prevent granting patents for 
obvious research results.
At last, the issue of the enablement of AI-related inventions deserves some thought. 
As reported, IBM, in its feedback to the USPTO Survey, observed that some AI in-
ventions could operate in a black box “because there is an inherent randomness in AI 
algorithms”.236 The resulting inability of users to explain why the algorithm favours a 
specific decision in case of inventions based on iterative learning capability, according to 
EPO experts, does not present a challenge for meeting the disclosure requirement under 
Article 83 EPC. This was so because the experts, as a rule, can reproduce the results of 
an AI algorithm based on the disclosure of the algorithm and training data available to 
the applicant.237 Nonetheless, it appears necessary that disclosure of such information 
is explicitly required either in the examination guidelines or even in the implementing 
regulations. Access to the training data is a crucial aspect of ensuring the reproducibility 
of an invention. Therefore, the disclosure of such data should be made mandatory.238 
234 J. Allan, 2020.
235 Cf. R. Clifford, 2018, p. 36.
236 Supra n. 34.
237 Y. Ménière and H. Pihlaiamma, 2019, p. 335.
238 As discussions in the scientific literature reveal, reproducibility of AI inventions presents an issue 
of great technical and legal complexity. This because proprietary rights of third parties in the data 
may be involved and constitute an obstacle for their disclosure (cf. B. Haibe-Kains, et al., 2020, pp. 
E14-E16; and S.M. Mckinney, et al., 2020, p. E17).
ZZR 2021.indd   54 10. 12. 2021   14:04:46
55
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
If necessary, the introduction and establishment of a depository system similar to that 
existing for biological material239 may offer a practicable solution for data access.
It transpires from the results of the overall feedback to the USPTO Survey, the position 
taken by the key patent offices and the conclusions reached by the Report “Trends and 
Developments in AI”,240 that the legal framework currently enabling/controlling the grant 
of patents for AI-related inventions is adequate. Nonetheless, also the need for fine-tun-
ing, including some important adjustments to the changing technological environment, is 
felt. However, for achieving the necessary adjustments, no legislative actions are required.
A suggestion made by the Committee on Legal Affairs of the European Parliament in 
its Report on Intellectual Property Rights for the Development of Artificial Intelligence 
Technologies (2020/2015 (INI))241 runs in the directly opposite direction. This Report 
proposes to create an operational and fully harmonised regulatory framework in the field 
of AI technologies in the form of a regulation to avoid fragmentation of the European dig-
ital single market and promote innovation.242 The motives of this suggestion are certainly 
laudable,243 and some reasons are given convincing. Nonetheless, the idea may not stand 
scrutiny. An attempt to successfully draft a regulation, which should take into account, for 
example, the degree of human intervention, the autonomy of AI, the importance of the 
role and the origin of the data244 and wrap these and other sophisticated considerations in 
a legally binding and enduring, but flexible form, seems nearly mission impossible. This 
would lead to protracted and controversial discussions within different EU fora and un-
necessarily politicise the topic. The unavoidable long-lasting discussions would generate 
legal uncertainty, i.e. precisely the opposite of intended and could eventually even fail.245 
Applicants, with rare exceptions, file applications for AI related inventions, with the EPO, 
which ensures a centralised and harmonised patent granting practice. The widely accepted 
EPO case law as regards computer-implemented inventions, including those AI-related, is 
a result of a decision of the EPO Enlarged Board of Appeal of May 10, 2010. Therein the 
Board rejected the Referral of the President as inadmissible because it could not identify 
any inconsistencies in the case-law of different Boards of Appeal as regards the patent-
239 See for details J. Straus, 2019, pp. 546–554, with further references.
240 Supra n. 220.
241 PE650.527v02-00, A9 - 0176/2020, published on 2 October 2020.
242 Ibid., no. 3 at p. 6.
243 The Report critically notes that in the EU Commission’s “White Paper” (supra n.1) “the issue of the 
protection of IPRs has not been addressed by the Commission, despite the key importance of these 
rights” (ibid, no.1 at p.6).
244 Ibid., no. 10, p. 7.
245 As it happened with the Proposal for a Directive of the European Parliament and of the Council 
on the patentability of computer-implemented inventions of 2002, which, after protracted con-
troversial discussions, was rejected by the European Parliament in July of 2005 (COM (2002) 92 
final – 2002/0047 (COD)).
ZZR 2021.indd   55 10. 12. 2021   14:04:46
56
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
ability of computer-implemented inventions, as alleged by the President. However, the 
Board used prudently the opportunity to circumscribe the frame of patent eligibility for 
computer-implemented inventions under the European Patent Convention.246
Prophets predicting that it is “only a matter of time until computers are responsible 
for most inventions”,247 are reminded of the fate of e.g. the 1930 prediction that “possibly 
Edison may be the last of the great heroes of invention.”248 To the great benefit of human-
ity, in reality, numerous great heroes of the invention followed Edison and the prospect 
that many more will follow him, no fantasy. AI technology has an impact on the patent 
law paradigms, but as in case of other revolutionary and disruptive technologies preced-
ing AI, it is an impact leading to evolutionary adaptations of their prerequisites. Whether 
the paradigm that only a natural person can be a “statutory” inventor could one day re-
quire a radical change will largely depend on the evolutionary adaptations of other patent 
law paradigms to the development of AI technologies. The understanding/definition of 
the patentable invention—inventive concept, and the inventive step, the non-obvious-
ness standard will be crucial, because they will control the question, whether a technical 
solution entirely autonomously, i.e. from the scratch generated without any human inter-
vention by an AI tool, can in principle fulfil all the patentability requirements.
Until that stage is reached, academic lawyers developing high flying thought for solv-
ing problems resulting from inventions “autonomously” generated by AI should over-
come their resistance to read and study original research papers reporting on achieve-
ments involving AI and the respective patent documents,249 instead to rely on reports of 
freelanced journalists in scientific or even popular journals.
References
2019 Revised Patent Subject Matter Eligibility Guidance (PEG), Federal Register – The 
Daily Journal of the United States Government, <https://www.federalregister 
.gov/documents/2019/01/07/2018-28282/2019-revised-patent-subject 
-matter-eligibility-guidance> (last accessed 31 August 2020).
Abbott, R. (2016) ‘I Think, Therefore I Invent: Creative Computers and the Future of 
Patent Law’, Boston College Law Rev., 57, 4, pp. 1079–1126.
246 G 3/08, OJ EPO 2011, p.10.
247 R. Abbott, 2016, p. 1080.
248 W. Kaempffert, 1930, p. 30, quoted from R. P. Merges, 2020, p. 111.
249 Reading patent documents, however, does not mean to quote a paragraph from the respective pat-
ent application. R. Deshpande and K. Karmath, 2020, pp. 879–884, for instance, based on what 
Thaler contended in its DABUS USPTO patent application (recited at p. 882), without going 
into any technical details decisive for the assessment, take it for granted that the DABUS machine 
autonomously generated the invention.
ZZR 2021.indd   56 10. 12. 2021   14:04:46
57
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
Abbott, R. (2019) ‘Everything is Obvious’, UCLA L Rev., 66 (2), pp. 2–52.
Allan, J. (ed.) (2020) Trends and Developments in Artificial Intelligence: Challenges to 
the Intellectual Property Rights Framework, Luxembourg: European Union 
Publisher.
Benaich, N. (2020) ‘Why Artificial Intelligence has disappointed in the Pandemic?’, 
Financial Times, 21 September, p. 21.
Bonadio, E., McDonagh, L., and Dinev, P. (2021) ‘Artificial Intelligence as Inventor: 
Exploring the Consequences for Patent Law’, I.P.Q., Issue 1, pp. 48–66.
Bratko, I. (1994) ‘Machine Learning and Qualitative Reasoning’, Machine Learning 14, 
pp. 305–312.
Clifford, R.D. (2018) ‘Creativity Revisited’, IDEA, 59, pp. 25–39.
Committee of Legal Affairs of the European Parliament, Report on “Intellectual Property 
Rights for the Development of Artificial Intelligence” of 2 October 2020, 
Explanatory Statement, pp. 12–13, PE650.527v02-00, A9 – 0176/2020, on 
2 October 2020.
Committee of Legal Affairs of the European Parliament, Report on “Intellectual Property 
Rights for the Development of Artificial Intelligence” of 2 October 2020, 
Explanatory Statement, pp. 12–13 (Doc. PE650.527v02-00, A9 - 0176/2020).
Court of Appeals for the Federal Circuit, In re Wands, 858 F.2d 731 (Fed. Cir. 1988).
Decision of the EPO of 27 January 2020, rejecting the application No. 18275174.3 (un-
der No. I, 1-5) (EPO Doc. PK23498), <https://www.epo.org/news-events/
news/2020/20200128.html> or <https://data.epo.org/publication-server/pdf 
-document?pn=3563896&ki=A1&cc=EP&pd=20191106> (last accessed 17 
August 2020).
Deshpande, K. Karmath (2020) ‘Patentability of Inventions created by AI – the DABUS 
claims from an Indian perspective’, JIPLP, Vol. 15, No. 11, pp. 879–884.
Dornis, T.W. (2020) ‘Künstliche Intelligenz als “Erfinder” – Perspektiven der Disruption 
im Patentrecht’, Part I, Mitteilungen der deutschen Patentanwälte, pp. 436–446.
England and Wales High Court (Patents Court), Decision of the of 21 September 2020 
– Stephen L Thaler v. the Comptroller-General of Patents, Designs and Trade 
Marks ([2020] EWHC 2412 (Pat)), para. 7.
EPO (2020) Patents and the Fourth Industrial Revolution: The Global Technology Trends 
Enabling the Data-Driven Economy, Munich, December.
EPO Enlarged Board of Appeal, G 3/08, OJ EPO 2011, 10.
ZZR 2021.indd   57 10. 12. 2021   14:04:46
58
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
EPO Patent Index 2020 (2020) Boom in Digital Technologies Continues, <https://
www.epo.org/about-us/annual-reports-statistics/statistics/2020/digital-tech-
nologies.html> (last accessed 13 April 2021), p. 4.
EPO Examination Guidelines Part G-II, 3.6, with references to the EPO case law. The 
Committee on Legal Affairs of the European Parliament in its Report on 
Intellectual Property Rights for the Development of Artificial Intelligence 
Technology (2020/2015/INI) of 2 October 2020 (PE650.527v02-00, A9 - 
0176/2020).
European Commission, White Paper on Artificial Intelligence – A European Approach to 
Excellence and Trust, February 2020, COM (2020) 65 final, p. 2.
Fabris, D. (2020) ‘From the PHOSITA to the MOSITA: Will ‘Secondary Considerations’ 
Save Pharmaceutical Patents from Artificial Intelligence?’, IIC 51, pp. 685–
708.
Ferragina, P., and Luccio, F. (2018) Computational Thinking, First Algorithm, then Code, 
Berlin: Springer.
Gen, M., and Lin, L. (2009) ‘Evolutionary Techniques for Automation, in: Nof, Sh. 
(ed.), Springer Handbook of Automation, Berlin: Springer, pp. 487–502.
Greenberg, A. (2020) ‘Protecting Virtual Things: Patentability of Artificial Intelligence 
Technology for the Internet of Things’, IDEA 60 (2), pp. 329–337.
Gurkaynak, G., Yilmaz, I. and Haksever, G. (2016) ‘Stifling Artificial Intelligence: Human 
Perils’, Computer Law & Security Review, 32, pp. 749–758.
Haibe-Kains, B. et al. (2020) ‘Transparency and Reproducibility in Artificial Intelligence’, 
Nature, 586, pp. E14-E16.
Higgins, B. (2019) ‘The Role of Explainable Artificial Intelligence in Patent Law’, 
Intellectual Property & Technology Law Journal, Vol. 31, No. 3, March, pp. 3–8.
Ireland, I., and Lohr, J. (2020) ‘‘DABUS’: The AI Topic that Patent Lawyers Should be 
Monitoring’, Managing IP.com, Autumn, pp. 23–25.
Japan Patent Office Administrative Affairs Division (2019) Comments to Patenting 
Artificial Intelligence Inventions submitted by the Examination Standards 
Office, of September 2019 to the United States Patent and Trademark Office 
(USPTO), Request for Comments on Patenting Artificial Intelligence Inventions 
of 21 August 2019, Federal Register/Vol. 84, No. 166 of 27 August 2019/
Notices 44889.
Jin, H.R. (2018) ‘Think Big! The Need for Patent Rights in the Era of Big Data and 
Machine Learning’, Journal of Intellectual Property and Entertainment Law, 7, 
Spring, pp. 78–110.
ZZR 2021.indd   58 10. 12. 2021   14:04:47
59
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
JPO Examination Guidelines, <https://www.jpo.go.jp/e/system/laws/rule/guideline/
patent/tukujitu_kijun/document/index/03_0100_e.pdf> (last accessed 18 
August 2020).
JPO Patent Examination Department (Electronic Technology) (2020) Recent Trends in 
AI-related Inventions – Report of July 2020.
Kaempffert, W. (2020) Invention and Society, 1930, p. 30, quoted from Merges, R.P. 
‘Patent Markets and Innovation in the Era of Big Platform Companies’, 
Berkeley Technology Law Journal, 35 (1), pp. 53–112.
Kim, K.G. (2016) ‘Deep Learning’, HIR Health Informatics Research 22 (4), pp. 351–354.
Kim, D. (2020) ‘AI- Generated Inventions: Time to Get the Record Straight’ GRUR 
International, 69(5), pp. 443–456.
Koay, N.Li-T. (2020) ‘Artificial Intelligence and Inventorship: An Australian Perspective’, 
JIPLP, 15, pp. 399–404.
Koza, J.R. (2010) ‘Human-Competitive Results Produced by Genetic Programming’, 
Genetic Programming & Evolvable Machines, 11, pp. 251–284.
Macleod, B.P. et al., Self-Driving Laboratory for Accelerated Discovery of Thin-Film 
Materials, <https://arxiv.org/ftp/arxiv/papers/1906/1906.05398.pdf> (last ac-
cessed 5 August 2020).
Marchant, J. (2020) ‘Powerful antibiotics discovered using AI’ Nature News, 20 February, 
<https://www.nature.com/articles/d41586-020-00018-3> (last accessed 5 
August 2020).
Marrafia, M. and Paternoster, A. (2017) ‘Models and Mechanisms in Cognitive Science’, 
in: Magnani, L. and Bertolotti, T. (eds.), Springer Handbook of Model Based 
Science, Berlin: Springer.
Ménière, Y., and Pihlajamaa, H. (2019) ‘Künstliche Intelligenz in der Praxis des EPA’, 
GRUR, pp. 332–336.
Merriam-Webster (2020) Artificial Intelligence, <https://www.merriam-webster.com/
dictionary/artificial%20intelligence> [https://www.perma.cc/24UH-MR3Z] 
(last accessed 5 August 2020).
McKinney, S.M. et al. (2020) ‘Reply to: Transparency and Reproducibility in Artificial 
Intelligence’, Nature, 586, p. E17.
Office of the Chief Economist IP Data Highlights (2020) Adjusting to Alice – USPTO 
Patent Examination Outcomes after Alice Corp. v. CLS Bank International, No. 
3, April, p. 2.
Open Letter to the European Commission. Artificial Intelligence and Robotics, <http://
www.robotics-openletter.eu> (last accessed 10 August 2020).
ZZR 2021.indd   59 10. 12. 2021   14:04:47
60
Zbornik znanstvenih razprav – LXXXI. letnik, 2021
Patent Handbook, Annex B, Chapter 1, No. 1.2.1.2 Points to Note, <https://www.jpo.
go.jp/e/system/laws/rule/guideline/patent/handbook_shinsa/document/in 
dex/app_b1_e.pdf> (last accessed 18 August 2020).
Proposal for a Directive of the European Parliament and of the Council on the patenta-
bility of computer-implemented inventions of 2002, (COM (2002) 92 final 
– 2002/0047 (COD)).
Ravid, Sh.Y., and Liu, X. (2018) ‘When Artificial Intelligence Systems Produce 
Inventions: An Alternative Model for Patent Law at 3A Era’, Cardozo Law 
Rev., 39, pp. 2215–2262;
Schulte, R., and Moufang, R. (2017) Patentgesetz mit EPÜ, 10th ed., Cologne: Carl 
Heymanns Publisher.
Schuster, W.M. (2018) ‘Artificial Intelligence and Patent Ownership’, Wash. & Lee L. 
Rev. 75, pp. 1945–2004;
Segler, M.H.S., Preuss, M., and Waller, M.P. (2018) ‘Planning Chemical Synthesis With 
Deep Neural Networks and Symbolic AI’, Nature, No. 555, p. 604.
Service, R.F. (2019) ‘AIs Direct Search for Material Breakthroughs – Decision Making 
Algorithms Transform How Robots Evaluate and Synthesize Solar Cells and 
More’, Science, 366, p. 1295.
Silver, D. et al. (2016) ‘Mastering the Game of Go with Deep Neural Networks and Tree 
Search’, Nature, 529, p. 484.
Silver, D. et al. (2019) ‘A General Reinforcement Learning Algorithm Masters Chess, 
Shogi and Go Through Self-Play’, Science, 362, pp. 1140–1144.
State Intellectual Property Office of People’s Republic of China (SIPO), Ordinance No. 
55 of January 2010.
Stergiou, C. and Siganos, D. (2014) Neural Networks, <http://srii.sou.edu.ge/neural-net-
works.pdf> (last accessed 5 August 2020).
Stokes, J.M. et al. (2020) ‘A Deep Learning Approach to Antibiotic Discovery’, Cell 180, 
20 February, pp. 689–690.
Straus, J. (2013) ‘Ordre Public and Morality Issues in Patent Eligibility’, in: Takenaka, T. 
(ed.), Intellectual Property in Common Law and Civil Law, Cheltenham (UK), 
Northampton (MA, USA): Edward Elgar Publisher, p. 19.
Straus, J. (2019) ‘Issues Surrounding Deposit and Release of Biological Material for 
Patent Granting Procedures’, EIPR, Issue 9, pp. 546–554.
Straus, J. (2021) ‘Artificial Intelligence – Challenges and Chances for Europe’, European 
Review, January, Vol. 29, No. 1, pp. 142–158.
ZZR 2021.indd   60 10. 12. 2021   14:04:47
61
Joseph Straus – Will Artificial Intelligence Change some Patent Law Paradigms?
Ullman, Sh. (2019) ‘Using Neuroscience to Develop Artificial Intelligence’, Science, 363, 
pp. 692–693.
US Patent and Trademark Office (2020) Manual of Patent Examining Procedure (MPEP), 
Ninth Edition, Revision 10.2019, Last Revised June 2020, <https://www.
uspto.gov/web/offices/pac/mpep/index.html> (last accessed 31 August 2020).
US Patent and Trademark Office (2020) Public Views on Artificial Intelligence and 
Intellectual Property Policy, Washington D.C. (USPTO Public Views), October.
US Patent and Trademark Office, Artificial Intelligence, <https://www.uspto.gov/initia-
tives/artificial-intelligence> (last accessed 24 November 2020).
US Patent and Trademark Office, Subject Matter Eligibility Examples: Abstract Ideas, 
<https://www.uspto.gov/sites/default/files/documents/101_examples_37 
to42_20190107.pdf> (last accessed 2 September 2020).
US Supreme Court, Mackay Radio & Telegraph Co. v. Radio Corp. of Am., 306 U.S. 86, 
94 (1939).
US Supreme Court, Benson, 409 U.S. 63, 71-72 (1972).
US Supreme Court, Parker v. Flook, 437 U.S. 584, 594 (1978).
US Supreme Court, Diamond v. Chakrabarty, 447 U.S. 303, 309 (1980).
US Supreme Court, Diamond v. Diehr, 450 U.S. 175, 191 (1981).
US Supreme Court, Bilski v. Kappos, 561 U.S. 593, 611 (2010).
US Supreme Court, Bilski v. Kappos, 561 U.S. 605 (2010).
US Supreme Court, Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, 71 
(2012).
US Supreme Court,  Alice Corp. Pty. Ltd. v. CLS Bank Int’l., 573 U.S. 208 (2014).
Wikipedia (2020) ‘Evolutionary Algorithm’, <https://en.wikipedia.org/wiki/Evolutio 
nary_algorithm> (last accessed 17 August 2020).
World Intellectual Property Organization (WIPO) (2019a) Technology Trends 2019: 
Artificial Intelligence.
World Intellectual Property Organization (WIPO) (2019b) Conversation on Intellectual 
Property (IP) and Artificial Intelligence (AI), Doc. WIPO/IP/AI/2/GE/20/1.
Yang, St. Amendments to the Guidelines for Patent Examination, <https://www.lexology.
com/library/detail.aspx?g=88cc460e-07d2-4d14-8e54-4f2c907ceec5> (last 
accessed 19 August 2020).
ZZR 2021.indd   61 10. 12. 2021   14:04:47