Christian Posthoff

Artificial
Intelligence
for Everyone
Artificial Intelligence for Everyone
Christian Posthoff

Artificial Intelligence for

Everyone
Christian Posthoff
Department of Computing and Information
Technology
The University of The West Indies
Trinidad, Tobago

ISBN 978-3-031-57207-4 ISBN 978-3-031-57208-1 (eBook)

https://doi.org/10.1007/978-3-031-57208-1

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2024
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

If disposing of this product, please recycle the paper.

Foreword

Artificial Intelligence (AI) is considered to be a key technology by industry and

scientists. A lot of money is spent for the development of AI systems which show
the enormous economic potential of AI.
The AI technology is already known since more than 50 years. The Japanese
fifth generation in the 1960s already covered topics like deductive and inductive
inferences and theorem proving.
However, sufficient computer power and enough memory space were not avail-
able. This changed during the last 10–15 years where big amounts of data and new
algorithms got available and led through a big breakthrough.
AI means the possibility of a computer to solve problems that, done by human
beings, requires human abilities such as thinking, learning, planning, and being
creative. Computers are able to observe their environment and have the ability to
reach a certain target. They can analyze situations that already have been met and
adapt their solutions more and more, until to a level that is higher than the level of
human beings. They are not in a competition to human beings, but they give them
support which very often reduces the required time, and reduces the costs required
to eliminate the problem.
Machine learning shows at present the most successful performance. A computer
system will learn by solving the problem as often as possible based on the outcome
of several situations. It is Important that there is no solution path modelled within the
algorithms; the computer analyzes the data and results on its own. Very successful
is the combination of robotics and intelligent computer systems.
Since machine learning is already a well-established area of research and
implementation (i.e., robotics), several other applications are under consideration
and of growing importance for all kinds of implementation.
For instance
– The financial business (high-speed share trading)
– Healthcare (image recognition-based diagnosis and therapy)
– Environmental modeling and decision support
– Education and training (optimization)

v
vi Foreword

– Infrastructure and urbanization

– Robotics
– Legal system and legal tech.
The discussions about the risk and danger of AI is a continuous topic of the
society in many countries.
Beside the potential of AI, there is significant challenge and danger for this
technology. And a lot of discussions along the society takes place. Examples are
possible: loss of jobs, security risks, lack of transparency, dependency on AI-
decisions, misuse of deep fakes, ethics, and privacy. However, the technology exists
and will generate a lot of progress. It is important that the challenges of using AI
lead to a comprehensive and coordinated approach. A responsible control for the
good of the society will be necessary. Ethic guidelines have to be implemented,
investments in training and education must be taken, a secure and transparent usage
of data is key, and diversity in the development has to be secured.
The book offers a basic knowledge about AI. After a definition of intelligence, it
starts with an extensive tour of the history of AI. It considers both the software and
the hardware aspects.
An overview of the combination of computer science and mathematics. The
presentation is very understandable, especially for non-scientists; they are a sound
foundation for further application.
A larger part of the book presents recent applications. It is very good under-
standable and shows the big range of application areas, and at the same time the
complexity of implementation details is hidden. It offers a good understanding of
future development directions. The progress is and will be enormous in the future.
Big yearly trade shows like the SES at Las Vegas present applications which were
not conceivable in nearly all areas of digitalization.
The book serves as a generally understandable source of information and is very
much recommended. It is important that new opportunities will be presented to the
society continuously in the future, maybe as an extension of a book like this.
The author presents complex facts in a popular science form effortlessly. His
didactic skills are obvious.
The list of references also offers a deeper dive into AI possibilities.
In summary, this is a very informative book for introducing AI where it comes
from and what is the direction for the future.

Joachim Reiss
Preface

The development of programs that are now called intelligent has accompanied
me since my student days. In the 1960s, computer science did not yet exist as a
subject of study. For mathematics students, there was an introductory course called
“Programming of Calculating Automata.” Programming was done in machine code,
and punched cards served as the input medium.
My PhD thesis had the title “The Application of Mathematical Methods in
Communicative Psychotherapy” (1976). It attempted to develop mathematical
methods to visualize and objectify the problems in groups of psychologically
vulnerable patients. Some computer programs supported the evaluation of the data.
Overall, the need for interdisciplinary cooperation could already be seen at this
stage.
In 1982, the Computer Science Department was founded at the TU Karl-Marx-
Stadt (today Chemnitz) in East Germany with a chair “Theoretical Computer

Fig. 1 The
Zeiß—Calculating
Automaton ZRA 1

vii
viii Preface

Science and Artificial Intelligence,” which I held from 1983 to 1993. At first, of
course, it was important to secure the teaching for the 5 years of study up to the
MSc level. The research work led to good results from 1986 on. In 1988, the setup
was essentially complete; everything that belongs to a normal university routine,
i.e., lectures, MSc as well as PhD theses, was in place.
In 1993, I switched to The University of The West Indies in Trinidad &
Tobago. Here the same sequence was followed. The teaching was developed into
a full-fledged course of studies in Computer Science (B.Sc., M.Sc., Ph.D.), and
after that I started again research on problems of artificial intelligence, without a
clear understanding or definition of the area. I simply followed the international
development and concentrated on some areas which I found interesting and useful.
The current discussions about future development are strongly driven by emo-
tions, wishes, fears, and premonitions that are not always based on knowledge and
not useful. But this is not a new phenomenon. There was such a wave in the 1980s
with computer chess. The development was followed with interest until the then
world champion G. Kasparov finally lost one game against the IBM system Deep
Blue in 1996. At first there were lively discussions, but subsequently the scene
calmed down, and a normal scientific-technical development began.
Autonomous driving of vehicles of various kinds is another focus of discussion.
Here, the focus is already on all possible real or fictitious dangers and possibilities,
without taking into account that robots have already been used for several years
without any problems.
There are currently very heated discussions about GPT-4 and similar programs.
This is a model that accepts text input and produces text output. Here one talks
almost exclusively about negative consequences—infringements of copyright, use
of plagiarism, distortion and falsification of facts, restrictions on creativity. From
this, one very often already deduces the necessity of legal steps, the justification of
bans, etc. But all these problems have existed before. The fact that people can also
use any technical progress in a negative way is not a new phenomenon, but it is
people themselves who are to blame.
To excuse the general public, it must of course be said that the media publish a
veritable deluge of opinions ranging from the destruction of humanity by artificial
intelligence to the colonialization of space. Since an ordinary mortal has only
taken a superficial note of the developments in computer science and is not
very knowledgeable, it naturally scares him very quickly, which is completely
inappropriate.
This book should enable many people to discuss and follow these developments
in an informed way and to draw conclusions for their own workplace and to acquire
the necessary new knowledge. This book is intended to be understandable for a wide
range of readers. If one wants to acquire special in-depth knowledge, then one must
resort to corresponding textbooks and courses. Many programs in the most diverse
fields are available online; one can then experiment with them at will.
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 How to Define Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Different Concepts for Defining Intelligence . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Intelligence Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 The Historical Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1 The Dual System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 The Dartmouth Conference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Algorithms and Programming Languages . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.1 The 1930s and 1940s: Logical Calculi . . . . . . . . . . . . . . . . . . . . . . 21
3.3.2 The 1950s: First Modern Programming Languages . . . . . . . . 22
3.3.3 The 1960s and 1970s: Development of New Paradigms . . . 23
3.3.4 Object-Oriented Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.5 The 1980s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.6 Further Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 The Turing Machine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 The First Computers: The Turing Test, the Single-Tape
Turing Machine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.6 LaTeX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.7 The Evolution of Computers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Mathematics on the Computer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1 Microsoft Mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Mathematica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Maple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5 Polynomial and Exponential Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.1 Introductory Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Mathematical Descriptions of Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . 44

ix
x Contents

5.3 Rules and Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5.4 Axioms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.5 Finding Contradictions: Inconsistency Proofs . . . . . . . . . . . . . . . . . . . . . . 55
5.6 Problems from Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.7 Coloring Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6 Game Theory, Cybernetics, and Information Theory . . . . . . . . . . . . . . . . . . 63
6.1 Game Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.2 Cybernetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Information Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7 Machine Learning and Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.1 Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.2 AlphaZero Wins at Chess, Go, and Shogi. . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3 Translation Between Different Languages . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8 Other Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.1 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2 Trade over the Internet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.3 Big Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.4 The Role of Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
9 Digitization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.1 The Basics of Digitization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.2 Some Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
10 Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10.1 Predicting Earthquakes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10.2 Tumors, Alzheimer’s Disease, and Heart Disease . . . . . . . . . . . . . . . . . . 125
10.3 Recognition of Viruses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
10.4 Medical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
10.5 Further Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
10.6 The Basics of Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
10.7 The Processing of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
11 Robotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
11.1 Robots in Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
11.2 Robots Help in Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
11.3 Other Interesting Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
12 Artificial Intelligence in Finance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Contents xi

13 Artificial Intelligence and Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

13.1 Education of AI Professionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
13.2 The Application of Artificial Intelligence in Education . . . . . . . . . . . . 179
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
14 Artificial Intelligence in the Arts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
14.1 Artificial Intelligence and Music . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
14.2 Artificial Intelligence and Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
15 Artificial Intelligence in Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
16 The Development in South and Central America
and the Caribbean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
16.1 Guyana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
16.2 The Caribbean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
16.3 Belize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
16.4 Argentina and Uruguay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
17 Future Developments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
List of Figures

Fig. 1 The Zeiß—Calculating Automaton ZRA 1 . . . . . . . . . . . . . . . . . . . . . . . . vii

Fig. 1.1 The power of computers is constantly increasing . . . . . . . . . . . . . . . . . . 2
Fig. 1.2 The brain works like a computer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Fig. 2.1 The mass of animals ensures the survival of most of them . . . . . . . . 7
Fig. 2.2 The structure of human intelligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Fig. 3.1 Gottfried Wilhelm Leibniz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Fig. 3.2 George Boole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Fig. 3.3 Claude Shannon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Fig. 3.4 A parallel circuit realizes the disjunction .a ∨ b . . . . . . . . . . . . . . . . . . . . 18
Fig. 3.5 A series connection realizes the conjunction .a ∧ b . . . . . . . . . . . . . . . . 18
Fig. 3.6 The Dartmouth Conference 1955 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Fig. 3.7 A Jacquard loom controlled by punch cards . . . . . . . . . . . . . . . . . . . . . . . 21
Fig. 3.8 Konrad Zuse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Fig. 3.9 Niklaus Wirth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Fig. 3.10 ENIGMA—the decoding equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Fig. 3.11 Alan Turing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Fig. 3.12 A single-tape Turing machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Fig. 3.13 A short part of a LaTeX text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Fig. 3.14 Moore’s law for supercomputers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Fig. 3.15 A supercomputer consists of many processors working
in parallel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Fig. 4.1 A mechanical calculator from the 1960s . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Fig. 4.2 The Microsoft Mathematics work screen . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Fig. 4.3 Mathematica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fig. 4.4 Maple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Fig. 5.1 The butterfly effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Fig. 5.2 The Lorenz attractor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Fig. 5.3 Miniaturization of circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Fig. 5.4 Exponential complexity of a family tree . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Fig. 5.5 Non-Euclidean geometry on a sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Fig. 5.6 The four-color problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

xiii
xiv List of Figures

Fig. 5.7 David Hilbert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Fig. 5.8 Four queens on a chessboard, not threatening each other . . . . . . . . . 56
Fig. 5.9 Eight queens on a chessboard, not threatening each other . . . . . . . . 58
Fig. 5.10 Nine queens and a pawn—ten queens and two pawns . . . . . . . . . . . . 58
Fig. 5.11 To be investigated is the sequence 1 – 2 – 3 – 4 – 1 . . . . . . . . . . . . . . . 59
Fig. 5.12 One solution for the matrix .18 × 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Fig. 5.13 There are many different tours of the same cities . . . . . . . . . . . . . . . . . . 61
Fig. 5.14 The Königsberg Bridge Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Fig. 6.1 John von Neumann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Fig. 6.2 Norbert Wiener . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Fig. 6.3 The Information Content approaches 0 for .p = 1 . . . . . . . . . . . . . . . . . 68
Fig. 7.1 Warren McCulloch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Fig. 7.2 Walter Pitts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Fig. 7.3 A possible structure of a perceptron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Fig. 7.4 Frank Rosenblatt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Fig. 7.5 A network of perceptrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Fig. 7.6 A section of a neuron network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Fig. 7.7 Correction of weights from right to left . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Fig. 7.8 The architecture of a deep CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Fig. 7.9 Computer versus Go—World Champion . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Fig. 7.10 A translation into Japanese. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Fig. 7.11 A translation into Chinese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Fig. 8.1 The first stamp of British Guiana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Fig. 8.2 All accessible data is stored . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Fig. 8.3 Waves of warming and cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Fig. 9.1 A Sony digital camera. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Fig. 9.2 A museum for digital art in Tokyo. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Fig. 9.3 Gravitational Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Fig. 9.4 The job profile is also changing in agriculture . . . . . . . . . . . . . . . . . . . . . 119
Fig. 9.5 The future of education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Fig. 10.1 Magnitude 3.6 earthquake in Switzerland . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Fig. 10.2 Magnetic resonance tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Fig. 10.3 Human internal organs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Fig. 10.4 Road traffic control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Fig. 10.5 A typical pixel image. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Fig. 10.6 A black and white representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Fig. 10.7 The use of colors with the value 30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Fig. 10.8 The use of colors with the value 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Fig. 10.9 The use of colors with the value 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Fig. 10.10 Diagonally below is an anamorphic image of the standing man . . 139
Fig. 10.11 Monstrosities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Fig. 10.12 Mae West Lips Sofa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Fig. 10.13 An impossible geometric construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Fig. 10.14 Distortions in all directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Fig. 10.15 A message hidden in the picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
List of Figures xv

Fig. 10.16 Pablo Picasso: The lonely crouching nude . . . . . . . . . . . . . . . . . . . . . . . . . 144

Fig. 10.17 A painting in the style of Modigliani (1884–1920) . . . . . . . . . . . . . . . . 144
Fig. 11.1 A water organ from ancient times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
Fig. 11.2 A drone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Fig. 11.3 Star Trek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Fig. 11.4 A healthy lung and a lung affected by cancer . . . . . . . . . . . . . . . . . . . . . . 163
Fig. 11.5 Black skin cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Fig. 12.1 This information regarding the DAX is offered in real time . . . . . . 172
Fig. 12.2 Benoît Mandelbrot (1924–2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Fig. 12.3 Average trend lengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Fig. 13.1 The 3D replica of a hand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Fig. 13.2 Complicated shapes do not cause difficulties . . . . . . . . . . . . . . . . . . . . . . 180
Fig. 14.1 The architecture of the future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Fig. 15.1 Computers and law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Fig. 16.1 Potato beetles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Fig. 17.1 There are no limits to the imagination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Chapter 1
Introduction

The digital revolution is leading to huge developments in all areas of science

and technology and will continue to change everyone’s lives quite significantly
in the coming decades. Artificial intelligence will intensify the economic and
social transformations already triggered by the first wave of digitization. There
are great opportunities, but risks are also emerging that should certainly not be
underestimated.
At present, one can see that the working speed of computers is constantly
increasing. Gordon Moore (1929–2023) published an article in 1965, in which he
established what is now called “Moore’s Law” for the preceding years (Fig. 1.1).
Using the data available at that time, he described a relationship between time and
the number of electronic components in an integrated circuit; he assumed an annual
doubling and asked what would happen if this continued for the next ten years. The
law was later slightly corrected; one currently assumes a doubling after 18 months.
In the last 20 to 30 years, the discussion about artificial intelligence has increased
in breadth and in depth. There are several reasons for this: first, there are such
surprising and far-reaching results that fundamentally change many things in
science and technology but also in everyday life. This is essentially due to the fact
that the working speed of computers has increased tremendously. The size and the
quantity of the available storage media reach astronomical orders of magnitude.
Many problems had been mathematically modelled or even solved in principle
for quite some time, but they were not computable until now, which precluded
their application. This has now changed quite significantly, and the possibilities are
expanding almost daily (Fig. 1.2).
We take as an example the competition for the largest prime number. On the
Internet, one can read that the number

M82589933 = 282589933 − 1
.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 1

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_1
2 1 Introduction

Fig. 1.1 The power of computers is constantly increasing

Fig. 1.2 The brain works

like a computer

is (probably) a prime number, the largest prime number found so far. Written out, it
would have over 24 million decimal places. The number was found by a participant
in the GIMPS (Great Internet Mersenne Prime Search) project, a network in which a
large number of participants on distributed computers search for so-called Mersenne
primes. These Mersenne numbers are of the form .2n − 1 and are named after the
French mathematician Marin Mersenne (1588–1648). Until 1947, the numbers

p1 = 22 − 1 = 3, p2 = 23 − 1 = 7, p3 = 25 − 1 = 31, p4 = 27 − 1 = 127
.

were known as Mersenne primes.

1 Introduction 3

Then computers began to be used, and the record given above shows the gap that
could be closed by computers. It is impossible for humans to verify such a value, and
the acceptance of such a solution becomes a matter of faith or trust. One must trust
both the programmer and the electronic engineer, and there must be no misbehavior
of any circuit during the calculations. We will encounter this situation of not being
able to check results very often.
The fact that technology is outperforming humans is not new. These problems are
now being discussed so fiercely mainly because the achievements of computers are
now playing a role in areas that were previously the exclusive preserve of humans,
their creativity and their intelligence. The intelligence of humans is more and more
supported by computers or even only effective with their help, which finally gave
rise to the name “artificial intelligence.”
It is probably quite natural that no completeness can be achieved in this
book. This area is already too large, and the number of qualified publications
grows exponentially. For some time now, it has been disintegrating into relatively
autonomous sub-areas:
• Application of data science in combination with AI and machine learning
methods,
• AI in medicine,
• Robotics,
• Speech and language technology,
• Educational technology,
• Innovative factory systems,
• Intelligent networks,
• Agents and simulated reality,
• Interactive machine learning,
• Augmented reality,
• Language technology and multilingualism.
The representation is further complicated by the fact that many areas overlap
and cooperate. It is precisely this cooperation that is an essential feature of artificial
intelligence. The book strives to provide basic knowledge that will objectify the
discussions and relieve some of the creepiness of utopian films. It must also be
understood that research results are a necessary condition for progress; they are
not sufficient until they can be translated into practice embedded in programs. This
difficult relationship between theory and practice has been known for a long time.
Chapter 2
How to Define Intelligence

The basic prerequisite for discussing intelligence is, of course, a healthy human
brain and associated nervous system. It has an enormous complexity: a human
being has about 100 billion brain cells that build up the central nervous system, our
brain, and they are interconnected. The number of these connections is estimated at
100 trillion. One speaks of artificial intelligence when the complexity of a solution
procedure on a computer is in similar dimensions. Microelectronic structures can
also reach this order of magnitude, and this is the fundamental prerequisite for
computer solutions to reach or increasingly surpass humanity.

2.1 Different Concepts for Defining Intelligence

Intelligence has been defined in many ways as the ability

• to think logically or critically,
• to grasp the meaning of something,
• to accept facts and persons,
• for self-awareness,
• for learning,
• for emotional knowledge,
• for reasoning,
• for planning,
• for creativity,
• for problem-solving.
Intelligence is most commonly studied in humans but also in animals and in
plants, although it is disputed whether some of these life forms exhibit intelligence.
Intelligence can also be seen in the behavior of computers or other machines, in
which case it is called artificial intelligence.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 5

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_2
6 2 How to Define Intelligence

In psychology, intelligence is a collective term for cognitive or mental perfor-

mance. The term refers primarily to the ability to use the totality of variously
expressed cognitive abilities to solve a logical, linguistic, mathematical, or meaning-
oriented problem. Since the individual cognitive abilities can vary in degree and
there is no agreement on how to determine and distinguish them, there is no
universally accepted definition of intelligence.
General psychology, differential psychology, and neuropsychology are all
concerned with intelligence. The study of intelligence in the field of general
psychology from the aspect of information processing is now often referred to
as cognitive psychology. This in turn is based on methods and findings of brain
research, developmental psychology, and increasingly also artificial intelligence.
• Individuals differ from one another in their ability to comprehend complex ideas,
adapt effectively to the environment, learn from experience, apply different forms
of reasoning, and overcome obstacles through reflection. Although these individ-
ual differences can be substantial, they are never completely uniform. A person’s
intellectual performance varies on different occasions, in different domains, and
is judged by different criteria. Concepts of “intelligence” are attempts to explain
and organize these complex phenomena. Although considerable clarity has been
achieved in some areas, none of these conceptualizations has yet answered all the
important questions, and none meets with universal approval.
• Human intelligence
Human intelligence is the intellectual achievement of humans characterized by
complex cognitive performance and high levels of motivation and self-awareness.
Intelligence enables humans to remember descriptions of things and to use those
descriptions to guide future behavior. It is a cognitive process. It gives humans
the cognitive abilities to learn, form concepts, understand, and reason, including
the ability to recognize patterns, innovate, plan, solve problems, and use language
to communicate. Intelligence enables people to have experiences and to think.
Intelligence is different from learning. Learning refers to the act of retaining
facts and information or skills and the ability to recall them for future purposes,
while intelligence is a person’s cognitive ability to perform these and other
processes. There have been various attempts to quantify intelligence through
testing and to calculate an intelligence quotient (IQ) based on the results of the
test.
It is controversial whether human intelligence is based on inherited factors
or environmental factors. Inherited intelligence is the theory that intelligence
is fixed at birth and cannot grow. Environmentally determined intelligence is
the theory that intelligence develops throughout life depending on the person’s
environment. An environment that cultivates intelligence challenges the person’s
cognitive abilities.
• Emotional intelligence
Emotional intelligence is the ability to communicate emotions to others in an
understandable way and to accurately read the emotions of others. Some theories
suggest that in addition to accuracy, higher emotional intelligence may lead
2.1 Different Concepts for Defining Intelligence 7

to faster generation and processing of emotions. It is also thought that higher

emotional intelligence helps us manage our emotions, which has a positive
impact on our problem-solving abilities. Emotional intelligence is important for
our mental health and has links to social intelligence.
• Social intelligence
Social intelligence is the ability to understand the social cues and motivations
of others and oneself in social situations. It is thought to be distinct from
other types of intelligence but has relationships with emotional intelligence.
Social intelligence has overlapped with other studies of how we judge others,
the accuracy with which we do so, and why people are considered to have
positive or negative social character. It is disputed whether these studies and
social intelligence are based on the same theories or whether there is a difference
between them, and they are generally thought to be two different schools of
thought.
• Collective intelligence
The individuals who are the cause of the collective intelligence phenomenon are,
in a sense, placed on the level of ants, which have a very limited behavioral and
response repertoire. In the animal world, this swarm intelligence is capable of
ensuring the survival of the majority of individuals in a hostile environment. At
the border between Tanzania and Kenya, many wildebeests gather at certain times
to cross a river. The crocodiles present there may catch and eat some wildebeest,
but the majority crosses the river without problems (Fig. 2.1).
In swarm behavior, also known as herd behavior, there is no central control
for the individuals. In financial markets, investors sometimes tend to behave like

Fig. 2.1 The mass of

animals ensures the survival
of most of them
8 2 How to Define Intelligence

herd in their buying and selling decisions and to invest or disinvest in a trading
object by a majority. Herd behavior is a manifestation of mass psychological
contagion effects and can thus be a cause of financial market crises or economic
crises. Hoarding purchases also exhibit herd behavior, as before natural disasters
or during the COVID-19 pandemic starting in March 2020, when there were shelf
gaps for certain goods (e.g., flour, pasta, toilet paper) in German stores.
Various mass-psychological or market-psychological causes may underlie
herd behavior. Consumers may be driven by the fear of not being able to meet
their needs in the face of shelf gaps if they do not buy immediately. A consumer’s
expectation that other consumers will also hoard after him also pushes him to
hoard purchases. Likewise, his fear that supply shortages may occur in the future
forces him to make purchasing decisions that are not in line with his needs.
Sometimes, consumers’ feelings of powerlessness are also seen as the cause.
This behavior is irrational, especially since food and beverages or toilet paper are
mass products that can be reproduced at any time. In France and Italy, one of the
products affected by hoarding is red wine, a product that cannot be reproduced at
any time.
The consequence of herd behavior is strong price fluctuations of the trading
object concerned. In addition, hoarding purchases accelerate the rotation of
goods and reduce the logistical range. As market behavior, herd behavior is
particularly known among noise traders, who are often guided by herd behavior
and motivated by sentiment or groups to buy or sell into falling ones. This is
what is known as “mood noise.” Rising or falling prices are an indication that
other market participants have previously made the same decision. This noise
can underlie both buy and sell decisions and also hold decisions. Herd behavior
is thus a sign of a lack of market efficiency.
Speculation only becomes problematic for a market when it is no longer
speculated on the basis of fundamental data but when herd behavior sets in. Then
speculative bubbles can arise, which are usually due to herd behavior. Speculative
bubbles can be justified by the expectation of the majority of market participants
of future profit opportunities.
Profit taking can also be based on herd behavior, when a large number of
investors take advantage of a high price level to sell and other investors join
in. The bank run is also a typical herd behavior, as investors observe a perhaps
random mass withdrawal of cash and blindly join it, trusting that it must have
a specific reason; the mass withdrawals eventually culminate in the domino
effect. Investors withdraw their deposit because they fear that, as a result of the
sequential payout principle (“first come, first served”), they will otherwise not
be able to withdraw this deposit because their cash reserves have been depleted.
Consequently, it is rational for any depositor to follow the herd. A bank run is
more likely the less informed bank customers are and the more they overreact.
Hoarding is likely to contribute to the scarcity of certain goods or services and
thus to market tightness by sharply increasing demand. Herd behavior can lead to
self-fulfilling prophecies: If market participants behave in a certain way, this can
cause the fundamentals underlying an investment to change as a result of the herd
2.1 Different Concepts for Defining Intelligence 9

behavior itself: they develop in the direction the herd is taking—consequently, it

is rational not to break away from the herd, which ultimately leads to the expected
outcome [1].
• Competence
The relationship between individual and collective intelligence is further overlaid
by the concept of competence. As an example, we choose Maxwell’s equations:

→ = ρ→
div D
.

div B→ = 0
∂ B→
rot E→ + =0
∂t
∂D →
rot B→ − = j→.
∂t

.E→ is the electric field strength, .D

→ the electric flux density, .H→ is the magnetic
field strength, .B→ the magnetic flux density. The charge density .ρ is the source of
the electric field, and the current density is denoted by .j→.
These four equations mean nothing at all to a normal mortal. Only an expert
can do something with it. This brings into play the concept of competence. Many
people can acquire competence, but only in a few fields, because the path to it
is long and arduous and usually requires university studies. This competence is
necessary for every science. But also all professions and the versatile everyday
life require competence in various fields.
It plays a role in processes where performance is sought from an application-
oriented point of view. Competences are mainly used in connection with the
development of educational standards in order to achieve educational goals.
The following characteristics can be listed:
1. Competencies as general cognitive performance dispositions that enable individ-
uals to accomplish very different tasks.
2. Competencies as context-specific cognitive performance dispositions that func-
tionally relate to specific classes of situations and requirements. These specific
performance dispositions can also be characterized as knowledge, skills, or
routines. They are functionally determined.
3. Competencies in the sense of motivational orientations necessary for accomplish-
ing demanding tasks.
4. Action competence as an integration of the first three concepts, related to the
requirements of a specific field of action.
5. Meta-competencies as the knowledge, strategies, or motivations that facilitate
both the acquisition and the application of specific competencies.
6. Key competencies but relevant to a relatively broad range of situations and
requirements. These include language or mathematical skills.
10 2 How to Define Intelligence

2.2 Intelligence Tests

The question immediately arises, of course, as to how to compare the various forms
of intelligence defined in the previous section. After all, they are not completely
congruent and are naturally expressed differently in each person. For this reason, a
variety of intelligence tests has been defined.
An intelligence test is a psychological diagnostic tool used to measure a person’s
intelligence. There are many different types of intelligence tests. Differences in
performance on intelligence tests are thought to reflect differences in cognitive
performance in daily life. Psychological tests are used as a tool to find an optimal
decision in the context of various diagnostic issues. For example, the assessment of
intelligence can be helpful in predicting career success or career aptitude (personnel
selection or career counseling) or in recommending a school education or choice of
study. Various clinical issues (presence of diseases such as dementia or reduced
intelligence or the application of vocational rehabilitation measures) may also
require assessment of intelligence.
The best known result of some of these tests is the intelligence quotient (IQ).
Because of the danger of making IQ an absolute designation for a person (like height
or weight—i.e., independent of a particular question), this measure is no longer used
in professional language, and other standard scales are used to describe intelligence
performance.
Intelligence tests are often controversial. This is due, among other things, to the
multitude of factors that influence intelligence, a possible link to heredity issues,
and doubts about objectivity. It is very easy to pass the time with intelligence tests
offered on the Internet, but it is not easy to draw a reasonable conclusion from
them. Very often, math comprehension is asked; also the ability to quickly remember
a larger set of figures and to classify them according to not immediately visible
features is helpful. Often, the time available also plays a role. If one performs the
same test with a lower or increased thinking time, the result can be quite different.
Even this brief overview shows that it is very difficult to find a generally valid
and accepted definition of intelligence that meets all requirements. Of course, this
also shows when one wants to speak of “artificial intelligence.”
There is still a fundamental quality to be considered that falls under the umbrella
of many discussions, but which is also part of the concept of competence. Very often,
people say a person can do this or that. But what is actually meant is there are some
people who can do this or that. These are the specialists for a certain activity, and
very often, they are extremely rare. If you want to introduce (intelligent) hardware
and software in a certain area, you cannot make sure early enough that specialists
are available (Fig. 2.2).
Reference 11

Fig. 2.2 The structure of

human intelligence

Reference

1. Gunter Dueck. Schwarmdumm—So blöd sind wir nur gemeinsam, Goldmann 2. Auflage, 2018,
978-3-442-15950-5
Chapter 3
The Historical Development

This chapter will show which mathematical and technical means were necessary so
that intelligent applications could be achieved in a wide variety of fields. It is also, it
should be emphasized here, not at all meaningful to speak of “artificial intelligence.”
There are many applications that are only related to each other in that they are based
on the use of computers.

3.1 The Dual System

Many modern scientific theories related to mathematics find their beginning with
Gottfried Wilhelm Leibniz (1646–1716, Fig. 3.1).
Some of his research findings and initiatives were:
• description of the dual system,
• development of decimal classification,
• plans for a submarine,
• improvement of technology of door locks,
• device for determining wind speed,
• advice to physicians to take fever regularly,
• establishment of a widows and orphans’ fund,
• proof of the unconsciousness of man,
• infinitesimal calculus (integral calculus and differential calculus),
• matrices and determinants,
• invention of the staggered roller for a mechanical calculating machine,
• development of the endless chain for mining ore extraction.
As a friend, professional colleague, and encourager of the European scribes of the
time and author of linguistically influential writings, he was one of the key initiators
in the founding of modern linguistics, particularly Indo-European studies.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 13

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_3
 14 3 The Historical Development

Fig. 3.1 Gottfried Wilhelm

Leibniz (1646–1716)

In the letter to Duke Rudolph August of January 12, 1697, Leibniz describes
and outlines for the first time his idea of the “dual system.” At the center of his
explanations is the Christian belief that one can create everything (1) from nothing
(0), which leads to the famous sentence omnibus ex nihilo ducendis sufficit unum.
Today we are familiar with the fact that any number greater than 1 can serve as
a base for the representation of any number. Currently, the number 10 is used in
human everyday life and the base 2 for all digital devices.
The decimal representation assigns powers of 10 to the positions from right to
left: .. . . 104 103 102 101 100 .

728 = 7 · 102 + 2 · 101 + 8 · 100 .

.

If you use base 10, you need the ten digits .0, 1, 2, 3, 4, 5, 6, 7, 8, 9.
The binary (dual, two-valued) representation of the same number:

728 = 1011011000 = 1·29 +0·28 +1·27 +1·26 +0·25 +1·24 +1·23 +0·22 +0·21 +0·20 .
.

We can see that the dual representation requires only the digits 0 and 1. But since
each power of 2 contributes much less to the value of the number than the powers
of 10, you need a larger number of powers of two, and the representation becomes
much longer.
The addition of dual numbers is quite simple:

0+0 = 0
.

0+1 = 1
1+0 = 1
1+1 = 0 carry 1 to the next position on the left

Multiplication does not cause any difficulties either:

3.1 The Dual System 15

Fig. 3.2 George Boole

(1815–1864)

0·0 = 0
.

0·1 = 0
1·0 = 0
1·1 = 1

The dual system can also be approached from a completely different angle.
The English mathematician, logician, and philosopher George Boole (1815–1864,
Fig. 3.2) founded in his work The Mathematical Analysis of Logic (1847), the first
algebraic logic calculus, which has fundamental importance for the development of
computer technology.
The premises for propositional logic are simple and self-evident:
• Repeating a statement does not change its truth value. A false statement remains
false, and a true statement remains true.
His further conclusion is brilliant: he transfers this fact to arithmetic, which
has worked without error for millennia. He introduces a logical variable x and
formulates:

x 2 = x.
.

This equation has the two solutions .x = 0 and .x = 1. Thus, if arithmetic is free of
contradictions, then a calculus of logics can have only two values.
This also has a (funny) side effect. Repeating a message adds nothing in terms of
truth or falsehood. A true message remains true, and a false message remains false.
16 3 The Historical Development

On these two values, he built the Boolean algebra. Today it is the basis
for the design of microelectronic circuits, but it also plays a very important role
in programming and in artificial intelligence. The following logical functions are
available:
• conjunction .∧,
• disjunction .∨,
• exclusive-or .⊕,
• equivalence .∼
• negation .¬.
The symbol .¬ is rarely used and is usually replaced by a horizontal line .x.

Conjunction und Disjunction

x y .x ∧ y .x ∨ y
0 0 0 0
0 1 0 1
1 0 0 1
1 1 1 1

The symbol .∧ is usually omitted (as is often the case with multiplication). So
we understand ab as .a ∧ b. The use of these operations in logic gave .∧ the name
and: the conjunction of two statements by and is true if either statement is true;
otherwise, it is false.
.∨ is denoted by or: to obtain a true statement, the first statement or the second

statement must be true; it is allowed that both statements are true. We can see that
the equivalence function is equal to 1 when x and y have the same value. The value
of the exclusive-or is equal to 1 if the values of x and y are different. It can be seen
that antivalence and equivalence merge by negation.

Exclusive-or and Equivalence

x y .x ⊕ y .x ∼ y
0 0 0 1
0 1 1 0
1 0 1 0
1 1 0 1

Negation
x .x
0 1
1 0

The comparison of the table for the addition of binary numbers with the table
for the exclusive-or shows that they are exactly the same; therefore, the formula
3.1 The Dual System 17

sign .⊕ is used for them. This is also true for the tables of conjunction and
multiplication of binary numbers. The disjunction and the negation can also be
expressed arithmetically:

.a ∨ b = a + b − a · b, x = 1 − x.

Mathematical modeling of circuits using Boolean algebra was introduced by

Claude Shannon (1916–2001, Fig. 3.3). He wrote his master’s thesis in 1937:
“A Symbolic Analysis of Relay and Switching Circuits.” Part of this thesis was
published in 1938. Shannon showed that Boolean algebra could be used to simplify
the number and arrangement of electromechanical relays then used in telephone
exchanges. In the final chapter, he presented diagrams of several circuits, including
a 4-bit full adder, all designed using Boolean algebra. Over the years, this calculus
became the mathematical basis for all digital systems.
Shannon also worked in many other fields and was one of the most prolific
scientists of all time.
Let us look at Fig. 3.4: a conductive connection is achieved when the upper
contact or the lower contact or are both closed. This situation is described by
disjunction. To achieve a conductive connection in series connection, the first
contact and the second contact must be closed. This state of affairs is modeled
by the conjunction (Fig. 3.5).
Boolean algebra provides many algorithms to minimize logical expressions. A
particular function to be realized by a circuit can be represented by many different
formulas, and the minimal representation results in the memory requirements of
such a circuit being as small as possible and the computational speed reaching its
maximum. Since the 1960s, an enormous development of digital technology started.

Fig. 3.3 Claude Shannon (1916–2001)

18 3 The Historical Development

Fig. 3.4 A parallel circuit

realizes the disjunction .a ∨ b

Fig. 3.5 A series connection

realizes the conjunction .a ∧ b

The available circuits became smaller and faster, and larger and larger amounts of
data could be stored.
The fastest supercomputer in the world at the moment of writing is probably
Fugaku at the RIKEN Center for Computational Science in Kobe, Japan. It
performs 415.5 petaflops with its 7,299,072 cores. One petaflop corresponds to a
speed of .250 floating-point operations (operations with real numbers) per second.
3.2 The Dartmouth Conference 19

The other side of the coin: there has been a huge development in presenting
information digitally, as was the case with numbers. Nowadays, one can assume
that any physical phenomena can be represented digitally and thus be displayed
and processed by computers: sound waves, electromagnetic waves, X-ray images,
ultrasound images, images from a tomographer, and images from space are stored
in digital form immediately after they are obtained and can be used accordingly.

3.2 The Dartmouth Conference

The first time the term artificial intelligence has been used explicitly was in the
proposal
• “A Proposal for the Dartmouth Summer Research Project on Artificial Intelli-
gence”
by J. McCarthy, Dartmouth College, M. L. Minsky, Harvard University, N.
Rochester, I.B.M. Corporation, C.E. Shannon, Bell Telephone Laboratories,
dated August 31, 1955 (Fig. 3.6).
The following topics were discussed:
• Automatic computers,
• How a computer can be programmed to use a language,
• Neural networks,
• Theory of the size of a computation,
• Self-improvement,
• Abstractions,
• Randomness and creativity.

Fig. 3.6 The Dartmouth—Conference 1955

20 3 The Historical Development

However, there were already some publications before that, which used terms
like thinking computer, thinking machines, or similar. It can therefore be assumed
that this idea of intelligent machines had already been around for a while but that
the conference immediately put it on the agenda.

3.3 Algorithms and Programming Languages

Many valuable programming environments (software that supports the input of

a program written in a specific programming language and programming to the
executable correct program) can be obtained at low prices or even no cost. Currently,
the main programming languages are C, C++, Java, and Python. They are
constantly being developed and updated. Moreover, they are independent of the
operating system used. Permanently, new languages are created, and it is always
a difficult decision whether to move to a new programming language or not.
With programming, essential aspects of the software quality are to be considered
and converted by the organization of the source code. Examples are programming
style, user-friendliness, reusability, and modularity as well as maintainability.
Programming in the extended sense covers numerous further activities apart from
the pure programming, for example, the testing (developer test) of the program or
the production of the program documentation. Other development activities such as
project management, requirement analysis, or data modeling can be separated from
programming.
Depending on the type and application environment of the software (system
software, software for games, standard software, graphics software, etc.), different
procedures and tools (such as programming languages and test procedures) may be
used for development or performed by specialized developers.
According to the organization of the work, programming activities take place
in temporally separated project phases in parallel or iteratively. In practice, pro-
gramming is often done in teamwork using modern development methods and
programming tools. If you don’t produce software yourself, you always have to
make sure that you have a reliable partner. You also have to make provisions for
what you will do if the supplier fails. This also applies if you produce the software
in your own company. Even then, you should always have a replacement at hand.
This area is also full of problems. Testing a program is always done using rep-
resentative examples. However, this famous sentence applies here: An example
is not a proof! This means that a successful run of a program does not
guarantee that it will be the same for a next run with other data.
There is a long list of famous computer scientists who contributed to the
development of all kinds of programming languages.
With the onset of the Industrial Revolution (around 1880), many formerly manual
tasks were increasingly put under the control of machines. These were particularly
designed for repetitive tasks due to their cyclic modes of operation. On the other
hand, it became increasingly important to provide alternatives to the processes
3.3 Algorithms and Programming Languages 21

Fig. 3.7 A Jacquard loom controlled by punch cards

carried out by machines, especially since these processes could also be carried out
even faster.
Exemplary for this necessity is the introduction of the programmable loom by
Joseph Marie Jacquard, (1752–1834), at the beginning of the nineteenth
century. The loom worked with punched cards, which contained the information
about the pattern to be woven. The craft of weaving could now be continued as a
“programming art” on a medium analogous to the fabric pattern (Fig. 3.7).

3.3.1 The 1930s and 1940s: Logical Calculi

The lambda calculus was developed by Alonzo Church (1903–1995) and Stephen
Kleene (1909–1984) in the 1930s. It was proved early on that the lambda calculus
is, in principle, a universal programming language.
With the completion of the first electronic calculators and the use of Boolean
algebra, the development of programming languages continued noticeably. Notable
in this regard are the patents of Konrad Zuse (Fig. 3.8) in about 1937, describing a
computer architecture that would later become known as the von Neumann machine.
In the 1940s, Zuse completed his programming language plan calculus, in
German Plankalkül for it, incorporating ideas from the lambda calculus. The
von Neumann architecture (VNA) is a reference model for computers, according to
which a shared memory holds both computer program instructions and data. von
Neumann systems belong to the class of SISD (Single Instruction, Single Data)
architectures according to Flynn’s classification, as distinct from parallel processing.
The von Neumann architecture forms the basis for the way most computers
known today work. It is named after the Austro-Hungarian mathematician John von
Neumann, who later worked in the USA and whose major work on the subject was
22 3 The Historical Development

Fig. 3.8 Konrad Zuse

(1910–1995)

published in 1945. It is sometimes called the Princeton architecture (after Princeton

University).

3.3.2 The 1950s: First Modern Programming Languages

Grace Hopper (1906–1992) was an American computer scientist. She came up

with the idea in the late 1940s to write computer programs in an understandable
language, that is, not to work directly at the binary level. This level is understandable
only for computers. Any high-level programming language must be translated into
this machine language using a translation program, a compiler. She was
instrumental in the pioneering Mark I, Mark II, and UNIVAC I projects. She
developed the first compiler (A-0) in 1952, and with the programming language
FLOW-MATIC and the associated compiler (1957), she did essential preliminary
work for the development of the programming language COBOL (nickname,
“Grandma COBOL”). Their insistence on the importance of a general-purpose
language gave rise to compilers, interpreters, and programming languages.
In the 1950s, the first three widely used, practical higher-level programming lan-
guages were developed in the United States: the oldest language still in widespread
use, FORTRAN (FORmula TRANslator), was designed in 1954 by John W.
Backus et al. In 1959, LISP (LISt Processor) was added by John McCarthy
et al. The aforementioned languages exist with their successors to this day.
Especially LISP strongly influenced the programming languages developed later
at American universities.
The next major milestone was set between 1958 and 1960, when an international
committee designed a “new language for algorithms” during a series of meetings,
later named ALGOL 58. The committee completed its development with the
3.3 Algorithms and Programming Languages 23

Revised Report on ALGOL 60. The report to the meeting incorporated

many ideas that were circulating in the professional community at the time.
One important innovation was the Backus-Naur Form (BNF) for context-free
description of programming language syntax. Nearly all subsequent programming
languages use the BNF to represent syntax as a context-free grammar.
The von Neumann architecture forms the basis for the way most computers
known today work. It is named after the Austro-Hungarian mathematician John von
Neumann, who later worked in the USA and whose major work on the subject was
published in 1945. It is sometimes called the Princeton architecture (after Princeton
University).
Although ALGOL 60 did not catch on in North America for political reasons,
mainly because IBM offered a counter-development in the form of PL/1 but
partly also because of the decision not to include input and output in the language
definition, ALGOL subsequently became the standard in the (Western) European
world. It influenced the education of a whole generation of computer scientists and
the design of later languages, in particular SIMULA 67, PASCAL, and SCHEME.
For quite some time, these languages were used almost everywhere; in the course
of time, many countries, especially China, Russia, and Japan, developed their own
programming languages and devices.

3.3.3 The 1960s and 1970s: Development of New Paradigms

Ken Thompson (* 1943) and Dennis Ritchie (1941-2011) are the

developers of the programming language C, which is still widely used today [1].
In the following time, a large number of programming languages were developed,
since the possibility and the need were given by the fast progress of the computer
technology. The greatest success was achieved by further developments of the
already-existing programming languages. For example, BASIC (Beginner’s
All-purpose Symbolic Instruction Code) was developed around
1964 to help students get started programming with ALGOL and FORTRAN.
In the mid-1960s, the phenomenon of the cost of software exceeding the cost of
hardware first appeared in the so-called software crisis. It was also during this time
that the first major failed software projects occurred.
BASIC eventually became popular in the more affordable home computers
built in the late 1970s. The C programming language, designed in 1972 for
the newly developed UNIX operating system, also has its roots in ALGOL. It
prevailed over BASIC for general application programs; the basic functional parts of
many operating systems are programmed in C. Both programming languages have
spawned many variants to this day.
24 3 The Historical Development

3.3.4 Object-Oriented Programming

However, new concepts also emerged during this time. Great importance was
given to object-oriented programming, which unites data, procedure,
and reference aspects in the single concept of the object. Thinking and concepts
of object orientation first appeared in SIMULA 67, a language for simulation
purposes, which was the first language to introduce object-oriented procedures. The
story goes that its developers, Ole-Johan Dahl (1931–2002) and Kristen
Nygaard (1926–2002), had been working on ship simulations. In the process,
the unmanageable number of parameter relationships resulted in a bewildering
variety of ways in which the various attributes of the different ships could influence
each other. So they came up with the idea of treating the different ship types
as independent objects, with each class of objects being responsible for its own.
However, in SIMULA, the new concepts were not yet clearly distinguished from
their implementation. Their concepts and procedures had already been refined and
implemented more consistently than in SIMULA since the early 1970s at the Xerox
Palo Alto Research Center with the language SMALLTALK. Smalltalk was finally
released to the public at large in the 1980s. SMALLTALK was designed as a fully
dynamic system, where objects could be created and changed interactively—in
contrast to the system of static programs used before. Remarkable also compared
to its successors is the integration of the language in an innovative graphical user
interface, which for the first time allowed real interaction.
After Niklaus Wirth (Fig. 3.9) was disappointed while working on ALGOL,
he developed PASCAL together with Kathleen Jensen and consistently used
PASCAL for teaching language concepts from 1971 on. After realizing how
difficult it was to implement software development of larger projects with multiple
developers, he published MODULA-2, a further development with a strict module
and interface concept, in 1978.

Fig. 3.9 Niklaus Wirth (1934–2024)

3.3 Algorithms and Programming Languages 25

Alain Colmerauer (1941–2017), Phillipe Roussel, and Robert Kowalski (b. 1941)
founded logical programming starting in 1972, which is expressed in several
realizations of the PROLOG language.

3.3.5 The 1980s

In the 1970s, the US Department of Defense expressed concern about the number of
programming languages used in its projects, which numbered over 450. Many of the
programming languages were also not standardized but rather vendor dependent. A
working group was tasked with clearing this jungle and finding a language that met
the department’s military requirements. Many existing languages were reviewed,
but in 1977, the working group concluded that none of the existing languages were
suitable. After inviting bids for four draft languages, it was decided in 1980 to
use the language ADA composed from the best ideas of the drafts. This draft was
standardized under the designation MIL-STD 1815, because in 1815, the eponym
Ada Lovelace (1815–1852) was born. The US Department of Defense temporarily
mandated that any software project with more than 30% of new code be written
in ADA. To support the spread of the standard, the US Air Force funded the
development of the free GNAT compiler. The number of different programming
languages in the Department eventually reduced to 36.
Object-oriented programming began to become more popular in the mid-1980s,
largely due to the influence of C++, which was designed as a syntactic extension
of the C language. Bjarne Stroustrup (b. 1950) had introduced C++ in
1983. Many existing programming languages received object-oriented extensions
since that time, such as PASCAL or LISP.
Especially for the development of artificial intelligence, the languages PROLOG
and LISP were intended. PROLOG realizes a logical programming based on facts
and rules, while LISP realizes a functional programming. It is used for problems
where a prototype of the solution must be created quickly.

3.3.6 Further Development

• Linus Torvalds Linus Torvalds (* 1969) created LINUX in his room

on the campus of the University of Helsinki. Today, the open-source operating
system is used not only by computer geeks but mainly in data centers or
server farms. To this day, he coordinates the development of the Linux kernel.
In addition, he is considered one of the central developers of the version
management system Git.
• Brendan Eich
26 3 The Historical Development

Former Mozilla CTO Brendan Eich developed the JavaScript precursor

LiveScript. JavaScript is currently the de facto standard for Internet page
development.
• Tim Berners-Lee
The British computer scientist Tim Berners-Lee is considered the inventor
of HTML and thus the founder of today’s Internet (World Wide Web). In
2004, Queen Elizabeth II awarded him a knighthood for his achievements. Sir
Tim Berners-Lee currently heads the World Wide Web Consortium (W3C) and
teaches as a professor at the Massachusetts Institute of Technology (MIT).
The world’s first Web server was developed and implemented by Berners-
Lee on a NeXTcube computer. One problem at CERN was that one part of
the laboratories was located on French territory and another on Swiss territory.
The two countries had different network infrastructures, which made it difficult
to exchange information. On March 12, 1989, Berners-Lee proposed to his
employer, CERN, a project based on the principle of hypertext that would facil-
itate the worldwide exchange and updating of information between scientists. In
1990, he published Robert Cailliau, a concept for a worldwide hypertext
project, the computer program Enquire. Subsequently, Berners-Lee developed
the page description language HTML, the transfer protocol HTTP, the URL, the
first browser, and the first Web server CERN httpd.
Berners-Lee created the first Web presence, info.cern.ch. This Web site still
exists today; however, only a copy from the 1992 original content still exists. It
explained, among other things,
• what the World Wide Web should be,
• how to get a Web browser,
• how to set up a Web server.
Originally, this was also the first simple search engine, because Berners-Lee was in
charge of other Web sites besides his own. The basic ideas of the World Wide Web
are comparatively simple, but Berners-Lee saw and linked them in a way whose
possibilities are still not fully exploited today.
In 1994, Berners-Lee founded the World Wide Web Consortium (W3C)
at the Massachusetts Institute of Technology. It was important that he did not patent
his ideas and technical implementations but passed them on freely. He also had a
strong influence on the maxim of the “World Wide Web Consortium” to adopt only
patent-free standards. In his book Weaving the Web, for example, the following is
emphasized: being able to edit the Web is as important as being able to browse the
Web. Computers can be used to perform tasks in the background so that groups can
collaborate better. Each area of the Web should have a network structure rather than
a tree structure. Notable exceptions are the Domain Name System and ICANN’s
rules for assigning domain names. Computer scientists have not only a technical but
also a moral responsibility.
3.4 The Turing Machine 27

3.4 The Turing Machine

Many questions related to algorithms require a precise mathematical definition of

an algorithm. Alan Mathison Turing (1912–1954; Fig. 3.11) was a British
logician, mathematician, cryptanalyst, and computer scientist. He is now considered
one of the most influential theorists of early computer development and computer
science. Turing created much of the theoretical foundation for modern information
and computer technology. His contributions to theoretical biology also proved
groundbreaking.
The computability model of the Turing machine he developed forms one of the
foundations of theoretical computer science. During World War II, he
was instrumental in deciphering German radio messages encrypted with the German
rotor cipher machine Enigma. Most of his work remained classified even after
the end of the war. The insights Turing gained in the cryptanalysis of Fish ciphers
later helped in the development of the first digital, programmable electronic tube
computer, ENIAC.
In 1953, Turing developed one of the first chess programs, whose calculations he
performed himself for lack of hardware. The Turing Award, the most important
award in computer science, is named after him, as is the Turing Test for proving
the existence of artificial intelligence.
In March 1952, Turing was sentenced to chemical castration for his homosexu-
ality, which was still prosecuted as a crime at that time. As a result of the hormone
treatment, Turing became ill with depression and died by suicide about two years
later. In 2009, then British Prime Minister Gordon Brown issued an official apology
on behalf of the government for Turing’s “abhorrent treatment” and acknowledged
his “extraordinary services” during the war; however, a pardon was still denied in
2011 despite a petition (Fig. 3.10).
Turing’s involvement as one of the key code breakers in deciphering the Enigma
was secret until the 1970s; not even his closest friends knew about it. The decryption

Fig. 3.10 ENIGMA—the

decoding equipment
28 3 The Historical Development

of secret German radio messages was a war-decisive component in the Allied

victory in the U-boat war and the African campaign.

3.5 The First Computers: The Turing Test, the Single-Tape

Turing Machine

From 1945 to 1948, Turing was employed at the National Physical Laboratory
in Teddington, where he worked on the development of the ACE (Automatic
Computing Engine). The name of the machine is derived from the analytical engine
of the mathematician Charles Babbage, whose work Turing admired throughout his
life (Fig. 3.11).
In 1948, Turing taught at the University of Manchester and became deputy head
of the computer department in 1949. Here he worked on the software for one of
the first real computers, the Manchester Mark I, while continuing to write various
theoretical papers. In Computing Machinery and Intelligence (Mind, October 1950),
Turing took up the problem of artificial intelligence and proposed
the Turing test as a criterion for whether a machine is capable of thinking like
a human. Since the thinking process cannot be formalized, the test looks only at
the responses of a machine in dialogue with a human, i.e., the communicative
behavior of the machine. If this appears indistinguishable from human behavior,
one should speak of machine intelligence. With his publication, he significantly
influenced the development of artificial intelligence. In 1952, he wrote the chess
program Turochamp. Since there were no computers with sufficient power to run
it, Turing took over its function and calculated each move himself. This took up
to 30 minutes per move. He lost the only game documented in writing against a
colleague.
From 1952 until his death in 1954, Turing worked on mathematical problems
in theoretical biology. In 1952, he published a paper on the chemical basis of
morphogenesis. In this work, he described for the first time a mechanism by
which reaction-diffusion systems can spontaneously evolve structures. This process,

Fig. 3.11 Alan Turing

(1912–1954)
3.5 The First Computers: The Turing Test, the Single-Tape Turing Machine 29

Fig. 3.12 A single-tape

Turing machine $ 0 1 2 3 4 5 6 7 8 9

work tape

control

known as the Turing mechanism, is still central to many chemical-biological theo-

ries of structure formation today. Another of Turing’s interests was the occurrence
of Fibonacci numbers in the structure of plants. Later work remained unpublished
until the publication of his collected works in 1992.
One way to define the notion of an algorithm in a mathematically precise way is
the single-band Turing machine (Fig. 3.12). It has a text unit containing the program
and also consists of an infinitely long memory tape with an infinite number of
sequentially arranged fields. Exactly one character from a predefined alphabet can
be stored in each field. As an additional character, a blank is allowed, which
corresponds to an empty field on the memory tape. A program-controlled read
and write head can move from field to field on the memory tape and change the
characters. A computation for an input word begins with the input word on the
tape and the read and write head on the first symbol of the input word. The Turing
machine then processes the input on the tape step by step according to the given
program.
At each step, the read-write head reads the current character, overwrites it with
another or the same character, and then moves left or right or stops. Which character
is written and which movement is executed depends on the character found at the
current position and on the state the Turing machine is currently in. At the beginning,
the Turing machine is in a certain start state and moves to a new state at each step.
The number of states the Turing machine can be in is finite. A state can be passed
through several times, and it says nothing about the characters present on the tape.
A Turing machine stops if no transition to a new state is defined for the current
state and the tape character read. So, in general, it depends on the combination
of state and symbol whether the Turing machine continues computing or stops.
States in which the Turing machine stops independently of the tape symbol read
are called final states. However, there are also Turing machines that never stop on
certain inputs.
A Turing machine is a tuple

M = (Z, E, A, d, q, F )
.

with
• Z a finite nonempty set of states of the control system,
• E the input alphabet,
30 3 The Historical Development

• A the working alphabet with .E ⊆ A ,

• d the transition relation, .d ⊆ Z × A × A' × Z; where .A' = A ∪ {L, R},
• .q ∈ Z is the starting state,

• .F ⊆ Z the set of final states.

The working alphabet A contains the special characters $ and the space; they do
not occur in the input alphabet. The characters L and R do not occur as alphabet
characters.
• .E ⊆ A means that every element of the set E is an element of the set A; possibly,
as a limiting case, .E = A.
• The sign .∪ means the union of two sets; one obtains the set .A' if one adds to the
set A the movements L (to the left) and R (to the right).
In addition to the computation of functions, the Turing machine is also used for
decision problems, i.e., for questions that have to be answered with “yes” or “no.”
Certain final states are defined as “accepting” while others as “not accepting.” The
input is accepted exactly when the Turing machine terminates in an accepting final
state.
With his model, Turing defined the terms “algorithm” and “computability”
as mathematical concepts. It is generally assumed that Turing’s computability
corresponds to the intuitive understanding of computability; this statement is known
as the Church-Turing thesis . The distinctive feature of a Turing machine
is its structural simplicity. It requires only three operations (reading, writing, and
moving the read-write head) to simulate all the operations of common computer
programs.

3.6 LaTeX

Scientific papers are very often difficult to write. Here we find two languages that are
extremely significant. LaTeX describes documents quite precisely and in all details.
The basic program of LaTeX is TeX and was developed by Donald E. Knuth during
his time as a computer science professor at Stanford University. Building on TeX,
Leslie Lamport developed LaTeX in the early 1980s, a collection of TeX macros
that simplified and enhanced its use for the average user compared to TeX. The
name LaTeX is an abbreviation for La—import—TeX. The details of the document
are described and converted by Tex into the document, into a form in which it is then
to be presented. To support the creation of the description, an editor that supports
the input of the LaTeX text is still necessary. For this book, MiKTex was used
as a program for translation, and TeXnicCenter is the used editor. However, there
are other equivalent systems available. They are all provided free of charge on the
Internet. In contrast to other word processing programs, which work according to
the what-you-see-is-what-you-get principle, LaTeX works with text files in which
3.6 LaTeX 31

passages or headings to be formatted differently within a text are described with

special commands.
Before the LaTeX system can set the text accordingly, it must process the
source code. The layout generated by LaTeX in this process is very clean, and the
formula set is of high quality. In addition, output to PDF, HTML, and PostScript
is possible. LaTeX is particularly suitable for extensive works such as theses and
dissertations, which often have to meet strict typographical requirements. Especially
in mathematics and the natural sciences, LaTeX facilitates the preparation of
documents through its convenient options for formulation compared to conventional
word processing systems.
The step-by-step work requires superficially in comparison with conventional
text processing systems on the one hand a longer training period; on the other hand,
the appearance of the result can be exactly specified. However, the longer training
period can be worthwhile, especially for subsequent projects with a comparable
scope or similar requirements. In the meantime, there are also graphical editors
that can work with LaTeX and offer WYSIWYG or WYSIWYM (what you see is
what you mean.) and can make it much easier for inexperienced users to get started
(Fig. 3.13).
LaTeX is an excellent example of collective intelligence. The basic functions
are extended by packages (packages), each of which serves a specific purpose.
Currently, 4000 packages are available. Whenever someone misses a particular
function, they can embed that function in a package and make that package available
to the general public.

Fig. 3.13 A short part of a LaTeX text

32 3 The Historical Development

Chemical formulas are also described simply and elegantly.

An extremely elegant and helpful extension of latex is called TikZ. It is probably
the most complex and powerful tool to create graphic elements in LATEX. We are
adding one more example. A useful collection of examples you will find at [5].

O
H
H
. The formula for water

The introduction of the euro sign was quickly taken into account, since it was
not present on the keyboards at the beginning: you simply write \euro{ } and get e.
There is plenty of teaching material on the Internet, which you have to learn step by
step.

3.7 The Evolution of Computers

Moore’s law shows that computing power is constantly increasing. Where have
we gotten to now? According to Wikipedia [2], the Japanese computer Fugaku
RIKEN from the Center for Computational Science, Kobe, Japan, is in first place,
with 415,530.00 teraflops. It is followed by two American and two Chinese systems.
The German system, Hawk, is located in Stuttgart, Germany, and has a performance
of 26,000 teraflops.
The SC21 supercomputing show comes with confirmation that China is running
the world’s first two exascale supercomputers: OceanLight and Tianhe-3. According
to this, OceanLight achieves a peak computing performance of 1.3 ExaFlops with
double precision (FP64) in the Linpack benchmark—over a longer period of time,
1.05 Exa-Flops are possible. If the National Research Center of Parallel Computer
Engineering and Technology (NRCPC) were to include the supercomputer in its
Top500 list, it would be around twice as fast as the previous leader, Japan’s Fugaku.
Development in the United States is proceeding in much the same way. A kind of
race and prestige thinking has emerged (see Figs. 3.14 and 3.15) [2, 3].
The costs are quite high: the cost of the Summit supercomputer was about 44
million euros, while the annual electricity costs are up to 4 million euros per year.
The next step might be quantum computers. Unlike the classical computer, it
operates on the basis of quantum mechanical states rather than electrical ones. Here,
firstly the superposition principle and secondly the quantum entanglement are of
importance.
Theoretical studies show that by exploiting these effects, certain problems in
computer science, such as searching extremely large databases and factoring large
numbers, can be solved more efficiently than with classical computers. This would
make it possible to significantly reduce the computation time for many mathematical
and physical problems.
3.7 The Evolution of Computers 33

Fig. 3.14 Moore’s Law for Supercomputers

Fig. 3.15 A supercomputer consists of many processors working in parallel

For a long time, the quantum computer has been a theoretical concept. There
have been various proposals on how a quantum computer could be realized; on a
small scale, some of these concepts have been tested in the laboratory, and quantum
computers with a few qubits have been realized. The record was 127 qubits for the
processor in November 2021. However, in addition to the number of qubits, other
34 3 The Historical Development

important factors exist, for example, a low error rate during computation and readout
and how long the states in the qubits can be maintained without errors.
qubits or quantum bits are the computational units of a quantum com-
puter. Qubits can store more information than just 1 or 0 compared to conventional
bits in computers. Qubits also exist in a superposition of these values—in what is
known in technical jargon as superposition. Unlike conventional bits, quantum bits
can assume any number of states simultaneously. Quantum computers thus achieve
great computing power with just a few qubits. Theoretically, the computing power
of a quantum computer increases exponentially with the number of qubits. Even
with just a few qubits, complex problems can be solved in a very short time.
In addition to the technology giant IBM, the industry itself has long recognized
the great potential of quantum computers and has joined forces in a consortium. The
aim of the consortium, called Quantum Technology and Application
Consortium (QUTAC), is to test industrial applications with high potential for
quantum computers.
Overall, quantum computers are expected to be used in the following areas:
• Logistics
Quantum computers could help optimally allocate limited resources in logistics,
traffic flow and transportation routes, network structures of telecommunication
companies, or the distribution of medical resources for hospitals.
• Artificial intelligence and machine learning
According to the Fraunhofer Alliance “Big Data and Artificial Intelligence,”
quantum computers could revolutionize artificial intelligence and machine learn-
ing: “Procedures of artificial intelligence and machine learning can be adapted
for quantum computers in such a way that they can follow several solution paths
at the same time. This allows quantum computers to process large data sets in
a single step, detect patterns in the data that classical computers cannot, and
provide reliable results even on incomplete or uncertain data.”
• Robotics
BMW’s example is just a taste of the potential of quantum computing for
robotics. Quantum computing could determine the most efficient paths for
employees or robots (AGVs, Autonomous Guides Vehicles) to navigate through
a warehouse.
• Chemistry, pharmacy, and materials science
The Fraunhofer “Cluster of Excellence Cognitive Internet Technologies” is
certain: “With the help of the simulation of molecules, catalysts could be
developed in a targeted manner in the future, for example, to make chemical
production processes more efficient. Opportunities of a similar magnitude arise
for the pharmaceutical industry.” And battery research also falls within the
scope of quantum computer-based simulations—the development of novel, more
powerful batteries for electric cars.
• Engineering
References 35

Are materials with certain properties desired and is a trade-off between stability
and weight important?—a case for quantum computing, because the properties of
any material depend fundamentally on its quantum mechanical building blocks.
• Finance
Faster financial simulations and better portfolio optimization solutions are
potential applications that make quantum computing attractive to the financial
industry.
• The German government has released a total of two billion euros for the devel-
opment of quantum computers for the international race for quantum computing.
The goal is to build a competitive quantum computer in Germany within the
next five years. So far, there is no quantum computer in Germany that has
been built entirely without technology from abroad. Currently, the German mid-
sized companies Trumpf and Sick are considered leading competence carriers in
quantum optical sensors [4].
So the path to ever more powerful computers is also far from having reached
its end. However, the design of circuits based on this principle is much more
complicated.

References

1. Bernd Steinbach, Christian Posthoff. Logic Functions and Equations—Fundamentals and

Applications using the XBOOLE-Monitor, Third Edition, Springer Nature, 2021.
2. https://www.heise.de/news/China-betreibt-schon-zwei-ExaFlops-Supercomputer-
3. https://www.computerbase.de/2021-10/intel-supercomputer-aurora-soll-2-exaflops-liefern-
zettaflops-schon-2027-kommen
4. https://www.mittelstand-heute.com/artikel/zukunft-mittelstand-wo-werden-quantencomputer-
eingesetzt
5. https://texample.net/tikz/examples/feature/coordinate-calculations/
Chapter 4
Mathematics on the Computer

The solution of mathematical problems with the help of a computer is a very original
concern of the development of calculating machines (Fig. 4.1) and computers. To the
extent that mathematics is applied in many fields, computers and their programs can
be applied as well.
At the beginning, it was essentially a matter of calculating with numbers, but
very quickly the programs were extended to symbolic arithmetic, solving equations,
differential and integral calculus, and all other areas of mathematics. There are now
a large number of such systems, usually grouped together under the name “systems
for computer algebra.”
It is difficult to choose. It is best to start with a smaller system and work your
way through it. This already costs a lot of effort. Later one can decide whether the
system is already sufficient or whether one needs a larger, more elaborated system.
Small systems have above all the advantage that one can use them without costs. The
existence of such systems will have a significant impact on mathematics education
in the coming years.
Before presenting individual software packages with which one can work,
it should be mentioned that one can obtain valuable information by searching
the Internet. Formulas and explanations are provided there for all subfields of
mathematics. A prudent high school graduate can complete his homework in a short
time.

4.1 Microsoft Mathematics

A good place to start is the system Microsoft Mathematics (Fig. 4.2). It is

freely available. The desktop is modeled after a calculator, but the system is much
more powerful. It covers all the material from eighth grade through high school
graduation to the first year of a mathematics or physics degree. A certain dilemma

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 37

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_4
38 4 Mathematics on the Computer

Fig. 4.1 A mechanical

calculator from the 1960s

Fig. 4.2 The Microsoft Mathematics work screen

arises: on the one hand, it is sensible not to spend a lot of time on calculations that
a PC has at its fingertips. But on the other hand, it could be that the independent
calculations also add to the amount of knowledge.
However, one can also consider that in this way, it is possible to no longer regard
mathematics as an “‘enemy” that one must avoid as soon as possible. One can
make the lessons much more interesting by including the relations of mathematics
to architecture, painting, and others. There are also two textbooks [1, 2] for this
system, which can be found on the Internet at bookboon.com.
4.3 Maple 39

Fig. 4.3 Mathematica

4.2 Mathematica

Mathematica is a software system of the Wolfram Research company and represents

one of the most widely used mathematical and scientific program packages. Author
and company founder Stephen Wolfram began development work in 1986, and the
first version of Mathematica was released in 1988 (Fig. 4.3).
The software package “Mathematica” contains among others:
• a computer algebra system for symbolic processing of equations and differential
equations,
• a numerical software for numerical solution or evaluation of equations,
• a visualization tool for displaying graphs and 3D/2D graphics,
• a programming language that combines elements of procedural, object-oriented,
functional, and rule-based programming.
A fantastic system that should be a companion to every mathematics course. One
saves a great deal of time that would have to be spent on calculations and has much
more time to devote to understanding the concepts and their solutions.

4.3 Maple

The first version of Maple was programmed in 1980 by Keith O. Geddes, Gaston
H. Gonnet, and their collaborators from the Symbolic Computation Group at the
University of Waterloo in the Canadian city of Waterloo (Ontario). At the end of
1987, Maple already existed in the version 4.2. (Fig. 4.4).
In 2005, Maple 10 introduced a new document mode within the standard version
of Maple. Since then, it is possible to edit Maple input in normal mathematical
notation. Here, texts and mathematical symbols can be combined in the same input
line.
40 4 Mathematics on the Computer

Fig. 4.4 Maple

The main component of the graphical user interface of Maple is the respective
worksheet, in which interactive work is done. It appears as a window in which
calculation instructions are entered. The Maple engine interprets these instructions
and returns corresponding outputs. Typical Maple outputs are numerical values,
terms, functions, tables, two- and three-dimensional graphics, animation objects,
and diagrams.
Maple includes a core of commonly used standard computational instructions
and additional packages loadable at runtime with the with command. It is exhilarat-
ing to have almost the “entire currently known mathematics” on your computer. A
technician, engineer, geologist, biologist, physician, etc. is of course faced with the
problem of selecting the subfields that are important to him. Therefore, in practice,
one already encounters some sub-domains such as business mathematics, financial
mathematics, statistics, etc., but further such specializations and narrowing down
will be necessary. Education can focus on teaching the basic ideas; the calculations
themselves and their presentation can be left to computer technology.

References

1. Christian Posthoff, Bernd Steinbach. Mathematik—Ohne Sorgen an der Uni I—Nutze Microsoft
Mathematics, bookboon.com, 2017, 978-87-403-1596-7
2. Christian Posthoff, Bernd Steinbach. Mathematik—Ohne Sorgen an der Uni II—Nutze Microsoft
Mathematics, bookboon.com, 2017, 978-87-403-1949-1
Chapter 5
Polynomial and Exponential Complexity

5.1 Introductory Considerations

The interrelationships presented in the next sections and chapters make it clear that
artificial intelligence is always up for discussion when one is faced with complex
systems whose investigation far exceeds the capabilities of humans. Worldwide
globalization, the fact that everyone is networked with everyone else, offers great
opportunities on the one hand but also harbors dangers. The possibilities are often
pushed into the distance, while disaster scenarios are immediately painted on the
wall when it comes to the risks.
The high level of interconnectedness means that a change at one point can
propagate throughout the system almost unhindered. Very often, a butterfly
effect occurs. It states that in complex, nonlinear dynamic systems, even the
smallest changes in the initial conditions, can lead to the fact that predictability
with respect to the further development of a system can basically be ruled out. The
butterfly effect has become known especially in connection with the topic weather
forecast. This is to express figuratively that already the wing beat of a butterfly at a
point X on earth can influence the weather event at another place Y. As astonishing
as this may seem at first glance, this effect is understandable. The simple scheme
cause .→ effect is not applicable to complex systems; even small causes can have a
large effect (Fig. 5.1).
The catchy term butterfly effect originated with US meteorologist Edward N.
Lorenz, who in 1972 gave a talk to the American Association for the Advancement
of Science titled “Predictability: Does the Flap of a Butterfly’s Wings in Brazil Set
Off a Tornado in Texas?” In its original form, however, it used the flap of a seagull’s
wings instead of the butterfly.
Preliminary work on the theory was done by Edward N. Lorenz (1917–
2008) with a 1963 paper in which he undertook a computer weather prediction
calculation. In connection with long-term weather forecasts, he used a simplified

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 41

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_5
42 5 Polynomial and Exponential Complexity

Fig. 5.1 The butterfly effect

convection model to investigate the behavior of liquids or gases when they are
heated: here, rolls initially form (hot gas rises on one side, loses heat, and sinks
again on the other side), which become unstable when further heat is added.
He characterized this behavior using three connected differential equations. He
projected the numerical result into phase space and obtained that strange attractor
that later became known as the Lorenz attractor: an infinitely long trajectory in three-
dimensional space that does not intersect itself and has the shape of two butterfly
wings.
Lorenz came across the chaotic behavior of his model rather by accident. In
order to save computing time, he had used intermediate results of already performed
calculations for the numerical solution of the equations but only took three decimal
places into account, although the computer calculated with an accuracy of six
decimal places. The differences between the old and the new calculations became
greater and greater, although they could only be based on deviations in the fourth
digit.
This led Lorenz to his statements about the sensitivity to the initial conditions.
From nearly the same starting point, the weather curves diverged until they finally
showed no commonality. In his first calculation, he gave an initial value for one
iteration to six decimal places (0.506127) and in the second calculation to three
(0.506), and although these values differed only by about .1/10,000, as time went
on, this calculation diverged sharply from the first.
If you consider that the circumference of the earth at the equator is 40,000 km,
then one ten-thousandth of that is equal to 4 km. A change in the landscape along
this distance is enough to trigger a chaotic event somewhere (Fig. 5.2).
The varying trajectories .x(t), y(t), z(t) are solution of the following differential
equation system:

dx
. = σ · [y(t) − x(t)]. (5.1)
dt
5.1 Introductory Considerations 43

Fig. 5.2 The Lorenz attractor

dy
= [ρ · x(t)] − [x(t) · z(t)] − y(t). (5.2)
dt
dz
= [x(t) · y(t)] − β · z(t). (5.3)
dt
Here .σ, ρ and .β are parameters to be defined beforehand. The nonlinearity can be
seen in the terms .[x(t) · z(t)] and .[x(t) · y(t)].
Because of the finite mantissa in floating point representation, numbers cannot
be represented with arbitrary precision on a computer. They must be rounded.
Instead of .x, the computer uses the number .rd(x) for another calculation. Today’s
computers usually work according to IEEE 754. The machine precision for the
data types used in this process is . = 2−24 ≈ 6 · 10−8 for single precision and
. = 2
−53 ≈ 1.1 · 10−16 .

Extending these considerations further, we arrive at chaos theory. Chaos

theory describes the temporal behavior of systems with deterministic chaotic
dynamics. If one tries to repeat experiments identically, this is not possible in prac-
tice, because due to unavoidable measurement inaccuracies—and due to noise—the
initial situation cannot be restored identically. If a system is deterministically
chaotic, then after a sufficiently long time, the system can lead to clearly different
final states or measurement results despite experimentally almost identical (or as
identical as possible) initial situations.
Contrary to the colloquial use of the term chaos, chaos theory does not deal
with systems that are subject to chance (i.e., stochastic systems) but with dynamic
systems that can be described mathematically and behave deterministically in
principle. Furthermore, chaos theory is to be distinguished from the theory of
complex systems, since even very simple systems can exhibit chaotic behavior.
44 5 Polynomial and Exponential Complexity

5.2 Mathematical Descriptions of Complexity

We begin with a simple example. We imagine a wall board with a row of books.
If one is now looking for a particular book, one starts from the left and checks one
book after another to see if it is the wanted book. If you are lucky, it is the first one,
but if you are unlucky, it is the last book in the row, or it is not there at all. So the
number of necessary steps is between 1 and n, if n is the number of books on the wall
board. To compare different algorithms in a meaningful way, one always uses the
worst case (worst-case complexity) and says that the linear search
has complexity .O(n).
In the next step, we imagine a shelf that contains n boards for books. So there are
.n · n books there, and in the worst case, it takes .n search steps. The complexity of
2
2
the search is of order .O(n ). If we now go to a library, there are many such shelves
in a row. We again assume that there are n pieces, and the complexity of the search is
now equal to .O(n3 ). However, one can also imagine a very long wall shelf where all
the books are in a row, and one uses the same linear search algorithm. Therefore, we
group all these problems into the set for which the running time .t (n) of a solution
algorithm is bounded by a polynomial of n-th degree:

t (n) ≤ nk .
. (5.4)

Polynomial time is considered as a boundary between practically solvable and

practically unsolvable problems. So it is written in all textbooks, but due to the
development of computational speed, this theorem can be safely deleted.
Whether a given problem is solvable in polynomial time is not always clear from
the outset. For example, for the problem of deciding whether a given natural number
is prime, an algorithm running in polynomial time was not given until 2002 by
Agrawal, Kayal, and Saxena (AKS prime test). The naive procedure of trying
all possible divisors is not feasible in polynomial time. In the rest of the text, a
more constructive point of view is taken. The only thing to consider is whether the
problem is solvable in reasonable time on a given computer (although “reasonable”
is not very clearly defined either).
A simple example will show that exponential complexity can be encountered
even in simple facts. Let a set M with three elements be given:

M = {a, b, c}.
.

For some reason, we are interested in determining the power set of M, that is, the
set of all subsets. This is not difficult.
5.2 Mathematical Descriptions of Complexity 45

⎛ ⎞
a b c Menge
⎜0 0 0 ∅ ⎟
⎜ ⎟
⎜1 0 {a} ⎟
⎜ 0 ⎟
⎜ ⎟
⎜0 1 0 {b} ⎟
⎜ ⎟
. ⎜0 0 1 {c} ⎟
⎜ ⎟
⎜1 1 0 {a, b} ⎟
⎜ ⎟
⎜1 0 1 {a, c} ⎟
⎜ ⎟
⎝0 1 1 {b, c} ⎠
1 1 1 {a, b, c}

You form a matrix where each row has four positions. The first three columns are
assigned to the three elements, a, b and c; the last column shows the set represented
by the row. The value 0 indicates that the element does not belong to the subset; the
number 1 indicates that it belongs to the subset. Since both possibilities are possible
for each element, there are .2 · 2 · 2 = 23 = 8 possibilities. In real life, you can see
this in a family tree: if you disregard all misfortunes and other abnormalities, you
have
• one child,
• two parents,
• four grandparents,
• eight great-grandparents.. . .
• sixteen great-great grandparents, etc.
So here one sees the same powers of the number 2 as before. More generally, the
power set of a set with n elements contains .2n elements.
We consider the function

f (a, b, c, d) = ab ⊕ (c ∨ d).
.

Here four variables occur with possible values of 0 or 1. A table of the function
values now has .24 = 16 rows. So the size of the tables for the values of binary
(logical) functions also grows exponentially.
46 5 Polynomial and Exponential Complexity

⎛ ⎞
a b c d f (a, b, c, d
⎜0 ⎟
⎜ 0 0 0 0 ⎟
⎜ ⎟
⎜0 0 0 1 1 ⎟
⎜ ⎟
⎜0 0 1 0 1 ⎟
⎜ ⎟
⎜0 0 1 1 1 ⎟
⎜ ⎟
⎜0 1 0 0 0 ⎟
⎜ ⎟
⎜0 ⎟
⎜ 1 0 1 1 ⎟
⎜0 ⎟
⎜ 1 1 0 1 ⎟
⎜ ⎟
. ⎜0 1 1 1 1 ⎟
⎜ ⎟
⎜1 0 0 0 1 ⎟
⎜ ⎟
⎜1 0 0 1 0 ⎟
⎜ ⎟
⎜1 0 1 0 0 ⎟
⎜ ⎟
⎜1 ⎟
⎜ 0 1 1 0 ⎟
⎜ ⎟
⎜1 1 0 0 0 ⎟
⎜ ⎟
⎜1 1 0 1 1 ⎟
⎜ ⎟
⎝1 1 1 0 1 ⎠
1 1 1 1 1

Since the field of circuit design makes exclusive use of binary functions, one
constantly encounters exponential complexity there (Fig. 5.3).
Even in everyday life, one is confronted with exponential complexity, as can
easily be seen in the family tree of a human being (Fig. 5.4).
It is an important characteristic of problems that are handled by means and
methods of artificial intelligence that they have a very high complexity. It is not

Fig. 5.3 Miniaturization of circuits

5.3 Rules and Constraints 47

Fig. 5.4 Exponential complexity of a family tree

always known which function describes the complexity; in any case, it is far above
the complexity of problems that can be solved by humans.

5.3 Rules and Constraints

Our life is full of rules [3].

• The easiest way to start is to look at the natural sciences and mathematics. Here,
the rules work automatically:
When it rains, the earth gets wet.
• Other rules arise because of social conventions.
People wash their hands before eating. Noncompliance leads to social exclusion
and criticism.
• Another group of rules is based on law.
On this place, parking is not allowed.
Compliance is enforced by imposing small, medium, or severe penalties for
noncompliance.
• Every game is based on rules.
The oldest collection of laws known in wording is the “Codex Hammurabi” by
King Hammurabi, who lived in the eighteenth century BC. The aim of these laws
was justice for all. In general, many people thought about what justice actually is.
The Greek philosophers Plato and Aristotle came to the answer about 1,400 years
later: justice is treating equal cases equally. The ancient Greeks contributed above
all to the further development of jurisprudence [4].
Contract law is particularly important for politics and business. We will return to
the connection between AI and law in a special chapter.
48 5 Polynomial and Exponential Complexity

The result is a huge web of relationships in which only individual parts can be
seen through separately. Not for nothing do court cases go on for years. Inheritance
disputes can destroy entire families and much more. And there is a long and
interesting history of how such systems of rules are created, introduced, or enforced.
Unfortunately, they do not always follow humane and correct principles either. In
any case, one can immediately see that this is the next extremely complex issue.
In mathematics and the natural sciences, implication is defined as a function to
formalize the application of rules. Linguistically, the implication is expressed by a
sentence with the structure “if assumption then conclusion.”

Implication
x y .x ⇒ y .x ∨ y

0 0 1 1
0 1 1 1
1 0 0 0
1 1 1 1

If one accepts a rule as valid and the assumption is true, then one receives with
certainty a true conclusion, about which one does not have to think further! If the
premise is not fulfilled .(x = 0), then the rule must not be applied.
Now, a scientific way of working is to have a set of rules to apply and a set of true
propositions. The application of the rules results in new true propositions, which one
adds to the set of true propositions one already has. You continue this process until
you have reached a certain goal.
And of course, it is questionable whether a theorem can really be proved with the
help of the premises and whether the proof is also correct. One famous example is
the proof of Fermat’s last theorem by Andrew Wiles.
Theorem No three positive integers a, b, and c can satisfy the equation

a n + bn = cn
. (5.5)

if n is an integer greater than two .(n > 2).

An overview of the proof can be found, for instance, in [5].
Implication can be expressed with the help of disjunction and negation:

(x → y) = (x ∨ y).
. (5.6)

If .x = 0, i.e., the assumption is false, but the implication is true and does
not depend on x. From a premise and its converse, two parties reach the same
conclusion. What is the effect of such a situation? The use of this rule is in the
given context without any meaning, the rule can be skipped.
When correct assumptions are used, the number of true conclusions should be
growing.
5.4 Axioms 49

5.4 Axioms

In mathematics in particular, it is customary to place the smallest possible set of

statements at the top without proof and to derive everything else from them. These
statements are called axioms. An axiom, postulate, or assumption is a statement that
is taken to be true, to serve as a premise or starting point for further reasoning and
arguments.
The precise definition varies across fields of study. In classic philosophy, an
axiom is a statement that is so evident or well established that it is accepted without
controversy or question. In modern logic, an axiom is a premise or starting point for
reasoning.
In mathematics, an axiom may be a “logical axiom” or a “non-logical axiom.”
Logical axioms are taken to be true within the system of logic they define and are
often shown in symbolic form (e.g., (.A ∧ B) ⇒ A), while non-logical axioms (e.g.,
a + b = b + a) are substantive assertions about the elements of the domain of a
specific mathematical theory, such as arithmetic.
Non-logical axioms may also be called “postulates” or “assumptions.” In most
cases, a non-logical axiom is simply a formal logical expression used in deduction
to build a mathematical theory and might or might not be self-evident in nature (e.g.,
the parallel postulate in Euclidean geometry). To axiomatize a system of knowledge
is to show that its claims can be derived from a small, well-understood set of
sentences (the axioms), and there are typically many ways to axiomatize a given
mathematical domain.
Any axiom is a statement that serves as a starting point from which other
statements are logically derived. Whether it is meaningful (and, if so, what it means)
for an axiom to be true is a subject of debate in the philosophy of mathematics.
The logico-deductive method whereby conclusions (new knowledge) follow
from premises (old knowledge) through the application of sound arguments (syl-
logisms, rules of inference) was developed by the ancient Greeks and has become
the core principle of modern mathematics. Tautologies excluded, nothing can
be deduced if nothing is assumed. Axioms and postulates are thus the basic
assumptions underlying a given body of deductive knowledge. They are accepted
without demonstration. All other assertions (theorems, in the case of mathematics)
must be proven with the aid of these basic assumptions. However, the interpretation
of mathematical knowledge has changed from ancient times to the modern, and
consequently the terms axiom and postulate hold a slightly different meaning for
the present day mathematician, than they did for Aristotle and Euclid.
The ancient Greeks considered geometry as just one of several sciences and held
the theorems of geometry on par with scientific facts. As such, they developed and
used the logico-deductive method as a means of avoiding error and for structuring
and communicating knowledge. Aristotle’s posterior analytics is a definitive expo-
sition of the classical view.
At the foundation of the various sciences lay certain additional hypotheses that
were accepted without proof. Such a hypothesis was termed a postulate. While
50 5 Polynomial and Exponential Complexity

the axioms were common to many sciences, the postulates of each particular
science were different. Their validity had to be established by means of real-world
experience. Aristotle warns that the content of a science cannot be successfully
communicated if the learner is in doubt about the truth of the postulates.
The classical approach is well illustrated by Euclid’s Elements, where a list of
postulates is given (common-sensical geometric facts drawn from our experience),
followed by a list of common notions (very basic, self-evident assertions).
• It is possible to draw a straight line from any point to any other point.
• It is possible to extend a line segment continuously in both directions.
• It is possible to describe a circle with any center and any radius.
• It is true that all right angles are equal to one another.
• (Parallel postulate) It is true that if a straight line falling on two straight lines
make the interior angles on the same side less than two right angles, the two
straight lines, if produced indefinitely, intersect on that side on which are the
angles less than the two right angles.
By making small changes in the axiom system, one can define a new subfield of
mathematics. One impressive example is the definition of non-Euclidean geome-
tries.
Non-Euclidean geometry arises by replacing the parallel postulate with an
alternative. In this case, one obtains hyperbolic geometry and elliptic geometry.
The beginning of the nineteenth century would finally witness decisive steps
in the creation of non-Euclidean geometry. Circa 1813, Carl Friedrich Gauss
and, independently around 1818, the German professor of law Ferdinand Karl
Schweikart had the germinal ideas of non-Euclidean geometry worked out, but
neither published any results. Schweikart’s nephew, Franz Taurinus, did publish
important results of hyperbolic trigonometry in two papers in 1825 and 1826, yet
while admitting the internal consistency of hyperbolic geometry, he still believed in
the special role of Euclidean geometry.
Then, in 1829–1830, the Russian mathematician Nikolai Ivanovich Lobachevsky
and, in 1832, the Hungarian mathematician János Bolyai separately and inde-
pendently published treatises on hyperbolic geometry. Consequently, hyperbolic
geometry is called Lobachevskian or Bolyai-Lobachevskian geometry, as both
mathematicians, independent of each other, are the basic authors of non-Euclidean
geometry. Gauss mentioned to Bolyai’s father, when shown the younger Bolyai’s
work, that he had developed such a geometry several years before, though he did
not publish. While Lobachevsky created a non-Euclidean geometry by negating the
parallel postulate, Bolyai worked out a geometry where both the Euclidean and
the hyperbolic geometry are possible depending on a parameter k. Bolyai ended his
work by mentioning that it is not possible to decide through mathematical reasoning
alone if the geometry of the physical universe is Euclidean or non-Euclidean; this is
a task for the physical sciences.
Bernhard Riemann, in a famous lecture in 1854, founded the field of Riemannian
geometry, discussing in particular the ideas now called manifolds, Riemannian met-
ric, and curvature. He constructed an infinite family of non-Euclidean geometries by
5.4 Axioms 51

giving a formula for a family of Riemannian metrics on the unit ball in Euclidean
space. The simplest of these is called elliptic geometry, and it is considered a non-
Euclidean geometry due to its lack of parallel lines.
By formulating the geometry in terms of a curvature tensor, Riemann allowed
non-Euclidean geometry to apply to higher dimensions. Beltrami (1868) was the
first to apply Riemann’s geometry to spaces of negative curvature.
It was Gauss who coined the term non-Euclidean geometry. He was referring
to his own work, which today we call hyperbolic geometry or Lobachevskian
geometry. Several modern authors still use the generic term non-Euclidean geometry
to mean hyperbolic geometry.
Arthur Cayley noted that distance between points inside a conic could be defined
in terms of logarithm and the projective cross-ratio function. The method has
become called the Cayley-Klein metric because Felix Klein exploited it to describe
the non-Euclidean geometries in articles in 1871 and 1873 and later in book form.
The Cayley-Klein metrics provided working models of hyperbolic and elliptic
metric geometries, as well as Euclidean geometry.
Klein is responsible for the terms hyperbolic and elliptic (in his system, he called
Euclidean geometry parabolic, a term that generally fell out of use). His influence
has led to the current usage of the term non-Euclidean geometry to mean either
hyperbolic or elliptic geometry.
Euclidean geometry can be axiomatically described in several ways. Unfortu-
nately, Euclid’s original system of five postulates (axioms) is not one of these, as his
proofs relied on several unstated assumptions that should also have been taken as
axioms. Hilbert’s system consisting of 20 axioms most closely follows the approach
of Euclid and provides the justification for all of Euclid’s proofs. Other systems,
using different sets of undefined terms, obtain the same geometry by different paths.
All approaches, however, have an axiom that is logically equivalent to Euclid’s
fifth postulate, the parallel postulate. Hilbert uses the Playfair axiom form, while
Birkhoff, for instance, uses the axiom that says that there exists a pair of similar
but not congruent triangles. In any of these systems, removal of the one axiom
equivalent to the parallel postulate, in whatever form it takes, and leaving all the
other axioms intact, produces absolute geometry. As the first 28 propositions of
Euclid (in The Elements) do not require the use of the parallel postulate or anything
equivalent to it, they are all true statements in absolute geometry.
To obtain a non-Euclidean geometry, the parallel postulate (or its equivalent)
must be replaced by its negation. Negating the Playfair’s axiom form, since it is a
compound statement (.. . . there exists one and only one .. . .), can be done in two
ways:
Either there will exist more than one line through the point parallel to the given
line or there will exist no lines through the point parallel to the given line. In the
first case, replacing the parallel postulate (or its equivalent) with the statement In
a plane, given a point P and a line l not passing through P, there exist two lines
through P, which do not meet l and keeping all the other axioms yields hyperbolic
geometry. The second case is not dealt with as easily. Simply replacing the parallel
postulate with the statement In a plane, given a point P and a line l not passing
52 5 Polynomial and Exponential Complexity

through P, all the lines through P meet l does not give a consistent set of axioms.
This follows since parallel lines exist in absolute geometry, but this statement says
that there are no parallel lines. This problem was known (in a different guise) to
Khayyam, Saccheri, and Lambert and was the basis for their rejecting what was
known as the obtuse angle case. To obtain a consistent set of axioms that includes
this axiom about having no parallel lines, some other axioms must be tweaked.
These adjustments depend upon the axiom system used. Among others, these tweaks
have the effect of modifying Euclid’s second postulate from the statement that line
segments can be extended indefinitely to the statement that lines are unbounded.
Riemann’s elliptic geometry emerges as the most natural geometry satisfying this
axiom.
On a sphere, the sum of the angles of a triangle is not equal to 180.◦ . The surface
of a sphere is not a Euclidean space, but locally, the laws of the Euclidean geometry
are good approximations. In a small triangle on the face of the earth, the sum of the
angles is very nearly 180.◦ . Models of non-Euclidean geometry are mathematical
models of geometries, which are non-Euclidean in the sense that it is not the case
that exactly one line can be drawn parallel to a given line l through a point that is
not on l. In hyperbolic geometric models, by contrast, there are infinitely many lines
through A parallel to l, and in elliptic geometric models, parallel lines do not exist.
Euclidean geometry is modeled by our notion of a flat plane. The simplest model
for elliptic geometry is a sphere, where lines are great circles (such as the equator or
the meridians on a globe), and points opposite each other are identified (considered
to be the same). The pseudosphere has the appropriate curvature to model hyperbolic
geometry (Fig. 5.5).
We note here that these axiom systems for geometry branch out.

Fig. 5.5 Non-Euclidean

Geometry on a Sphere
5.4 Axioms 53

Until the 1970s, it was completely self-evident that proofs should consist of
logical steps that approached the goal to be proven. But suddenly computers came
into play. Their high speed made it impossible to retrace the steps individually;
only the result counted. This was initially controversial among mathematicians, but
since a correct result was kept in mind, such proofs were later accepted without any
problems.
In mathematics, the four-color theorem, or the four-color map theorem, states
that no more than four colors are required to color the regions of any map so that no
two adjacent regions have the same color. Adjacent means that two regions share a
common boundary curve segment, not merely a corner where three or more regions
meet. It was the first major theorem to be proved using a computer. Initially, this
proof was not accepted by all mathematicians because the computer-assisted proof
was infeasible for a human to check by hand. The proof has gained wide acceptance
since then, although some doubters remain.
The four-color theorem was proved in 1976 by Kenneth Appel and Wolfgang
Haken after many false proofs and counterexamples (unlike the five-color theorem,
proved in the 1800s, which states that five colors are enough to color a map). To
dispel any remaining doubts about the Appel-Haken proof, a simpler proof using
the same ideas and still relying on computers was published in 1997 by Robertson,
Sanders, Seymour, and Thomas. In 2005, the theorem was also proved by Georges
Gonthier with general-purpose theorem-proving software.
The intuitive statement of the four-color theorem—given any separation of a
plane into contiguous regions, the regions can be colored using at most four colors
so that no two adjacent regions have the same color—needs to be interpreted
appropriately to be correct.
First, regions are adjacent if they share a boundary segment; two regions that
share only isolated boundary points are not considered adjacent. (Otherwise, a
map in a shape of a pie chart would make an arbitrarily large number of regions
“adjacent” to each other at a common corner and require arbitrarily large number
of colors as a result.) Second, bizarre regions, such as those with finite area but
infinitely long perimeter, are not allowed; maps with such regions can require more
than four colors. (To be safe, we can restrict to regions whose boundaries consist of
finitely many straight line segments. It is allowed that a region has enclaves and that
it entirely surrounds one or more other regions.) Note that the notion of contiguous
region (technically, connected open subset of the plane) is not the same as that
of a country on regular maps, since countries need not be contiguous (they may
have exclaves, e.g., the Cabinda Province as part of Angola, Nakhchivan as part of
Azerbaijan, Kaliningrad as part of Russia, France with its overseas territories, and
Alaska as part of the United States are not contiguous). If we required the entire
territory of a country to receive the same color, then four colors are not always
sufficient (Fig. 5.6) [6].
However, the verifiability of a proof by humans also has its rough edges. This is
shown by the results of A. Wiles in the proof of Fermat’s Great Theorem. In number
theory, Fermat’s Last Theorem (sometimes called Fermat’s conjecture, especially
in older texts) states that no three positive integers, a, b, and c, satisfy the equation
54 5 Polynomial and Exponential Complexity

Fig. 5.6 The Four-color

Problem

a n + bn = cn for any integer value of n greater than 2. The cases n = 1 and n = 2

.

have been known since antiquity to have infinitely many solutions.

The proposition was first stated as a theorem by Pierre de Fermat around 1637
in the margin of a copy of Arithmetica. Fermat added that he had a proof that was
too large to fit in the margin. Although other statements claimed by Fermat without
proof were subsequently proven by others and credited as theorems of Fermat (e.g.,
Fermat’s theorem on sums of two squares), Fermat’s Last Theorem resisted proof,
leading to doubt that Fermat ever had a correct proof. Consequently, the proposition
became known as a conjecture rather than a theorem. After 358 years of effort by
mathematicians, the first successful proof was released in 1994 by Andrew Wiles
and formally published in 1995. It was described as a stunning advance in the
citation for Wiles’s Abel Prize award in 2016. It also proved much of the Taniyama-
Shimura conjecture, subsequently known as the modularity theorem, and opened up
entire new approaches to numerous other problems and mathematically powerful
modularity lifting techniques.
The unsolved problem stimulated the development of algebraic number theory
in the nineteenth and twentieth centuries. It is among the most notable theorems in
the history of mathematics and prior to its proof was in the Guinness Book of World
Records as the “most difficult mathematical problem,” in part because the theorem
has the largest number of unsuccessful proofs.
Starting in mid-1986, based on successive progress of the previous few years
of Gerhard Frey, Jean-Pierre Serre, and Ken Ribet, it became clear that Fermat’s
Last Theorem could be proven as a corollary of a limited form of the modularity
theorem (unproven at the time and then known as the “Taniyama-Shimura-Weil
conjecture”). The modularity theorem involved elliptic curves, which was also
Wiles’s own specialist area.
The conjecture was seen by contemporary mathematicians as important but
extraordinarily difficult or perhaps impossible to prove.
Despite this, Wiles, with his from-childhood fascination with Fermat’s Last
Theorem, decided to undertake the challenge of proving the conjecture, at least to
the extent needed for Frey’s curve. He dedicated all of his research time to this
problem for over six years in near-total secrecy, covering up his efforts by releasing
prior work in small segments as separate papers. In June 1993, he presented his
proof to the public for the first time at a conference in Cambridge.
He gave a lecture a day on Monday, Tuesday, and Wednesday with the title
Modular Forms, Elliptic Curves, and Galois Representations. “There was no hint in
5.5 Finding Contradictions: Inconsistency Proofs 55

the title that Fermat’s last theorem would be discussed,” Dr. Ribet said. .. . . Finally,
at the end of his third lecture, Dr. Wiles concluded that he had proved a general
case of the Taniyama conjecture. Then, seemingly as an afterthought, he noted that
meant that Fermat’s last theorem was true. q.e.d.
In August 1993, it was discovered that the proof contained a flaw in one area.
Wiles tried and failed for over a year to repair his proof. According to Wiles,
the crucial idea for circumventing—rather than closing—this area came to him on
September 19, 1994, when he was on the verge of giving up. Together with his
former student Richard Taylor, he published a second paper that circumvented the
problem and thus completed the proof. Both papers were published in May 1995 in
a dedicated issue of the Annals of Mathematics.
The problems with the proof of the abc assumption were even more difficult. The
abc conjecture refers to numerical expressions of the type .a + b = c. The statement,
which comes in several slightly different versions, concerns the prime numbers that
divide each of the quantities a, b, and c. Every whole number, or integer, can be
expressed in an essentially unique way as a product of prime numbers—those that
cannot be further factored out into smaller whole numbers: for example, 15 = 3 .× 5
or 84 = 2 .× 2 .× 3 .× 7. In principle, the prime factors of a and b have no connection
to those of their sum, c. But the abc conjecture links them together. It presumes,
roughly, that if a lot of small primes divide a and b, then only a few, large ones
divide c.
The publication of the proof comprised 520 pages and is not yet fully confirmed.

5.5 Finding Contradictions: Inconsistency Proofs

If a great many rules are to hold simultaneously, it is an extraordinarily difficult task

to determine whether this results in something at all. It can be that no application of
the rules results in true.
Many popular antinomies and paradoxes do not refer to a calculus but are based
on intuitive, opaque, illicit ways of reasoning. Therefore, it is important to regulate
logical reasoning in calculi; only then the inconsistent steps leading to paradoxes
become clearly visible, for example, in the liar paradox: “This sentence is false.”
The sentence is true exactly when it is false.
In logic, the proof of non-contradiction succeeds gradually in the twentieth
century. David Hilbert (Fig. 5.7) published his program to prove that arithmetic
is free of contradictions starting in 1918; for this purpose, he outlined methods
that are more demanding because contradictions must be excluded for all possible
derivations.
This is a quite fundamental realization of great importance. As long as the
statements are clearly defined and linked by propositional operators, one moves on
the safe side! Therefore, logical functions and equations are a core area of artificial
intelligence [1].
56 5 Polynomial and Exponential Complexity

Fig. 5.7 David Hilbert (1862–1943)

Fig. 5.8 Four queens on a

chessboard, not threatening
each other

In the daily business life, it is always necessary to define all the objects included
and the respective laws and rules as precisely as possible.

5.6 Problems from Combinatorics

Many problems that have a combinatorial character are defined by rules; sometimes,
they look like a game. When the implications are transformed into disjunctions,
then the classical search algorithms disappear, and they can be solved by logical
equations. This is very typical for the results of artificial intelligence; some fields
are completely replaced by other methods.
Example We take a chessboard of size .4 × 4 (Fig. 5.8). We want to position four
queens on this board so that none threatens another. We know that a queen can
move horizontally, vertically, and diagonally, so there must be only one queen in
each horizontal, each vertical, and each diagonal. This can be expressed by rules:
.a1 = 1 expresses the fact that there is a queen on this field and .a1 = 0 the fact that

there is not a queen on this field.

5.6 Problems from Combinatorics 57

solution space for the problem with four queens

a1 a2 a3 a4 b1 b2 b3 b4 c1 c2 c3 c4 d1 d2 d3 d4
1 0 0 0 0 0 – – 0 – 0 – 0 – – 0
0 1 0 0 0 0 0 – – 0 – 0 – 0 – –
0 0 1 0 – 0 0 0 0 – 0 – – – 0 –
0 0 0 1 – – 0 0 – 0 – 0 0 – – 0
0 0 – – 1 0 0 0 0 0 – – 0 – 0 –
0 0 0 – 0 1 0 0 0 0 0 – – 0 – 0
– 0 0 0 0 0 1 0 – 0 0 0 0 – 0 –
– – 0 0 0 0 0 1 – – 0 0 – 0 – 0
0 – 0 0 0 – – – 1 0 0 0 0 0 – –
– 0 – 0 0 0 0 – 0 1 0 0 0 0 0 –
0 – 0 – – 0 0 0 0 0 1 0 – 0 0 0
– 0 – 0 – – 0 0 0 0 0 1 – – 0 0
0 – – 0 0 – 0 – 0 0 – – 1 0 0 0
– 0 – – – 0 – 0 0 0 0 – 0 1 0 0
– – 0 – 0 – 0 – – 0 0 0 0 0 1 0
0 – – 0 – 0 – 0 – – 0 0 0 0 0 1

Then the following equations can be built:

1. For each vertical, you write down the possibilities you have:

a1 ∨ a2 ∨ a3 ∨ a4.
.

In the first line a, the queen can be on one of the four squares, a1, a2, a3, or a4.
2. One formulates the constraints that eliminate impermissible solutions:

a1 a2 a3 a4 ∨ a2 a1 a3 a4 ∨ a3 a1 a2 a4 ∨ a4 a1 a2 a3
.

∨a1 b1 c1 d1 ∨ a2 b2 c2 d2 ∨ a3 b3 c3 d3 ∨ a4 b4 c4 d4
∨a1 b2 c3 d4 ∨ a2 b1 b3 c4 ∨ a3 b2 c1 b4 ∨ a4 b3 c2 d1.

Thus, a queen on a1 prevents a queen on one of the squares a2, a3, a4 b1, c1,
d1, b2, c3, and d4.
Analogously, one proceeds with the other verticals b, c, and d. The conjunctions
for a vertical are connected by .∨, and the verticals themselves are connected by .∧.
So we get an equation

.disj unction1 ∧ disj unction2 ∧ disj unction3 ∧ disj unction4 = 1.

If we use the representation as a matrix, then the solution is almost given.

The symbols of a line have a very descriptive meaning; the value 1 occurs only
once in each line and indicates the position of a lady. The value 0 shows the squares
58 5 Polynomial and Exponential Complexity

that are no longer available due to the position of the queen. The positions of the
queen on b2, b3, c2, and c3 leave only four squares open; all other positions still
allow the occupation or non-occupation of six squares.
If one uses an appropriate solution program [2], then this is a simple matter. A
search for a solution can even be done by hand.
• Queen on a1: the next possibility is a queen on b3; with this, you can’t place
anything on the vertical c.
• Queen on a2: the next possibility is a queen on b4; this commits one to c1 and a3
and gives the solution shown in Fig. 5.6.
• In a mirror image, one gets the second solution Qa3 – Qb1 – Qc4 – Qd1.
• There are no further possibilities.
Following this methodology, one can extend this to chessboards of any size
(Fig. 5.9). Modifications of the problem are also possible (Fig. 5.10).
One suspects that there is no solution with a queen on a1 or a2. For reasons of
symmetry, the squares a7 and a8 are also omitted. A modification of the problem

Fig. 5.9 Eight queens on a

chessboard, not threatening
each other

Fig. 5.10 Nine queens and a pawn—ten queens and two pawns
5.7 Coloring Problems 59

is, for example, to place additional pawns on the board, in the first example on d7
and in the second on c3 and e5. These pawns interrupt the lines of action of the
queens, so one suspects that the number of queens on the board might increase. You
can see that one pawn allows nine queens; ten queens are possible with two pawns.
The methodology remains completely the same; only the corresponding equation
changes.

5.7 Coloring Problems

The following problem shows how far the possibilities of computers reach today.
The background is a problem from electronics. On two conducting rails, there is a
row of contacts. Each contact of one rail is conductively connected to each contact
of the other rail. Each connection has one of four possible properties, which is
described by the four colors red, green, blue, and amethyst. The circuit
does not work if four edges (1 – 2 – 3 – 4 – 1) have the same properties, i.e., the
same color. It is important that the last edge returns to the starting point, so the edge
train must be closed (Fig. 5.11).
In 2010, it was known that for 16 contacts on both tracks, it is possible to avoid
such conflicts. It was also known that for 19 contacts, it is not possible to work
without conflicts. The interesting question was how it looks like for 17 and 18
contacts. A logical model for this problem is constructed in the following way with
a matrix .18 × 18:
• the one track is assigned to the columns of a matrix;
• the second track is assigned to the rows of a matrix.
The difficulty of the solution is due to the exponential complexity of the problem:
there are .18 × 18 = 324 positions in the matrix. Each position can be colored with
one of four colors, resulting in .4324 = 2648 possibilities for possible colorings. This
number is about twice the number of atoms in the universe.
One proceeds exactly as in the checker problem. One begins with a matrix .18 ×
18, and first write down what possibilities you have.
• For each position (m,n) from (1,1) to (18,18) holds:

Fig. 5.11 To be investigated 2 4

is the sequence 1 – 2 – 3 – track 1
4–1

track 2
1 3
60 5 Polynomial and Exponential Complexity

(xm,n red ∨ xm,n green ∨ xm,n blue ∨ xm,n amethyst )

.

This results in .4 · 324 = 1296 variables.

• The selection of two horizontal and two vertical lines leads to
18 · 17 18 · 17
. · = 23,409
2 2
rectangles.
• The vertices of a rectangle must not have the same color. As an example, we use
the top left rectangle with the points 11 – 12 – 21 – 22.

x11red ∧ x12red ∧ x21red ∧ x22red = x11red ∨ x12red ∨ x21red ∨ x22red = 1

.

is the constraint that must be satisfied based on the rule: not one rectangle with
the same color. The same procedure is used for the other colors and all possible
rectangles. The formulas of all rectangles are again linked by .∧.
It is absolutely impossible for a person to find a solution with paper and pencil,
but he can still check the solution. With other problems, even that is no longer
possible.
Here, another interesting feature of such problems becomes apparent. Every
solution for the matrix .18 × 18 is also a solution for matrices that are smaller. You
just have to delete the corresponding number of rows and columns. The also open
question for .17 × 17 did not have to be examined at all, because one only has to
delete one row and one column of the solution for .18 × 18.
The solutions make up only a very small part of all possibilities; nevertheless,
there are very many solutions, because each interchange of rows and columns results
in a solution again. In addition, you can also swap the colors with each other. This
results in a total of

.18! × 18! × 4!

solutions. The solution of the problem took the years 2010–2012. The basic logical
algorithm still had to be supplemented by many subtleties, but in the end, a solution
with tolerable effort was found: 20 graphics processors working in parallel and
1 week of computing time were sufficient for the solution (Fig. 5.12).
What comes out of this solution: the original search algorithms for this problem
belong to AI. The development of this area made it a mathematical problem, which
is solved by deterministic algorithms. And still, a lot of computer power is required.
There are many other combinatorial problems of this kind.
• The problem of the traveling salesman
Figure 5.13 shows the optimal travel route of a traveling salesman through the
15 largest cities in Germany. The given route is the shortest of 43,589,145,600
possible routes. The problem of the traveling salesman is a com-
5.7 Coloring Problems 61

Fig. 5.12 One solution for

the matrix .18 × 18

Fig. 5.13 There are many

different tours of the same
cities

binatorial optimization problem in operations research and theoretical computer

science. The task is to choose an order for visiting multiple places such that
no station but the first is visited more than once, the total travel distance of the
traveling salesman is as short as possible, and the first station is equal to the last.
Since its first mention as a mathematical problem in 1930, many researchers
have studied it and developed and tested new optimization methods, which are
currently being used for other optimization problems. Today, a large
number of heuristic and exact methods are available, with which even difficult
cases with several thousand cities have been solved optimally.
The traveling salesman problem already appears in its pure form in many
practical applications, for example, in route planning or in the design of
microchips. Even more often, however, it appears as a subproblem, for example,
in the distribution of goods, in the planning of tours of a customer, or breakdown
service (Fig. 5.13). Here, the terms “city” and “distance” are not to be taken
literally; rather, cities represent, for example, customers to be visited or spare
parts depots, while distance stands for travel time or cost. In many practical
applications, additional constraints such as time windows or limited resources
must also be taken into account, which makes solving the problem much more
difficult.
• Hamiltonian and Eulerian graphs also involve round trips but deal with different
problems. Again, locations and connections between them are given.
An Euler circle is a cycle in a graph containing all edges of the graph; the last
edge must lead back to the starting point (Fig. 5.14).
62 5 Polynomial and Exponential Complexity

Fig. 5.14 The Königsberg

Bridge Problem

Leonhard Euler was confronted with the problem in Königsberg whether it is

possible to cross all bridges from one starting point without using a bridge twice.
That is why such paths are named after him.
In a Hamiltonian graph, each place is entered exactly once, but not
all edges have to be used. The Eulerian path corresponds to the behavior of a
letter carrier, the Hamiltonian path more to the behavior of a company supplying
branch offices. Of course, this is not a serious problem for individual localities
nowadays, but if you consider the network of airports or container loading and
unloading points, these problems have a great economic significance.
• The backpack problem
This problem often arises when preparing for air travel and knowing that only
20 kg of luggage per person is allowed in the cabin. From a set of objects, each
having a weight and a utility, a subset is to be selected whose total weight does
not exceed a given weight bound. Under this constraint, the utility of the selected
objects is to be maximized.

References

1. Bernd Steinbach, Christian Posthoff. Logic Functions and Equations—Fundamentals and

Applications using the XBOOLE-Monitor, Third Edition, Springer Nature, 2021.
2. Christian Posthoff, Bernd Steinbach. Mathematik—Ohne Sorgen an der Uni II—Nutze Microsoft
Mathematics, bookboon.com, 2017, 978-87-403-1949-1
3. https://www.juraforum.de/lexikon/regel
4. https://kinder.wdr.de/tv/wissen-macht-ah/bibliothek/kuriosahbibliothek-wer-hat-das-recht-
erfunden-100.html
5. https://en.wikipedia.org/wiki/WilesproofofFermat’sLastTheorem
6. https://www.en.wikipedia.org/wiki/Four-color-theorem https://www.ki-strategie-deutschland.
de/home.html
Chapter 6
Game Theory, Cybernetics, and
Information Theory

6.1 Game Theory

Two other outstanding scientists, Norbert Wiener and John von Neumann (Fig. 6.1),
add to the list of famous mathematicians and computer scientists who are of
outstanding importance in the field of artificial intelligence.
Work on quantum mechanics established John von Neumann’s reputation in
America. From 1933, he worked at the newly founded, sophisticated Institute
for Advanced Study in Princeton as a Professor of Mathematics. His
colleagues there included Albert Einstein and Hermann Weyl. Like them, von
Neumann finally emigrated to the USA after Hitler came to power.
John von Neumann made outstanding contributions in many areas of mathemat-
ics. As early as 1928, an essay by the mathematician Émile Borel on minimax
properties had given him ideas that later led to one of his most original designs,
game theory. von Neumann proved the minimax theorem in 1928 for the
existence of an optimal strategy in “zero-sum games”. Zero-sum games
describe situations in which the sum of all players’ combined wins and losses is
zero. Together with the economist Oskar Morgenstern, he wrote the classic
Theory of Games and Economic Behavior in 1944, which also deals with the
generalization to n-person games, which is important for economics. Decisive
for representation and solution is the level of information of the players. Three
notions are distinguished here, complete, perfect (or perfect) information, and
perfect memory, depending on whether the player is aware of the rules of the
game, the moves of the other players, and his own past information. Standard is the
game with complete information as well as perfect memory. Perfect information is
not one of the standard assumptions, as it would be obstructive in explaining many
simple conflicts.
Complete information, the knowledge of all players about the rules of the game,
is an assumption that is considered a prerequisite for joint play in normal games.
Perfect memory is each player’s knowledge of all information that was already

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 63

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_6
64 6 Game Theory, Cybernetics, and Information Theory

Fig. 6.1 John von Neumann

(1903–1957)

available to him in the past. Although this assumption always seems to be fulfilled,
at least in principle, there are counterexamples: If a game involves a team of
cooperating players, as in Skat, the individual player at the time of his own decision
no longer knows the information context of past moves made by a partner on the
basis of his cards.

6.2 Cybernetics

Norbert Wiener (1894–1964) was an American mathematician and philosopher

(Fig. 6.2). He is known as the founder of cybernetics, a term he coined in his work
Cybernetics: Or Control and Communication in the Animal and the Machine (1948).
He thus established the scientific and mathematical basis for control theory and
control engineering to calculate the dynamics and stability of feedback systems and
described their analogies to the human brain (through feedback via sensory organs)
and to social organizations (through feedback via communication and observation).
The book contains the following main topics:
• Newtonian and Bergsonian time
• Groups and Statistical Mechanics
• Time Series, Information and Communication
• Feedback and Oscillation
• Computing Machines and the Nervous System
• Gestalt and Universals
• Cybernetics and Psychopathology
• Information, Language and Society
• On Learning and Self-Reproducing Machines
• Brain waves and Self-Organizing systems
6.2 Cybernetics 65

Fig. 6.2 Norbert Wiener

(1894–1964)

If one wants to single out one chapter in particular, it is perhaps the chapter
Computing Machines and the Nervous System. There he justifies the
necessity of working with the dual system, but from a completely different angle. He
looks at the cost of storing and processing numbers and reasons that the dual system
is the least costly in both technical realization and work, at a time when there were
no computers!
He writes: “The ideal computing machine must then have all its data inserted at
the beginning, and must be as free as possible from human interference to the very
end. This means that not only must the numerical data be inserted at the beginning,
but also the rules for combining them, in the form of instructions covering every
situation which may arise in the course of computation.”
He justifies the dual system by basing it on the way neurons (nerve cells) work:
They are either in a resting state or, when they fire, they undergo a series of changes
that are almost independent of the nature and intensity of the stimulus.
The chapter “Cybernetics and Psychopathology” is extremely interesting and
reaches far into the future. He poses the question, “How the brain avoids gross
errors, gross malfunctions resulting from the malfunction of individual compo-
nents.” Similar questions, relating to computing machines, are of great practical
importance. “It is quite possible for a chain of arithmetic operations to consist of .109
individual steps. Under these circumstances, the chance that at least one operation
will go wrong is anything but negligible.”
And he is not at all afraid to consider social problems; remember that these
sentences were written over 70 years ago:
66 6 Game Theory, Cybernetics, and Information Theory

The psychology of the fool has become a subject worthy of the attention of scoundrels.
Instead of keeping his own interest in mind, as von Neumann’s player does, the fool acts
in a way that is, on the whole, as predictable as the struggles of a rat in a maze. This
policy of lies - or rather, of truth-telling claims - will get him to buy a particular brand
of cigarettes; this policy will get him, the party hopes, to run for a particular office -
or to participate in a political witch hunt. A certain mix of religion, pornography, and
pseudoscience will sell an illustrated newspaper. A certain mix of ingratiation, bribery,
and intimidation will persuade a young scientist to work on guided missiles or the atomic
bomb. To find this out we have our machinery of radio fan ratings, straw-man polls, opinion
surveys, and other psychological investigations which have the common man as their object;
and there are always statisticians, sociologists, and economists offering their services to
these undertakings.

One can only admire his enormous foresight. In fact, everything he described
with foresight has come to pass. In the penultimate chapter, written in 1961, he
writes:
There is nothing more dangerous than World War III. It’s worth thinking about whether part
of the danger might not lie in the ill-considered use of learning machines. Time and again
I’ve heard the statement that learning machines can’t expose us to new dangers because
we can turn them off when we feel like it. But can we? In order to effectively shut down
a machine, we must be in possession of information about whether the danger point has
been reached. The mere fact that we built the machine does not guarantee that we have the
right information to do so. This is already contained in the statement that the chess playing
machine can defeat the man who programmed it. Moreover, the very speed of operation of
modern digital machines is opposed to our ability to perceive and think through the signs of
danger.

He also discusses the question of whether the machines can create successors
that possess an even higher level of intelligence than their parent generation, and his
answer is yes! The intelligent systems can continue to improve on their own.
All of these problems still exist today and have been exacerbated many times
over by the technical and social developments from 1948 to the present day.

6.3 Information Theory

We return once again to C. Shannon, who also founded information theory, a math-
ematical theory from the field of probability theory and statistics. It deals with con-
cepts such as information and entropy, information transmission,
data compression, encoding, and related topics. In addition to mathemat-
ics, computer science, and communications engineering, the theoretical consider-
ation of communication by information theory is also used to describe commu-
nication systems in other fields; as examples, we see the media in journalism,
nervous systems in neurology, DNA and protein sequences in molecular biology,
knowledge in information science, and documentation. Shannon’s theory uses the
term entropy to characterize the information content (also called information
density) of messages. The more non-uniform the structure of a message is, the
higher is its entropy.
6.3 Information Theory 67

A technical breakthrough occurred in the late 1930s when pulse-code modulation

made it possible to represent a message, which exists as a continuum, discretely
to a satisfactory approximation. With this method, it became possible to telegraph
speech. Shannon, who worked for Bell Telephone Laboratories, was familiar with
the technical development. The great importance of his theory for technology
lies in the fact that he defined information as a physical quantity with a unit of
measurement or counting, the bit. This allowed to compare quantitatively exactly the
effort for the technical transmission of information in different forms (sounds, signs,
pictures), to determine the efficiency of codes as well as the capacity of information
memories and transmission channels.
The core idea of information theory is that the informational value of a
communicated message depends on the degree to which the content of the message
is surprising. If a highly likely event occurs, the message carries very little
information. On the other hand, if a highly unlikely event occurs, the message is
much more informative. For instance, the knowledge that some particular number
will not be the winning number of a lottery provides very little information, because
any particular chosen number will almost certainly not win. However, knowledge
that a particular number will win a lottery has high informational value because it
communicates the outcome of a very low probability event.
The information content, also called the surprisal or self-information, of an event
E is a function that increases as the probability .p(E) of an event decreases. When
.p(E) is close to 1, the surprisal of the event is low, but if .p(E) is close to 0, the

surprisal of the event is high. This relationship is described by the function

 
1
. log (6.1)
p(E)

log is the logarithm function that gives 0 surprise when the probability of the event
.

is 1. In fact, the logarithm function is the only function that satisfies this specific set
of characterization. Hence, we can define the information, or surprisal, of an event
E by the entropy (Fig. 6.3)

I (E) = − log2 (p(E)).

.

Shannon himself calls his work a mathematical theory of communication. He

explicitly excludes semantic and pragmatic aspects of information, i.e. statements
about the “content” of transmitted messages as well as their meaning for the
receiver. This means that a meaningful message is transmitted as conscientiously
as a random sequence of letters. Thus, although Shannon theory is usually referred
to as information theory, it makes no direct statement about the information content
of transmitted messages.
If you want to summarize all the knowledge that has been conveyed so far, then
you will certainly be amazed at how many things had to come together to reach this
level. And it goes on like this.
68 6 Game Theory, Cybernetics, and Information Theory

Fig. 6.3 The Information Content approaches 0 for .p = 1

Chapter 7
Machine Learning and Neural Networks

Amazing results have been obtained in recent years by using neural networks.
They are currently the key to great successes in many fields, combined with the
hope of obtaining further important results. Behind this is the effort to develop
mathematical models that will allow the activity of neurons to be described and
then recreated on computers.

7.1 Neural Networks

A neural network is a mathematical model that mimics the way the human
brain works. Of course, the replication of its functioning on computers is of interest
to artificial intelligence.
In 1943, mathematicians Warren McCulloch (Fig. 7.1) and Walter
Pitts (Fig. 7.2) introduced the neuron as a logical threshold element with multiple
inputs and a single output. It could take on the states true and false as Boolean
variables and fired when the sum of the input signals exceeded a threshold
(Fig. 7.3).
The neuron can be divided into different sections [1]:
• the neuron,
• dendrites,
• the nucleus,
• the soma,
• the axon,
• the axon hillock,
• the myelin sheath,
• the Ranvier’s lacing ring,
• the Schwann cell,
• synaptic terminal knob.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 69

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_7
70 7 Machine Learning and Neural Networks

Fig. 7.1 Warren McCulloch

(1898–1969)

Fig. 7.2 Walter Pitts

(1923–1969)

The dendrites are cell extensions of the cell body and form the contact with other
cells or neurons. They receive the initial excitation signals and transmit them to the
cell body.
The soma is the cell body of the neuron. It forms an area filled with cytoplasm
that contains the important cell organelles. These include, for example, the cell
nucleus, the mitochondria, the endoplasmic reticulum, and the
Golgi apparatus.
The axon hill forms the transition from the soma to the axon. Electrical
signals are collected and summed here until a certain threshold or threshold potential
is exceeded. Only then is a signal transmitted to the axon. These signals are called
7.1 Neural Networks 71

Fig. 7.3 A possible structure of a perceptron

action potentials. This prevents our body from passing on even the smallest
signal. Without this filtering mechanism, processing the relevant stimuli would be
impossible.
The long extension of the nerve cell that emerges from the axon hillock is
called axon or neurite. The task of the axon is to transmit action potentials
to nerve or muscle cells. To ensure that the transmission of electrical signals works
as quickly as possible and without losses, the axon is insulated like an electrical
cable. For this purpose, the process is sheathed by supporting or sheath cells. In the
peripheral nervous system (outside the brain and spinal cord), they are also called
Schwann cells. In the central nervous system (brain and spinal cord), they are called
oligodendrocytes.
As an electrical insulation layer, the sheath cells form a so-called myelin sheath
around the axons. The sheath is repeatedly interrupted by exposed axon areas. The
non-enveloped area of an axon is called Ranvier’s lacing ring. They allow
the speed of excitation conduction to be increased. This is because excitation can
also be “jumped” from one lacing ring to the next. Thus, the long sheathed areas are
simply bridged.
The synaptic terminal knobs form the end of a neuron. The electrical signal is
transmitted here to the next neuron or, for example, to a sensory or muscle cell. For
this purpose, the electrical signal is usually converted into a chemical signal. The
connection at the end of a nerve cell with another cell is called synapse. In most
cases, these are chemical synapses. The terminal button brings chemical molecules
into the synaptic cleft, the gap between the two cells. There they meet receptors and
pass on the excitation. Electrical signals are thus triggered again in the next nerve
cell.
According to their function, neurons can be divided into three different types:
72 7 Machine Learning and Neural Networks

Fig. 7.4 Frank Rosenblatt

• sensory neurons conduct information from sensory organs or other organs in the
body to the brain;
• motor neurons conduct information from the brain or spinal cord to muscles or
glands;
• interneurons, as part of the central nervous system, interconnect neurons with
each other; they have a mediating function.
These concepts are the neurobiological analogy of an action potential, that a
neuron emits upon a critical change in its membrane potential. McCulloch and Pitts
showed that by suitably combining several such neurons, any propositional function
(.∧, ∨, ¬) is realizable.
In 1957, Frank Rosenblatt (Fig. 7.4) published the perceptron model,
which is still the basis of artificial neural networks (Fig. 7.5). It consists in the
basic version (simple perceptron) of a single artificial neuron with adjustable
weights and a threshold.
A human is assumed to have 100 billion dendrites with a total length of many
hundreds of kilometers. We consider a single neuron with input values .x1 . . . xn , a
threshold .θ , and weights .w1 . . . wn for the connections of the input values with the
output.
⎧ n
⎪ 
⎪
⎨ xi · wi + b > θ
o=
. i=1 (7.1)
⎪
⎪
⎩
0 otherwise

The number of synapses in the brain of an adult is about 100 trillion (.1014 )—
related to a single neuron, it varies between 1 and 200,000. Despite this large
number, it is amazing at first glance that these simple linear operations form the
basis of intelligent behavior.
7.1 Neural Networks 73

InputLayer HiddenLayers OutputLayers

Fig. 7.5 A network of perceptrons

In the single-layer perceptron, there is only a single layer of artificial

neurons, which also represents the output vector. Each neuron is represented by a
neuron function and receives the entire input vector as a parameter.
In Fig. 7.5, the yellow layer shows the neurons that receive the input information
from their environment. Each neuron in this layer is connected to each neuron in the
next layer. The values arriving at the neurons of the red layer, with the appropriate
weights, then give the output signal.
If the outputs are only connected to inputs of a subsequent layer, so that the
information flow is only in one direction, we speak of feed-forward networks. The
following topologies have proven themselves:
• fully connected: Neurons in one layer are connected to all neurons in the
layer immediately following.
• Short cuts: Some neurons are not only connected to all neurons of the next
layer but furthermore to further neurons of the layers after next.
• If there are neurons in the network whose outputs are connected to neurons of
the same layer or a previous layer, it is a recurrent neural network.
• Multi-layer perceptrons require more complex learning rules than single-layer
perceptrons. Backpropagation (feedback) is a possible algorithm for supervised
learning. The extension of these network topologies to include more hidden
layers and introduction of other architectures (e.g., recurrent neural networks),
which are also trained by feedback, is nowadays grouped under the heading
Deep Learning.
The backpropagation algorithm uses supervised learning: the algorithm
is presented with examples of the inputs and outputs that the network is to compute,
and then the error is computed. The backpropagation algorithm starts with random
weights, and the goal is to adjust them to reduce the error until the network has
learned the training data. Feed-forward networks often have one or more hidden
layers of sigmoid neurons followed by an output layer of linear neurons. This
74 7 Machine Learning and Neural Networks

Fig. 7.6 A section of a

neuron network

structure allows the network to learn nonlinear and linear relationships between
input and output vectors. The linear output layer allows the network to obtain values
outside the range of .−1 to .+1 (Fig. 7.6).
For the learning process, the data must be split into two sets: the training
dataset, which is used to calculate the deviations and update the weights, and the
validation dataset , which is used to select the optimal number of iterations.
As the number of iterations increases, the training error decreases, while the error of
the validation dataset first decreases, then reaches a minimum, and finally increases.
If the learning process continues after the validation error reaches a minimum,
this will result in overlearning. Once the learning process is complete, another
dataset (test dataset) is used to validate and confirm the prediction accuracy
(Fig. 7.7).
The difficulty in understanding exactly how changing weights and biases affect
the overall behavior of an artificial neural network was one of the factors that
prevented a wider application of neural networks until the early 2000s. Today,
backpropagation algorithms are applied in many areas of artificial intelligence,
for example, optical character recognition, natural language processing, and image
processing.
This is how you learn as a child; after touching a hot stove top once, you either
keep your distance or use an oven mitt, which you most likely learned from your
parents. After a few tries, animals learn to avoid an electric fence. Pain is considered
unpleasant or harmful in many biological systems, and the neural network of
humans or animals seeks a way out.
There are different types of neural networks adapted to different tasks. One such
type is convolutional neural networks (folding neural networks, CNN).
It is a concept inspired by biological processes in the field of machine learning. It is
primarily applied to machine processing of image or audio data.
7.1 Neural Networks 75

Fig. 7.7 Correction of weights from right to left

Fig. 7.8 The architecture of a deep CNN

Yann LeCun (* 1960) is considered the founder of CNNs. He developed new

methods for machine learning, especially for image processing, recognition of
handwritten text, and single characters.
The arrangement in Fig. 7.8 for image classification shows the structure of a
typical CNN. This network has multiple filter kernels per convolutional
layer, resulting in layers of feature maps that each receive the same input but
extract different features based on different weight matrices. Architecturally, three
major differences can be noted compared to the multilayer perceptron:
• two- or three-dimensional arrangement of neurons,
• shared weights,
• local connectivity,
• convolutional layer.
76 7 Machine Learning and Neural Networks

Typically, the input is in the form of a two- or three-dimensional matrix (e.g.,

the pixels of a grayscale or color image). Accordingly, the neurons are arranged
in the convolutional layer. The success of CNNs can be explained by the compact
representation of the weights to be learned. This is based on the assumption that a
potentially interesting feature, for example, edges, is interesting at every position
of the input signal. While a classical two-layer perceptron with 1000 neurons per
layer requires a total of 2 million weights to process a 32 .× 32 image, a CNN with
two repeating units consisting of a total of 13,000 neurons requires only 160,000
(shared) weights to be learned, most of which are in the posterior (fully-connected
layer).

7.2 AlphaZero Wins at Chess, Go, and Shogi

Google subsidiary DeepMind has developed AlphaZero, a system that learned

chess, Shogi, and Go on its own. After a short time, it played these games better
than the strongest previous programs and better than human players (Fig. 7.9).
Chess, Shogi (the Japanese chess), and Xiangqi (the Chinese chess) all derive
from the Indian Chaturanga and, together with Go, are considered to be the games
that pose a great challenge to humans and computers.
The program learned the games within 24 hours, with a massive amount of
computing power behind it: it used 5000 first-generation Tensor Processing Units
(TPUs) and 64 second-generation TPUs. TPUs are special chips for neural network
computations. The mentioned first generation performs 92 tera operations per
second; no exact information about the second one has been published yet.

Fig. 7.9 Computer versus Go—World Champion

7.2 AlphaZero Wins at Chess, Go, and Shogi 77

The fully trained neural network then plays the game on a single machine with
only four TPUs, which is a reasonably fair comparison with the respective strongest
chess and shogi programs, which had 64 CPU cores available after all. An important
point of view is shown here: the training requires a very large effort, but the
application of the finished net is possible with much less effort.
After the comparison with human players has become irrelevant, only one
interesting question remains: Is there an optimal strategy for these games?
• Can White necessarily win from the initial position, for any of Black’s moves?
One might assume so, since White makes the first move and is always one move
ahead in mobilizing his forces;
• Does Black own a move for any move by White, so that the game ends in a draw?
• Can Black necessarily win from the initial position, for every move by White?
It could be possible, because White may irreversibly weaken himself on the first
move.
The question of an optimal strategy has already been answered by the German
mathematician Georg Cantor (1845–1918). He referred to the following steps:
• Build up all the positions in which White has mated, one after the other.
• One move back. White can mate in one move.
• Black takes back a move, and white again chooses a move in which he can
forcibly mate. This results in an optimal strategy. But it cannot be traced back
to the initial position, because that is of course too many possibilities.
The first ending tablebases—for all four-piece endgames—were built by the end
of the eighties. In the beginning of the nineties, the same task was done for five
pieces. In 2005, six-piece endings were solved in Nalimov Tablebases, which are
now used by many professional chess programs and services.
Experts didn’t expect seven-piece endings to be cracked and catalogued until
after 2015, but Convekta Ltd, namely, programmers Zakharov and Makhnichev,
managed to solve this task in just 6 months using a new algorithm designed
specifically for this purpose and run on the Lomonosov supercomputer based in
the Moscow State University, Lomonosov Tablebases.
As a result, we now have 525 tablebases of the 4 vs. 3 type and 350 tablebases of
the 5 vs. 2 type. The calculations for six pieces playing against a lone king weren’t
done because the outcome is rather obvious.
The total volume of all tablebases is 140 000 gigabytes, which is obviously
too much for personal computers. Lomonosov Tablebases will be accessible online
from the Aquarium interface and on the ChessOK website. All users of ChessOK
Aquarium, Houdini Aquarium, and Chess Assistant products get free access to the
service for the period specified in their product description.
Very interesting theoretical and mathematical questions appear here again. In
all previous chess programs, the moves were selected with evaluation functions
whose components had a chess background. For example, the value of the pieces,
the safety of the king, the occupation of the center, etc. were evaluated. The move
that corresponded to the best value of the evaluation function was selected.
78 7 Machine Learning and Neural Networks

The training of neuronal networks generates the weights .wi , which obviously
guarantee an excellent playing style but where one does not know what the
individual weights mean for chess. The requirement that the results of AI programs
must be explainable cannot be realized. Either you believe the computer and act
according to its suggestion or you leave it alone. The result in many areas will
be a computer faith of which we do not yet know where it will lead. In any case,
one must keep in mind that all the existing knowledge of a field has been used.
New knowledge can easily be added, and the finished systems can in principle be
effortlessly duplicated and made available to a wide range of users.
The algorithm for creating the elements of a database and for proving the
existence of an optimal strategy was published by the mathematician Ernst Zermelo
at a mathematics congress in 1912. Later, it found itself as a special case in
mathematical game theory.
Poker has always been considered a game between people. It differs from other
games in that psychological thinking plays a role here, at least when there are only
humans at the table. In a casino in the American city of Pittsburgh, a computer
program beat four of the best poker players in the world. For nearly three weeks,
four poker professionals each spent many hours a day competing with the software
called “Libratus,” betting chips, analyzing their strategies, and still lost. “It felt like
I was playing against someone who was cheating, like they could see my cards,”
said 28-year-old American-Korean poker player Dong Kim.
Thus, neural networks form a central role in the development of other intelligent
applications. The strategic games are just to try out the techniques. The results are
easily grasped by humans. One can now trust that similar results can be achieved
for other problems and other fields.

7.3 Translation Between Different Languages

German technology company DeepL released DeepL, a machine translation system

that delivers translations of unprecedented quality. The DeepL translator can be
tested at www.DeepL.com/translators.
The programs for DeepL run on a supercomputer in Iceland capable of 5.1
PetaFLOPS (5 100 000 000 000 operations per second). This computer was used
to train neural networks with a collection of multilingual texts. The networks look
at a large number of translations and learn independently how to translate correctly
and choose good phrases. The DeepL translator currently (June 2023) supports 31
languages, which can be combined in any way.
An impressive example: we take a previously written paragraph and enter it into
DeepL in the left window; in the right window, we can then set another language.
German
Die Existenz der maschinellen Übersetzung wird oft entweder übersehen (weil
man sich eigentlich längst daran gewöhnt hat und sie benutzt, ohne darüber
7.3 Translation Between Different Languages 79

nachzudenken) oder etwas zu kritisch gesehen (weil sie angeblich alle Übersetzer
um ihre Arbeit bringen wird). Der weltweite Übersetzungsmarkt verzeichnet jedes
Jahr einen Umsatz von etwa 40 Milliarden Dollar und wird bis 2025 voraussichtlich
auf über 60 Milliarden Dollar anwachsen. Durch die fortschreitende Globalisierung
rücken Regierungen, Unternehmen und sogar Privatpersonen in verschiedenen
Ländern immer enger zusammen, und die ständig steigende Nachfrage nach Über-
setzungen kann nicht mehr von Menschen allein gedeckt werden.
English
The existence of machine translation is often either overlooked (because people
have actually long been used to it and use it without thinking about it) or viewed
somewhat too critically (because it will supposedly deprive all translators of their
jobs). The global translation market records about 40 billion Dollars in revenue
every year and is expected to grow to over 60 billion Dollars by 2021. Governments,
companies, and even private individuals in different countries are growing ever
closer together as a result of ongoing globalization, and the ever-increasing demand
for translations can no longer be met by people alone.
Latvian
Mašintulkošanas pastavešana bieži vien tiek vai nu ignoreta (jo cilveki jau sen pie tas
ir pieraduši un izmanto to, par to nedomajot), vai ari uz to raugas parak kritiski (jo
ta it ka atnems darbu visiem tulkotajiem). Pasaules tulkošanas tirgu katru gadu tiek
registreti aptuveni 40 miljardu dolaru ienemumi, un paredzams, ka lidz 2025. gadam
tie pieaugs lidz vairak neka 80 miljardiem dolaru. Globalizacijas rezultata dažadu
valstu valdibas, uznemumi un pat privatpersonas klust arvien ciešak saistitas, un
arvien pieaugošo pieprasijumu pec tulkojumiem vairs nevar apmierinat tikai cilveki.
Hungarian
A gépi fordítás létezésérol gyakran vagy nem vesznek tudomást (mert az emberek
már régóta hozzászoktak, és gondolkodás nélkül használják), vagy túlságosan
kritikusan szemlélik (mert állítólag minden fordítót megfoszt a munkájától). A
globális fordítási piac évente mintegy 40 milliárd dollár bevételt könyvel el, és
2025-re várhatóan 80 milliárd dollár fölé no. A különbözo országok kormányai,
vállalatai, sot magánszemélyei is egyre közelebb kerülnek egymáshoz a globalizáció
következtében, és a fordítások iránti egyre növekvo igényt már nem lehet egyedül
kielégíteni.
Translations into Japanese and Chinese are also possible (Figs. 7.10 and 7.11).
Of course, there are still problems where improvements may be necessary:
• problems with technical terminology and with terminology consistency,
• problems with homonyms and context recognition.
And if you are still not convinced, you can look for a Chinese or Japanese journal
on the Internet and transfer the text to DeepL’s window with Ctrl + C and Ctrl
+ V. There, it will be translated, and you can easily see if you understand the text.
After doing this several times, one is convinced that such systems already have a
80 7 Machine Learning and Neural Networks

Fig. 7.10 A translation into

Japanese

Fig. 7.11 A translation into Chinese

very good level that far exceeds the level of a normal mortal. You can now (as a
scientist, as a reporter, as someone interested in everyday life in foreign countries)
get information that previously would only have been possible with enormous effort.
Google has also invested a lot in translation programs. There are more than
100 languages offered, including many languages that are very rare, especially
from Africa and from Asia, up to Xhosa, Yoruba, and Zulu. The Yoruba are a
West African people living mainly in southwestern Nigeria. The Xhosa are a South
African people who linguistically belong to the Bantu. The Zulu are a subgroup of
the Bantu with over eleven million people today and are the largest ethnic group in
South Africa.
When travelling to Africa, you can be pretty sure to encounter knowledge of
English, French, Spanish, or Portuguese, depending on the former colonial powers.
You get a very friendly welcome if you make an effort to learn the language of the
local people. You write down on your computer what you want to know and then
present it in the local language. And there are also occasions when you have to rely
on communicating with a local who doesn’t know another language.
• Where is the next opportunity to buy something to eat or drink here?
• Dab tsi yog qhov tuag nyob rau hauv Gelegenheit, thiab yog dab tsi.
Hmong is an ethnic group living in southern China.
Reference 81

• Je! Ni nani aliyepata kufa Gelegenheit, na hii inaweza kusababisha nini?

Suaheli is widespread in East Africa.
For the blind, a whole new world has opened up with automatic translations into
Braille . With the help of computers, texts from many languages can be translated
into Braille and vice versa. So it is quite possible for a blind German to communicate
with a blind Japanese.
Even for people who have completely lost their hearing, the possibilities
are increasingly expanding. There are already smartphone apps offered as
translators for sign language and for written and for spoken texts.
Deep learning methods are currently being used very successfully in many fields
of science, technology, and art. Some examples are presented in the corresponding
application works. One can also find a whole range of companies selling applicable
solutions on the Internet.
You can see here how the situation will change in the future. The number of
people translating routine texts in the context of their professional work or private
life will decrease. Only a few linguists will remain who work on high-quality or
the most difficult texts. However, the number of people who have to find their way
around in a foreign environment will have a much easier time of it. I was recently
able to observe a scene at a German bank counter where a foreigner spoke his
request on a cell phone; there it was presented in writing and translated, and the
bank clerk was able to understand the request and take the appropriate actions and
answer the questions.
It is easy to see that these opportunities also have implications for education.
Learning foreign languages becomes a hobby for the individual.
A whole set of caveats to these learning procedures is that you can have negative
and very negative goals. But this is firstly not new in the history of mankind and
secondly depends completely on which goals are given.

Reference

1. https://studyflix.de/biologie/nervenzelle-2737
Chapter 8
Other Applications

8.1 Databases

Databases represent a frontier area with regard to artificial intelligence. Here, one
uses the large storage possibilities available and the search algorithms available.
A database consists of two parts:
1. the management software, the database management system (DBMS),
2. the set of data to be managed, the database (DB) in the narrower sense,
often also called “database.” The management software internally organizes the
structured storage of the data and controls all read and write access to the
database.
3. A database system offers a database language for querying and managing the
data. SQL (System Query Language) is a database language for defining
data structures in relational databases and for editing (inserting, modifying,
deleting) datasets.
SQL is based on a relational algebra, its syntax is relatively simple and
semantically based on English colloquial language. A joint committee of ISO and
IEC standardizes the language with the participation of national standardization
bodies such as ANSI or DIN. The use of SQL aims to make applications independent
of the database management system used.
A very natural form of databases are, for example, dictionaries. The dictionary
dict.cc [2] offers for German as source language 26 languages on the opposite
side. The corresponding page allows you to select a language and provides the
answer when you enter a German word. The advantage of such databases is obvious.
You save a lot of space, a lot of time, and a lot of money. The databases available at
linguee work closely with DeepL.
The existence of dictionaries on the Internet has a major impact on the business
world. Fewer dictionaries are printed, as the Internet provides much more compre-
hensive dictionaries for many more languages. The Duden and its special editions

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 83

C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_8
84 8 Other Applications

are still considered the norm for spelling. However, sales have also declined, as it is
of course also available on the Internet at https://www.duden.de.
For generations, the Brockhaus Encyclopedia has been synonymous with the
world’s collected knowledge, with lexical competence at the highest level. It is
the basis for work, education, and cultural experience and for discussions in
everyday life. But it has also adapted to technical developments: The “Brockhaus
Enzyklopädie Audiothek” is a novelty: over 70 hours of audio examples on two
audio DVDs—one for PC, one for the DVD player—make audible for the first time
what can only be described in words to a limited extent: sounds of nature, speeches,
pieces of music, animal voices, literary lectures, etc.
The online portal https://brockhaus.de ensured that the entire keyword inventory
was accessible to encyclopedia owners anytime and anywhere free of charge until
the end of 2010. In addition, the content is constantly updated here, so that the
Brockhaus Enzyklopädie is and remains not only the most comprehensive
but also the most up-to-date German-language reference work. The same principle
is followed by Wikipedia on the Internet.
There are many more databases on the Internet that have made many branches of
business more or less superfluous. At www.stampworld.com, you can find a stamp
catalogue for all countries and all times. It was a huge job to produce this relational
database, and it is no easy task to keep it constantly updated [3].
In the simplest case, one proceeds in the following way:
• Given a single object selected from a set of similar objects, we take as an example
a single stamp, which was issued somewhere at some time. All stamps of all
countries and times form the set M.
For stamps, one now defines sets of properties that can be attributed to a stamp.
In Fig. 8.1, there are specified the following items:

Fig. 8.1 The first stamp of

British Guiana
8.1 Databases 85

• ∅—no special designation

.

• date of its issue: first issue July 1, 1850,

• colored paper,
• no watermark,
• no perforation,
• issue number 1,
• value 2 cents
• color black, paper dark pink
• price unused e 200.000 ,
• price used e 150.000,
• a picture of the stamp.
The term “relational database” can be easily explained by these properties. For
each point, there is a basic set. There is a wide range of names. The names
of commemorative days, organizations, politicians, and many others appear. The
numbering of the stamps of a country always begins with 1; the upper limit is
with British Guiana the number 241. If one knows this value, then the stamp is
immediately identified.
The paper can be of different colors; today, white is the standard. Watermarks are
a special characteristic of the paper on which the stamp was printed. They should
serve as protection against counterfeiting. Perforations also play a given role.
In older days, very often, the stamps had no perforation. Lateron stamps normally
have a perforation; the existing perforations are distinguished according to their
fineness.
After that, the issue value of the stamp is given in the respective national
currency. The rest of the points are self-evident. The price, of course, varies within
certain limits and embodies the rarity of the stamp.
An element of the relation is given, if for a stamp in each of the given sets
one element was selected. Although hardly any letters are written anymore, stamp
collecting is still alive; for postal administrations of many countries, it is an
additional source of money.
Such databases are very common—car numbers, account numbers, credit card
numbers, insurance policy numbers, tax numbers, or the entries at the public order
office serve the same purpose of unambiguous characterization of certain objects
and their affiliation.
In particular, the links between different databases are also interesting and
important. In Switzerland, medical information is linked to a person. Any entries
that a doctor makes for a patient are immediately available to all other doctors
dealing with that patient. Thus, a user group is created. This avoids duplicate testing,
conflicting medications, unnecessary pathways, and costs. In recent years, a new
approach has emerged on this basis, called cloud computing.
Cloud computing (computer or data cloud) describes a model that provides
shared computer—resources as a service, for example, in the form of servers, data
storage, or applications, in a cloud on demand—usually via the Internet and device-
independent, in a timely manner and with little effort, and charges according to use.
86 8 Other Applications

The offer and use of these computing resources are defined and usually take place
via an application programming interface (API) or, for users, via a Web site or app.

8.2 Trade over the Internet

With the emergence and development of the Internet, many new interlinked
industries have also emerged, which are heavily dependent on databases but
now combine many other characteristics, advantages and disadvantages. They are
summarized under the name “E-commerce”. The Internet is the basis for extensive
sales operations. Potentially, “online banking” also belongs here, but this has already
become a separate industry. Almost every bank today has an online presence.
E-commerce has developed very strongly over the last 30 years and now domi-
nates many business areas where traditional retailing has been severely impacted. A
well-known example is the “Kindle ebook.” Books offered there are cheaper than in
paper format, and they do not require space in the home. However, large pictures and
maps are usually not very clear, beautiful bindings are also lost, and the profession
of bookbinder is more or less dying out. The best known are probably “Amazon”
and “eBay”, but every industry is included. Many manufacturers of certain products
sell via the Internet but also run retail stores.
Amazon.com, Inc. (Amazon for short) is a listed US online mail order company
with a wide range of products. According to its own statements, Amazon, as the
market leader of commerce on the Internet, has the world’s largest selection of
books, CDs, and videos [4]. Via the integrated sales platform “Marketplace”, private
individuals or other companies can also offer new and used products as part of
online trading. Under its own brand, the company sells the “Amazon Kindle” as
a reading device for electronic books, the “tablet computer” Amazon Fire HD, the
“set-top box Fire TV” as well as the “HDMI stick Fire TV Stick,” and the voice
recognition system “Echo.” Through Amazon Web Services, the company is also
one of the leading service providers for cloud computing. The company generated
sales of $386 billion in 2020. In Germany, Amazon was the top-selling US company
in 2020 with sales of 29.5 billion euros. It is almost inconceivable that this company
was only founded in 1994 and has achieved this level of performance within just
under 30 years.
The special characteristics are market transparency and the absence of pref-
erences. In the creation and use of such systems, many previously unimagined
problems arise. For example, it is of great importance for the sales of a product when
an offer appears on the Internet. Research has shown that a customer searching for
something will only look at three to four pages and then make a decision or abandon
the search and postpone the decision. That’s why advertising on the Internet plays a
crucial role. It enables the relatively precise measurement of a specific advertising
success. However, it has probably often reached a stage where this form becomes
extraordinarily exaggerated. However, it is gradually replacing advertising with
catalogs, advertisements, and brochures.
8.2 Trade over the Internet 87

The main advantages are as follows:

• new communication possibilities with customers,
• opening up new sales channels,
• acquisition of new customers,
• increase in sales,
• efficiency increase,
• automated business processes independent of working hours.
However, e-commerce also has disadvantages that affect costs:
• additional sales or procurement risks,
• decreasing customer value (if personal service is not provided),
• decreasing customer satisfaction (reactance reactions to creation of customer
profiles),
• increased risks of payment transactions (risk of fraud),
• additional effort due to maintenance of a mail-order typical organization (ware-
housing, accounts receivable control, etc.),
• return without cost to the customer.
There are also quite a few smaller companies that also use the Internet to
sell their products. Internet retailers have the advantage that they do not need a
physical salesroom, which is available virtually as a Web site. Some online retailers
therefore also do without their own warehouses, and orders are forwarded directly
to the producer or wholesaler, who takes care of picking and shipping the goods.
Companies that ship goods free of charge and also provide fast delivery have thus
become serious competitors to brick-and-mortar retailers. Even Internet booksellers,
who are subject to fixed book prices in Germany and therefore cannot run discount
campaigns, benefit by saving customers a trip to the bookstore for the same price.
In some cases, stores serve simultaneously as a stationary store and as a pick-up
or return station for merchandise ordered by the customer on the Internet. In this
way, service in the local store is linked with online retailing. A study from October
2015 to January 2018 showed that traditional retail can compete with online retail,
especially in cities.
In the case of digital products such as music files, software, or online books,
the customer can download the product directly after purchase. Physical shipping is
eliminated in that case, and the customer can use his purchase immediately.
Most e-shop systems have the following basic software components:
• store database with product information,
• administration database,
• presentation system,
• recommendation engine for recommending a product,
• payment gateway (handling of the payment process),
• other functionalities (tools),
• Web tracking system.
88 8 Other Applications

Amazon obviously proceeds in two steps: first, products are searched for that
semantically basically match the search query. This excludes the possibility that, for
example, a slipper is suddenly offered, although the customer was actually looking
for motorcycle accessories. In the second step, the products pre-filtered in this way
are then sorted by an algorithm according to various criteria. In this step, the most
important factors are “early reviews,” “product images,” or “availability.”
These factors are measured in the units “click rate” and “conversion frequency.”
The click rate shows how many people visited the product page. The second factor
shows how many people actually bought the product. Before you start searching for
a product, you should do some thinking yourself: what would you ask a saleswoman
in the store when searching for a certain product? Using this method, you can come
up with a few matching keywords to the product relatively quickly and build on
that list. This list can then be entered into the Amazon search box to find out more
matching terms through Amazon’s recommendations.
One can secure a place at the top of the search results through Amazon’s
“sponsored products,” which are based on keywords and auctions. The merchant
with the highest bid gets the coveted spot at the top of Amazon’s product search.
Anyone who has a professional Amazon Seller account has the opportunity to place
such ads.
The goal of Google’s search engine is to provide its users with the best possible
information in a fast and uncomplicated way. For this purpose, billions of Web
pages are searched. Numerous factors are used in this search: the words used in
a search query, the relevance and usefulness of pages, the expertise of sources, and
the location and settings.
It is becoming increasingly clear that what matters most is large amounts of data.
The existing algorithms for classifying data and finding factors relevant to decision-
making are already working very successfully. It’s almost a given that learning
programs will also be used to constantly improve advertising.
And, of course, one must not forget the entertainment industry. Here, YouTube
occupies a top position. YouTube is a video portal founded in 2005 by the US
company YouTube, LLC, a subsidiary of Google LLC since 2006. Users can watch,
rate, comment, and upload video clips themselves on the portal free of charge.
YouTube’s main source of income is advertising. It also earns through monthly
subscription services such as YouTube Premium. In 2019, YouTube generated $15
billion in annual revenue. Most of its revenue is generated by playing commercials.
YouTube offers three forms of accessibility: public, unlisted, or private. Public
videos are the default. These can be viewed by anyone and found in search results.
There is an infinite amount of material available, books, plays, classical music, or
anything your heart desires, and search time is very short.
Netflix, Inc. is a US media company engaged in paid streaming and production
of movies and series. It follows the same principles as Amazon and eBay but has a
more limited business scope [19]. Founded by Reed Hastings and Marc Randolph
in Los Gatos, California, in 1997, the company initially operated as an online video
store, shipping movies on DVD and Blu-ray to its subscribers. In 2007, Netflix
entered the video-on-demand business, making its content available to subscribers
8.2 Trade over the Internet 89

via streaming. By March 2020, the media service had offered nearly 3,000 movie
and TV shows for streaming. By the end of June 2021, Netflix had signed up more
than 200 million paid subscriptions since the company’s founding, including more
than 70 million in the United States and Canada. In 2020, the company’s stock
market value exceeded that of the Walt Disney Company for the first time, at just
under $195 billion.
Netflix categorizes its customers as starters who have watched only two
minutes of a movie or series, watchers who abandon a movie or series episode
after 70%, and completers who have watched movies or series seasons 90% of
the time. The numbers in the three categories are reported back to producers within
the first 7 days and 28 days after the release of a new movie or series season. Metrics
for production success beyond 1 month after release are not collected.
Through content-based and collaborative filtering, Netflix is able to make
recommendations to users about series that are tailored to them. To this end,
multilayered data about user behavior is collected each time Netflix is used. In
addition to the plot, actors, film location, etc., this also includes the exact length
of time spent consuming films.
Every driver appreciates the database hidden in his navigation device. It knows
every city, every there, every street, and every house number in every country. It’s
hard to remember that you used to have to carry a mountain of maps or a car atlas
with you. However, the art of finding your way around maps is slowly being lost.
The company eBay follows the same principles. eBay Inc. is a US company that
operates one of the largest online marketplaces. It was founded on September 3,
1995, by Pierre Omidyar in San José (California) under the name AuctionWeb.
The company also offers its services in Europe, for example, in Germany, Switzer-
land, and Austria. Until July 17, 2015, the payment service PayPal also belonged
to the company. Here, then, the character of an auction is added to Amazon’s
business approach. Anyone can offer a product on eBay for a certain commission,
along with a minimum amount at which you want to sell the object. Customers can
then raise the price, as in an auction. A closing date then ends this auction, and the
highest bid wins.
Over the years, the offer expanded from a marketplace with a flea market-like
character to a platform that also offers new goods from commercial dealers, as
opposed to secondhand goods from private sellers. The dependence on databases
and the Internet is quite obvious, and the intelligence here lies, among other things,
in the business idea.
The tourism industry is also firmly in the hands of the Internet and has very
much marginalized local travel agencies. The American company Tripadvisor
is an example of this. It is a US tourism Web site that provides users with
individual testimonials to help plan vacations [5]. There are reviews and testimonials
on 1.8 million stores, 4.2 million restaurants, 1.1 million hotels, and 830,000
vacation rentals, as well as 730,000 attractions in 135,000 destinations (as of mid-
2016). Offered are testimonials and reviews from travelers; links to reports from
newspapers, magazines, and travel guides; as well as travel forums (electronic
“bulletin boards”).
90 8 Other Applications

China is one of the most attractive e-commerce markets in the world. In 2018,
online retail in China generated sales of around 530 billion euros. The number of
online shoppers has increased more than tenfold in the past ten years. Today, nearly
85% of China’s Internet users already shop online. The company Alibaba is quite
comparable to Amazon.

8.3 Big Data

Big data refers to large volumes of data originating from sectors such as the
Internet and mobile communications, the financial industry, the energy industry,
healthcare, and transport and from sources such as intelligent agents, social media,
credit and customer cards, smart metering systems, assistance devices, surveillance
cameras, and aircraft and vehicles, which are stored, processed, and analyzed
using special solutions. It is about “grid search,” “(inter-)dependency analysis,”
“environmental” and “trend research” as well as “system” and “production control.”
The main concern is the discovery of new knowledge. The worldwide volume of
data has swelled to such an extent that unprecedented opportunities are being opened
up. The networking of data sources is also leading to new kinds of applications, as
well as risks for users and organizations. Here we are only at the beginning of the
development. The far-reaching linkages are illustrated in Fig. 8.2.
The economy promises itself new insights into prospective customers and
customers, their risk potential, and their purchasing behavior and generates personal
profiles. It seeks to optimize production and make it more flexible and to better
bring innovations to markets through forecasting. Science studies climate change
and the occurrence of earthquakes and epidemics, as well as mass phenomena such
as population migrations and traffic congestion. With the help of supercomputers,

Fig. 8.2 All accessible data is stored.

8.3 Big Data 91

it simulates both atomic bomb drops and meteorite flights and their impacts.
Authorities and intelligence agencies use enormous amounts of data to detect
deviations and anomalies that could give criminals and terrorists away, as well as
similarities that allow groups to be identified and narrowed down.
Big data is a challenge for data protection and privacy. Often, there is no consent
from the data subject for the use of the data, and often, they can be identified and
controlled. The linking of information that is in itself unproblematic can lead to
problematic findings, so that one suddenly belongs to the circle of suspects, and
statistics can make someone appear uncreditworthy and risky because one lives in
the wrong neighborhood, uses certain means of transportation, and reads certain
books. Information ethics asks about the moral implications of big data, in terms
of digital paternalism, informational autonomy, and informational justice. Business
and legal ethics are also required. Data protection laws and institutions can go some
way toward preventing excesses and ensuring consumer protection.
The volume of data has assumed immense proportions: The amount of data that
mankind accumulated from the beginning until the year 2002 was already created
within just ten minutes in 2014. According to forecasts, this mountain of data
will continue to grow and double worldwide every two years. The flood of data
results from the increasing digitization of all areas of life. According to the big
data definition, the volumes of data are so extensive that normal software is simply
overwhelmed and the usual technology cannot process the volume efficiently. So
when dealing with mass data, special technical requirements arise for the software
used. Only special frameworks can analyze the data. To do this, the software must
be able to process as many datasets as possible at once and import large amounts of
data as quickly as possible. In addition, the software should make the data volumes
available to the user in real time if possible and, if necessary, also be able to answer
several database queries simultaneously.
Here it is particularly important to be able to process very large amounts of data.
To calculate the correlation between two data sequences, one uses the correlation
coefficient, for example. This is not very difficult mathematically [6].
We assume two datasets .(x1 . . . xn ) and .(y1 . . . yn ). First, we calculate the two
mean values:

1 1
n n
x=
. xi y= yi (8.1)
n n
i=1 i=1

Then calculate the covariance in the following way:

n
i=1 (xi − x)(yi − y)
cov(x,y) =
. . (8.2)
n
Covariance can be used to determine how the relative positions (deviations from
the mean) of paired measurements from two variables relate to each other.
92 8 Other Applications

• If deviations from the mean are predominantly of the same sign, the covariance
becomes positive (if positive deviations in one variable are usually accompanied
by positive deviations in the other variable).
• If deviations from the mean are mostly opposite in sign, the covariance becomes
negative (if positive deviations in one variable are mostly accompanied by
negative deviations in the other variable).
• When deviations vary non-systematically, the covariance is close to zero (when
positive deviations in one variable are sometimes accompanied by negative
deviations and sometimes by positive deviations in the other variable).
Then we calculate the standard deviations

n
i=1 (xi − x)
2
.sx = . (8.3)
n−1

n
i=1 (yi − y)
2
sy = . (8.4)
n−1

Finally, we obtain the product-moment correlation

cov(x, y)
r(x,y) =
. (8.5)
sx · sy

The resulting characteristic value r(x,y) is called Pearson correlation

coefficient.
Like the covariance, it allows an interpretation of the direction of the correla-
tion. In addition, however, it also provides information about the strength of the
correlation. This is because standardization allows the correlation coefficient to
be interpreted uniformly (regardless of which variables are considered). By it, the
correlation coefficient .r(x,y) can take only values between .−1 and +1, where
• the value .−1 indicates a perfectly negative correlation between the variables,
• the value .+1 shows a perfectly positive correlation between the variables,
• the value 0 indicates that there is no relationship between the variables.
In the first two cases, all points lie on one degree, which, in practice, however, is
very rarely the case. For the interpretation of the values between 0 and 1 or 0 and
.−1, one uses the following expression:

• from r = 0.10 or .−0.10 one speaks of a weak correlation,

• from r = 0.30 or .−0.30 is called a medium correlation,
• from r = 0.50 or .−0.50 is a strong correlation.
Very often, however, it happens that the data is not described by numbers but only
characterized by rankings. Here the Pearson’s rank correlation helps.
One uses the formula for .r(x,y) but replaces the values for the .xi with rankings [6].
In the example, the running time over 100 m is shown together with the age.
8.3 Big Data 93

Person Age Age rank Running time in seconds Time rank

1 59 6 14,61 6
2 35 3 11,80 2
3 43 5 14,34 5
4 23 1 13,03 3
5 42 4 14,18 4
6 27 2 11,02 1

It can be seen that the order determined by age differs from the order given by
the running times; three people are not in the place corresponding to their age in
the running times. If we calculate the correlation coefficient, we get r = 0.73, which
means that age determines the place in this run only in about 75% of the cases. This
value can be further refined by using additional data.
Regression analysis provides very extensive possibilities. With all these methods,
one can determine extensive correlations between very different pieces of infor-
mation. For example, Google has filtered out 45 factors for the occurrence of an
epidemic from several million queries.
Overall, statistics and probability theory provide many sophisticated, well-
studied means and methods for evaluating mass data that can provide very accurate
information about relationships and expected events. However, one must be very
careful in interpreting results. There are very many (intentional or unintentional)
misinterpretations.
Big data is used in the most diverse areas of life. Even everyday things that almost
every Internet user knows are based on it. An example of big data evaluation from
the area of online shopping: Anyone who has ever shopped on the Internet will
be familiar with the section “Customers who bought product XY also bought.. . ..”
These recommendations result from the evaluation of millions of purchase data from
other customers.
Other areas to benefit from big data:
1. Medical research: By analyzing mass data, physicians can find the best therapy
solutions for patients.
2. Industry: Thanks to the use of their own machine data, companies can increase
the efficiency of their production.
3. Business: Big bata enables companies to get to know their customers well and
tailor offers to them.
4. Energy: To tailor energy consumption to individual needs, you have to know
those needs. Collected consumption data ensures an appropriate energy supply
in the long term.
5. Marketing: Big data is often used in the marketing sector to determine targets.
Mostly, it is about improving relationships with customers.
6. Crime prevention: Government and state security are also turning to big data; the
mass use of surveillance cameras has reduced petty crime quite substantially.
 94 8 Other Applications

7. Analysis method for finding patterns: It identifies clusters of similarities and then
forms groups of objects that are very similar in terms of certain aspects; unlike
classification, the groups (or clusters) are not predefined and can take different
forms depending on the data analyzed.
8. Association analysis: It uncovers correlations between two or more independent
factors, which, although not directly related, are increasingly found together.
9. Regression analysis: It uncovers relationships between a dependent variable (e.g.,
product sales) and one or more independent variables (e.g., product price or
customer income); it is used, among other things, to make predictions about the
dependent variable (e.g., a sales forecast).
10. Predictive analysis: This is actually a higher-level task aimed at making pre-
dictions about future trends; it works with a variable (predictor) measured for
individuals or larger entities.
The applications related to big data are very broad and sometimes show quite
surprising results [1]. Very often, there is no causal relationship between the
phenomena, but the results are quite surprisingly accurate.
In 2009, the H1N1 virus (responsible for bird and swine flu) was detected in the
United States. A few weeks before this virus hit the headlines, engineers working
at Google published an article in the journal Nature. There, the authors showed
how Google could predict the spread of the virus, not only at the country level but
also at the state level. They looked at correlations between the frequency of certain
searches and the temporal and spatial extent of infections. The software found 45
search terms that were used together in a mathematical model. The predictions made
on this basis and the actual national numbers were in high agreement.
This form of data analysis does not look for causal relationships cause .→
effect; it focuses its attention on the fact that many factors, each in itself, may
contribute a tiny bit to the final facts.
A company called 23andMe has been working in Silicon Valley since 2007. The
company’s name refers to the 23 pairs of chromosomes in a human being. Starting
in December 2007, it offered private individuals an examination of their genetic
information. The offer was initially aimed only at customers from the United States
but was expanded over time to numerous other countries. It is available in Germany,
Austria, and Switzerland, among other countries.
The saliva sample sent in is tested for around 200 genetic diseases and 99
other predispositions. Information on geographical origin is also provided. More
than 960,000 sections of the human genome are examined, which identify single
nucleotide polymorphisms and make up the personal characteristics [15, 19].
With the help of artificial intelligence, doctors could soon be able to more
accurately detect malignant tumors during a colonoscopy and better distinguish
them from harmless growths. Currently, all polyps (growths) are removed and then
examined histopathologically. With the help of artificial intelligence, an evaluation
should be available virtually in real time. Another advantage of artificial intelligence
is that it is capable of learning. This will enable the detection rate of polyps to
continue to improve. Small adenomas in particular are flat and often hidden in the
8.3 Big Data 95

folds of the intestinal wall. They may well be missed even with high-resolution
endoscopy equipment. So the use of AI should further improve the overall quality
of screening. It is also possible that the number of histopathological examinations
required may decrease, leading to a decrease in the overall cost of screening.
With the help of artificial intelligence, doctors could soon be able to more
accurately detect malignant tumors during a colonoscopy and better distinguish
them from harmless growths. Currently, all polyps (growths) are removed and then
examined histopathologically. With the help of artificial intelligence, an evaluation
should be available virtually in real time. Another advantage of artificial intelligence
is that it is capable of learning. This will enable the detection rate of polyps to
continue to improve. Small adenomas in particular are flat and often hidden in the
folds of the intestinal wall. They may well be missed even with high-resolution
endoscopy equipment. So the use of AI should further improve the overall quality
of screening. It is also possible that the number of histopathological examinations
required may decrease, leading to a decrease in the overall cost of screening.
Several tools for data analysis are already available and offered over the
Internet. Hadoop, MapReduce, and Spark are three offerings from the Apache
Software Projects. These, as well as other software solutions, can be used to
deploy big data in an enterprise.
• Hadoop is an open-source software solution for big data. Hadoop’s tools
help distribute the process load. This allows massive datasets to be run on
many separate computing nodes. Instead of transferring a petabyte of data to
a small processing site, Hadoop does the opposite. It thus provides much faster
processing of information.
• MapReduce supports the execution of two functions: compiling and organizing
(mapping) datasets and then refining them into smaller, organized datasets to
respond to tasks or queries.
• Spark is also an open-source project from the Apache Foundation. It is an
ultrafast distributed framework for big data processing and machine learning.
Spark’s processing—engine can be used as follows: as a stand-alone installation
and as a cloud service, in all popular distributed IT systems like Kubernetes or,
Sparks’ predecessor, Apache Hadoop.
Cloud technologies are constantly evolving and leading to ever greater floods
of data. In order to build forward-looking digital solutions, this information must
be processed. For virtual transactions, inventories, and IT infrastructures, a well-
thought-out big data approach with data from numerous sources is therefore
required. Only in this way can big data provide a holistic overview. Sources can
include:
• virtual network protocols,
• security-related events and patterns,
• global network traffic patterns,
• anomaly detection and resolution,
• compliance information,
96 8 Other Applications

• customer behavior and preference tracking,

• geolocation data,
• social channel data for brand sentiment or tracking,
• inventory and shipment tracking.
A good example is the analysis of data generated when using a navigation system.
This provides information on millions of different car trips and can predict to the
minute and to the kilometer where traffic jams will occur, where there are dangers in
winter, where bypasses need to be built, where lanes might need to be widened, etc.
The same applies to the use of streetcars, buses, and railroads. Here, for example,
the frequency of trips can be increased or decreased so that the vehicles are neither
overcrowded nor undercrowded, or the waiting time of passengers is minimized.
Even cautious big data trend analyses assume a continuous reduction of local,
physical infrastructures and an increase in virtual technologies. This creates a
growing dependency on various tools and partners. The use of big data is more likely
to increase than decrease in the future. The way in which companies, organizations,
and their IT experts solve tasks will be based on developments in data and IT
technology. This means that there will always be new solutions with which big data
can be stored, analyzed, and processed.
Big data played a major role in the prevention and control of COVID-19 in China.
Software was used for tracking infected people, monitoring and early warning
in places with many people, detecting virus sources, drug screening, medical
treatment, resource allocation, and production recovery. Investment opportunities
for big data technologies in China are expected to witness a 30% growth during the
forecast period (2021–2026). Big data has tremendous potential to revolutionize the
Chinese economy. China has the largest consumer market in the world, producing
countless goods for export. Big data can provide crucial insights into China’s
consumer market and help Chinese companies engage in important or necessary
economic activities.
Another factor that underscores the importance of big data for Chinese com-
panies is the amount of data China now produces as the world’s second largest
economy. Tencent Holdings, for example, states that its data trove contains
15 times as much information as the world’s largest library. The WeChat app alone,
Tencent’s flagship product, has around 760 million registered users [7].
WeChat is China’s most popular messaging app with a monthly user base of
more than one billion people. It is owned by Tencent, one of Asia’s largest
companies by market capitalization. What started as a messaging service has
evolved into an app that can be used to do everything from making payments
to offering rides and even booking flights. One of the main uses of WeChat is
messaging. Just like WhatsApp, you have a list of conversations that you are
involved in. You can add people in different ways.
When exchanging contact information in China, you often see one person
scanning the other person’s phone. Each WeChat user has a unique barcode, called
QR code. A person can scan the other user’s QR code to add them to WeChat.
You can also use a phone number or ID to add a person and search for people
8.3 Big Data 97

nearby. WeChat is one of the most important communication tools in China. Even
for business matters, people prefer WeChat to email.
There is also a social feature called Moments. Users can upload a series of
pictures or videos, and their friends can view and comment on the post. From large
supermarkets to the smallest street vendors and cabs, you can pay almost anywhere
in China with WeChat if you have a Chinese bank account. There are two ways
to pay for something through the app. First, the store can scan the unique WeChat
barcode, or second, the buyer scans the merchant’s barcode.
When buying something online in China, one has the option to pay with
WeChat Pay. One has to enter a passcode or use a biometric authentication tool
to authorize the transaction. Instant money transfers to WeChat contacts can also be
made using the messaging feature, making it easier to share bills or move money
around China. It is possible to live almost cashless in China and go out all day
without a wallet.
The main competitor of WeChat Pay is Alipay, a company of Alibaba
subsidiary Ant Financial.
WeChat and Alipay are often referred to as Super Apps because everything is
integrated into a single service. Businesses can choose to launch mini-programs—or
apps within WeChat—instead of a stand-alone app. The program allows companies
to send promotional messages directly to users through WeChat and reach the app’s
more than one billion users.
The Chinese Big Data market is in a highly competitive phase. In the last ten
years, several start-ups have emerged and made their mark on the market. For
example, the big data company Terark, founded in 2015, has developed the
Terark Database, which reportedly outperforms similar products from Facebook
and Google. Several such start-ups have challenged the previous dominance of
a few major players and fragmented the market. This has also led to significant
consolidation in this industry as major players seek to gain market share and smaller
companies try to scale up. Increasing competition from the West has also forced
Chinese companies to look for opportunities around the world rather than limiting
themselves to China. Some of the recent developments are as follows:
• April 2019—ByteDance, known as the parent company of apps TikTok and
Toutiao, announced the acquisition of Beijing-based big data company Terark
for an undisclosed sum, allowing the company to expand its customer base to
market leaders such as Alibaba Group. Terark has solutions to compress data in
a much smaller space and with higher search speed.
• January 2019—Alibaba acquires Berlin-based start-up Data Artisans for
$103 million. Data Artisans provides distributed systems and large-scale data
streaming services for enterprises. Data Artisans was founded in 2014 by the
team leading the development of Apache Flink, a widely used open-source
technology for big data processing.
Perhaps the most big data evaluation software offerings are in Japan [8], with
239 systems on the market in the business intelligence category alone.
98 8 Other Applications

In Singapore, there are six companies producing software for big data applica-
tions.
Africa and Latin America are also making great strides in this area. Africa will
establish its first big data hub, which will promote knowledge sharing and the
extraction of information from complex datasets. The hub will enable the continent
to access and analyze up-to-date data related to the Sustainable Development Goals
(SDGs) to make evidence-based decisions, says Oliver Chinganya, director of the
Africa Statistics Centre at the United Nations Economic Commission for Africa
(UNECA).
According to one study, big data has a positive impact on almost all areas of
life, including health, aviation, banking, military intelligence, and space exploration.
The study says big data hosted in Africa could be a focal point for healthcare,
supporting diagnoses, treatment prescriptions, health systems monitoring, and
digital surveillance to track epidemics.
During the Ebola crisis in West Africa, data scientists tapped into phone
companies’ extensive data to visualize the population movements of Ebola patients
and predict the spread of the disease. Detailed maps showing the movements of
Ebola patients were created and used by authorities to find the best places to set up
treatment centers and develop means to limit travel to contain the disease.
In a recently signed memorandum of understanding between Rwanda and
Ethiopia-based UNECA, the data center will be established at the National Institute
of Statistics of Rwanda (NISR). “The establishment of this data hub will enable
Africa not to be left behind in the data revolution,” says Yusuf Murangwa, director
general of NISR. “At the moment, we have problems in Africa with access to big
data volumes.” Chinganya tells SciDev.Net that the center will promote knowledge
sharing on newly developed statistical methods and tools while providing training on
the use of big data and data science for development. Users will include members of
the African regional statistical community, the private sector, academic institutions,
and civil society.
The project is still in the early stages. The center is expected to be operational once the
implementation strategy is finalized by the end of the second quarter of 2020, but that
depends on many factors, including the COVID-19 global pandemic, which is restricting
movement and work patterns.

The Kigali center joins others planned in Brazil, China, and the United Arab
Emirates, according to a report by the Global Working Group on Big Data for
Official Statistics released in December by the United Nations Economic and Social
Council’s Statistical Commission.
The regional center in Rwanda “will support capacity building in Big Data
for official statistics in Africa, in collaboration with international and regional
agencies.” The biggest challenge in the field of data science is the lack of capacity
and appropriate infrastructure to support Big Data initiatives, as well as insufficient
investment. Under the global platform, many data sources will be openly accessible.
Mechanisms will also be developed to enable selective data sharing between the
various nodes.
8.3 Big Data 99

The repository will allow computer systems to share and use information; legal
regulations for access will be developed by the Economic Commission for Africa.
Data will be collected based on the UN Basic Principles for Official Statistics, which
protect civil liberties.
Dubai plans to use new technologies and mass data collection to optimize the
mobility of its citizens. The idea is to identify scenarios that will enable better traffic
flow while making all traffic in the city cleaner. The Roads and Transport Authority
(RTA) is an independent government agency responsible for Dubai’s roads and
traffic. It has recently made several announcements aimed at optimizing traffic flow
and making traffic cleaner throughout the emirate [9].
This extensive project is based on the collection of all kinds of data to transform
public and other transport. By collecting data on public transport, cab traffic, or hotel
occupancy, the “Dubai in Motion” program will be able to create optimized mobility
scenarios and thus make adjustments to improve people’s mobility and develop even
cleaner and more sustainable transport. In the long term, this program will also help
plan future green spaces and infrastructure for soft mobility.
But the RTA isn’t stopping there. The agency also plans to use artificial
intelligence for its bike lanes to count the number of users and monitor compliance
with traffic rules and basic safety regulations. Another upcoming project is a new
generation of smart displays that will provide information to public transportation
users and help them better organize their journeys. Dubai also aims to convert 25%
of its entire transportation system to autonomous modes by 2030, whether on the
ground or in the air! [22]
The corona crisis accelerates the use of artificial intelligence in Russia. First
applications to combat the pandemic are coming to market (Gerit School, Moscow,
4/29/2020).
• Pandemic increases the need for digital technologies. Financial resources for this
could be significantly increased in the wake of the corona crisis. The dramatic
drop in oil prices and the pandemic-induced need to work on the move are
leading to an increased demand for digital products in Russia. The Russian
Digital Ministry is proposing to spend up to an additional 67 billion rubles (804.4
million euros) over and above the planned budget.
• Russia’s companies have already presented the first applications for virus control.
The startup company Promobot from Perm offers service robots that measure
body temperature or sterilize surfaces without contact. Sberbank has developed
similarly adaptive robots that autonomously disinfect rooms with UV irradiation.
• The country’s largest financial institution is one of the pioneers. It initiates
important innovations and at the same time provides money for new projects.
Its partner is the Russian Direct Investment Fund (RDIF), which says it has
commitments of more than US $2 billion from foreign donors looking to invest
in artificial intelligence projects. Together with Yandex, Gazprom Neft, Mail.ru
Group, and mobile operator MTS, Sberbank and RDIF formed an “Alliance for
the Development of Artificial Intelligence” in November 2019.
100 8 Other Applications

• RDIF plans to invest, among other things, in analytical methods that will
make it possible to distinguish between COVID-19 disease and pneumonia.
The parastatal Rostelecom Group has unveiled an image recognition method
that evaluates X-ray images of lungs for symptoms of COVID-19 disease. The
technology is being developed with state-owned holding Rostec. RDIF also
plans to work with the city of Moscow to deploy cameras for contactless fever
measurement in urban areas.
• Russian Internet company Yandex and Nvidia, the US manufacturer of pro-
cessors and chipsets, have been providing Russian companies with software
libraries to work on artificial intelligence since mid-March 2020. Among them
is the Parabricks application for genomic analysis, which is used in COVID-
19 research. Yandex plans to invest at least US $40 million in building cloud
capacity for AI applications.
• That the corona pandemic could accelerate the market penetration of AI solutions
in Russia’s industry is shown by the example of Sibur. The country’s largest
chemical company is cooperating with Italian software provider Brochesia
to introduce augmented reality (AR, computer-assisted augmentation of reality
perception) in its factories. According to Sibur, AI-enhanced data glasses can
be used to monitor production processes more efficiently and perform many
maintenance tasks remotely. This has reduced the business travel activity of
technicians during times of pandemic.
• Moscow will become a pilot region from the summer of 2020. There, applications
will be used to control exit restrictions during the corona quarantine. For this, the
metropolis is relying on its area-wide camera network. On July 1, 2020, a 5-year
experiment started with a flexible legal framework designed to facilitate the use
of AI applications while ensuring the protection of personal data.
• According to the “Roadmap for the development of AI technologies in Russia,”
the share of products related to artificial intelligence in the gross domestic
product will increase noticeably, to 0.8 percent by 2024 and to 3.6 percent by
2030, which would be above the global average. According to research by market
research firm IDC, Russian companies invested around US $139 million in AI in
2019, half of which was in computing technology. The market for such products
is forecast to grow strongly.
• Russia’s government supports the new technology with the “National Strategy
for the Development of Artificial Intelligence by 2030.” It lists as its main goals
prosperity gain, higher quality of life, national security, competitiveness of the
Russian economy, and an international leadership role in artificial intelligence.
According to this strategy, the following main application areas are defined:
• production
• more efficient planning processes, supply chain optimization,
• predictive maintenance of machinery and equipment,
• automation of routine production,
• increased use of robots,
• improvement of logistics management,
8.3 Big Data 101

• improvement of risk management in factories (fewer occupational accidents),

• higher customer satisfaction through individualized products,
• optimization of work and duty schedules and employee training and develop-
ment,
• healthcare,
• improvement of diagnostics and prophylaxis through evaluation of imaging
procedures,
• optimal dosage of medicines,
• containment of pandemic risks,
• automation and greater precision in surgical procedures,
• government policies to support artificial intelligence,
• long-term financial support for research activities,
• equipping research labs with advanced computing technology,
• development of globally accessible program libraries for the application of AI,
• tax incentives for private investment in AI,
• simplified access to big data,
• specialist training,
• adequate payment of specialists,
• attracting international specialists in the field of AI (including bringing home
emigrated Russian experts through financial incentives),
• ensuring international exchange (conferences, stays abroad, simplified visa
regulations for foreign specialists),
• export promotion for Russian AI technologies,
• increased public procurement of AI products,
• grants for the development of high-performance energy-saving processors,
• construction of data processing centers to which Russian AI development studios
will have discounted access.
An interesting effect occurs during the realization of such projects. In the
beginning, considerable human and financial resources have to be allocated for the
development and commissioning of such systems. However, once they have been
implemented, a lot of financial resources flow back, and the work in the respective
area reaches a higher level. Importantly, with smaller users, it is always important
to ensure that there are sufficient funds for basic hardware and software.
Chile was still chosen as an example of the situation in other countries. In
Chile, almost only start-ups are advancing artificial intelligence. It has the best
digital networking in Latin America [15]. Mining, retail, and the financial and
health sectors set the tone. As a start-up hub, Santiago is considered one of Latin
America’s Silicon Valleys in the region. Many industries recognize the need to invest
in artificial intelligence. However, implementation is still in its infancy. This is also
due to deficits in the digital infrastructure and the supply of skilled workers—even
though Chile scores well in a Latin American comparison.
• The Chilean government is planning a virtual healthcare system [10]. In 2015,
it published the Agenda Digital 2020, which includes 63 measures to
advance digitization. Of these, it has implemented 23 so far. In addition to
102 8 Other Applications

projects on connectivity and strengthening competition, six of the 63 projects

relate to the healthcare system. The focus is on the collection of health data
available to various medical service agencies [16].
In October 2018, the Ministry of Health announced the digital
hospital as an additional initiative that aims to relieve the burden on
other health centers in the country: The ministry intends to offer care services
online and collect patients’ data in a cloud. It also plans to invest in monitoring
equipment for chronically ill patients. The virtual hospital is based on machine
learning and provides reports and forecasts in real time.
• The Ministry of Education, under the National Commission for Scientific and
Technological Research, offers CONICYT Funding Opportunities for Technical
Equipment for Research Projects (FONDEQUIP) [17].
• In the February 2018 digital connectivity ranking by the Economist Intelligence
Unit, Chile was in eighth place out of 86 countries, just behind France and the
United Kingdom. Compared to the previous year, the country had moved up
twelve places. The Andean country is still ahead of Germany (twelfth place)
and leads within Latin America. Nevertheless, 44 percent of Chilean households
do not have a permanently installed Internet connection.
Most Chileans give out their tax number without hesitation. In Chile, cashiers
ask for the personal tax number when shopping at the supermarket or pharmacy.
Companies can use this for marketing purposes and to identify target groups.
The customer usually benefits from bonus programs. In a survey conducted by
Ceret (Centro de Estudios de Retail), the Center for Retail Sector Studies at the
University of Chile, with more than 2,000 respondents, 70% said they provide
their tax number at the supermarket.
• In May 2018, the Senate approved a law reform on personal data protection.
This stipulates that sensitive personal data can only be used with consent. It also
addresses the establishment of a data protection agency under the Ministry of
Economy. Accordingly, habitual patterns of citizens count as sensitive data.
Until now, the rule has been that published data does not constitute sensitive
information and therefore may be reused without consent. The reform defines
certain data sources that continue to apply under this rule. However, reuse of all
published data is no longer generally permitted.
• Only 10% of professionals with technology-oriented backgrounds are knowl-
edgeable about computer and network security, according to the Foundation
for Progress (Fundación para el Progreso). One study says universities have
barely trained artificial intelligence specialists, even though the need is immense.
Universities are slowly getting on board: Universidad de Chile, Inacap Technical
University, and Pontificia Universidad Católica in Valparaíso already offer a
graduate course in AI. Research is also not yet well advanced. Most government
programs are based mainly on the application of conventional technologies for
small- and medium-sized enterprises.
The Universidad Católica established a laboratory for AI in the Computer
Science Department in 2018. The team developed a robot that checks inventory
and prices at the supermarket. An application will detect false news in the future.
8.3 Big Data 103

Engineers want to better analyze the mechanisms of intelligent apparatus and

eliminate errors. PwC (PricewaterhouseCoopers) opened the first AI and applied
analytics center in Chile in September 2018 as a new division of the multinational
consulting group.
• Start-ups offer solutions ranging from workforce decisions to fiber-optics.
According to Alejandro Legazcue, chairman of digital architecture at technology
consulting firm everis, 63% of companies in the AI sector are start-ups
founded after 2012. Chile is the country with the most start-ups among OECD
(Organization for Economic Cooperation and Development) countries, and
Santiago is known in the region as a center for start-ups. However, financing
options are often limited, as Chilean banks are reluctant to lend to start-ups.
The Chilean startup DART specializes in ophthalmology and is optimizing
a process for diagnosing diabetic retinopathy, one of the most common causes
of eyesight loss. For this purpose, medical professionals collect examination
results from patients with diabetes mellitus on an Internet platform. An algorithm
determines necessary appointments and tests to filter out patients at risk of
blindness.
• A study by consulting firm IT Hunter reveals that demand for AI experts is
highest in banking and retail. Banks are focusing on digital, personalized services
and machine learning. Using machine learning algorithms improves the success
rate for forecasting by 25 to 50%, according to Leandro Fernández, president of
the AIS group in Chile, which specializes in AI.
For retailers looking to calculate and forecast product demand, consulting
firm Accenture Chile expects a significant increase in the use of smart
technologies. A study by Accenture found that 42% of consumers hope for new
standards in services and communications.
• In mining, automation of processes and maintenance of machinery play a
major role. Industry experts see promise in smart sensors to analyze processes
and digital tools for greener practices. Automated vehicles can minimize risks
to workers. Chairman of FLSmidth’s digital department, Mikael
Lindholm reports great benefits of digital twins of machines to simulate
processes. Different scenarios increase productivity.
Innovative technologies are already being used to develop mining areas with
low mineral content. However, many of the corporations in Chile are conservative
in their approach and are not yet open to digitalization.
• In northeastern Santiago, the affluent municipality of Vitacura introduced secu-
rity robots in October 2018 to monitor the streets and provide assistance to
residents. The robots autonomously travel pre-programmed routes, and residents
can communicate through them with the municipality’s control center in the event
of an emergency. The municipality plans to use the data for safety maps and
further analysis.
In Asia, along with China, Taiwan, South Korea, and Japan are also taking the
front seat in AI research. Among other things, Japan is increasingly focusing on
care robots; this is particularly important in Japan because the Japanese population
104 8 Other Applications

is getting older. As part of a pilot project, South Korea plans to use thousands of
surveillance cameras and facial recognition to track contacts of corona-infected
people, according to Reuters. The test will be conducted in the South Korean city of
Bucheon and is scheduled to begin in January.

8.4 The Role of Simulation

This is an excellent way to obtain large amounts of data [18]. One models a problem
on the computer, where the process has a number of parameters that one can change.
A large set of changes produces quite different positive or negative results. The best
results then yield values for the parameters one can work with [11]. It is remarkable
that the results are available in a short time. There can be several reasons for using
simulations: an investigation on the real system would be too complex, too
expensive, ethically unacceptable, or too dangerous:
• driving simulator (too dangerous in reality),
• flight simulator for pilot training, reenactment of critical scenarios (engine
failure, emergency landing—too dangerous in reality),
• medical simulation for training and education of clinical staff, especially emer-
gency situations or complications,
• simulators in medical training (training on real patients is not ethically justifiable
in some areas),
• a power plant simulator in which mainly the operating crews of nuclear power
plants train the control of incidents up to GAU (too dangerous in reality),
• crash test (too dangerous or too costly in reality),
• simulation of manufacturing plants before a rebuild (multiple rebuilds of the plant
in reality would be too complex and too expensive),
• Simulation models can be modified much easier than the real system.
• The real system does not (yet) exist. Example: Wind tunnel experiments with
aircraft models before the aircraft is manufactured.
• The real system cannot be observed directly.
• The real system works too fast: simulation of circuits.
• The real system works too slowly: simulation of geological processes.
• The real system is not understood or is very complex.
• The real system is understood in its elementary dynamics, but the time evolution
is too complex; respectively, an exact solution of the equation of motion is not
possible. Examples are the three-body problem, the double pendulum, molecular
dynamics, and nonlinear systems.
• Simulations are reproducible.
• Technical simulations, for example, for strength calculation (finite element
method), flow simulation, of factory processes and complex logistic systems, for
virtual commissioning or circuit simulation.
8.4 The Role of Simulation 105

• Scientific computing, with applications in physics, chemistry, biology, and

meteorology,
• Simulations for education and training, for example, business simulation games
or medical simulations,
• Simulation games, for example, flight simulations, racing simulations, or busi-
ness simulations.
Simulation can therefore be very helpful when special skills are to be acquired.
Only when these skills are sufficiently available do you move on to real systems.
Working with them is much more expensive. However, simulation can also be
used very well in preventive protection against disasters. Situations are simulated
in which special system parameters are outside the normal operating mode and
possible protective measures are set up.
Generally understandable should be the simulation of the last ice age [20]. The
cited paper is a Max Planck Institute publication: “The Simulation of Ice Age
Cycles with a Complex Earth System Model,” with authors Uwe Mikolajewicz,
Matthias Gröger, Jochem Marotzke, Schurgers, Guillaume Schurgers, and Vizcaíno,
Miren from the Dept. of Ocean in the Earth System (Marotzke) (Prof. Dr. Jochem
Marotzke) and MPI for Meteorology, Hamburg.
Why ice ages and warm periods follow one another is one of the greatest
unsolved mysteries of Earth system research. Presumably, interactions between the
atmosphere, ocean, ice, and oceanic and terrestrial biosphere contribute decisively.
Here we describe the fundamentals and initial results of what is currently the
most complex Earth system model for simulating long time scales. Interactive
land vegetation amplifies climate changes caused by variations in Earth’s orbital
parameters and thus causes a positive feedback even in a complex Earth system
model.
Probably the most important task of climate research for society is the prediction
of climate changes that are to be expected as a result of anthropogenic emissions of
greenhouse gases. A major problem here is the validation of the models with which
the predictions are made. On the one hand, the existing time series of measured
climate data are too short (with few exceptions shorter than 100 years), and on the
other hand, they do not cover the extent of the expected climate changes. To find
sufficiently large climate signals, one must go back at least to the last Ice Age.
For these times, of course, no measured climate data exist. These must therefore
be reconstructed, from marine sediment cores, isotope distribution in ice cores, or
pollen distributions in limnetic sediments. In addition to the validation aspects of
the climate models, there is the scientifically appealing aspect that the reasons for
the regular alternation between long ice ages and relatively short warm periods are
only very incompletely understood. While there is general consensus that variation
in summer insolation at high latitudes (and thus the amount of potential snowmelt)
plays a significant role, this effect is far too weak to explain the strength of the
climate signal (about 4 .◦ C global mean) alone. Thus, a strong positive feedback in
the Earth system is still needed, but it may also consist of the interaction of several
feedback mechanisms to explain the observed cooling.
106 8 Other Applications

The current Earth system models used for the predictions of anthropogenic
climate change are designed to perform simulations of typically several centuries in
length. However, simulating a transition from a warm period to an ice age requires
about 10,000 years of simulation time, and simulating a complete glacial cycle
requires more than 100,000 years. For these much longer integrations, current Earth
system models are far too computationally intensive to perform the corresponding
experiments on today’s computers. In addition, key components that are essential to
this problem are missing, such as an explicit representation of the ice sheets or the
interaction between carbon dissolved in the deep ocean and marine sediment.
Previous attempts to solve the problem have been carried out with highly
simplified models. For example, the dependence on the geographical longitude is
neglected, i.e., only the so-called zonal mean is represented, or the atmosphere is
extremely simplified. Since probably not all relevant feedback mechanisms have
been identified yet, the simplifications could suppress important processes. We
therefore pursue the approach of developing a coupled Earth system model in which
the individual components should be as realistic as possible. This should capture all
important interactions between the different subcomponents as far as possible.
Our model will consist of the following components: the atmospheric circulation
model ECHAM, the oceanic circulation model LSG, the marine carbon cycle model
HAMOCC, the ice sheet model SICOPOLIS, and the terrestrial dynamic vegetation
model LPJ. The model will incorporate a closed carbon cycle. In particular, this
has the effect that the .CO2 concentration of the atmosphere will be simulated
(predicted) by the model, unlike most complex models where it must be prescribed.
Currently, the model is still in the testing phase. The last interglacial period
(about 125,000 years ago) and the subsequent transition to an ice age are used as
the test simulation. The baseline simulation was performed using only the coupled
atmosphere-ocean model. Earth orbit parameters were varied during integration,
starting with the year 132,000 BCE to 112,000 years BCE.
About 125,000 years ago, a warmer climate in the northern hemisphere compared
to today is documented in many geological climate archives. The eccentricity and
the earth’s slope were slightly increased, and the greatest approach to the sun
took place in the northern summer. Overall, this results in greater solar radiation
during the northern summer. On the northern hemisphere it becomes predominantly
warmer—compared with the present condition—particularly in the Arctic and on
the continents. The warming is strongest in summer and autumn, which also
corresponds to the change in insolation. A major positive feedback mechanism here
is the reduction in snow and sea ice cover. The darker surface absorbs a greater
fraction of the solar radiation striking the surface, which increases warming and
further reduces snow and sea ice cover. This interaction between ice cover and
albedo is an important positive feedback effect in the climate system. Over the
Sahara, the warming of the land surface leads to an intensification of monsoon
precipitation, especially in the Sahel region. A similar effect occurs in southern Asia.
The accompanying denser cloud cover leads to cooling at about 10 .◦ N to 20 .◦ N.
Albedo is the measure of diffuse reflection of solar radiation and is measured on a
8.4 The Role of Simulation 107

scale from 0, corresponding to a black body absorbing all incident radiation, to 1,

corresponding to a body reflecting all incident radiation.
10,000 years later, the insolation conditions are almost opposite. The Earth is
farthest from the Sun in the northern summer, resulting in weak summer insolation.
Geologic data also indicate that this is the time when glaciation begins over North
America and the Ice Age begins. At high northern latitudes, the differences between
seasons are smaller, and summer is much colder. On land surfaces in the northern
hemisphere, summers are typically 5 .◦ C to 6 .◦ C colder on a zonal average. As a
result, snow melts much later and sometimes not at all. The lighter surface reflects
more sunlight, which exacerbates the cooling. The north polar region is typically
4 .◦ C to 5 .◦ C colder throughout the year. The monsoon over Northwest Africa and
South Asia is also much weaker, with lower precipitation over the Sahel and India.
In the simulations described so far, the vegetation was prescribed according
to current conditions. In shorter test simulations with interactive vegetation, clear
effects are apparent. The warmer climate 125,000 years ago results in a significant
northward shift of the vegetation boundary in Eurasia and North America. At high
latitudes, the forest spreads farther north. As a result, the surface becomes darker
overall. This effect is exacerbated by the fact that the forest is much darker than
grassland under similar conditions, even under heavy snow cover. This significantly
enhances the albedo effect and thus warming, which in turn allows the forest
to advance further north. The interactive vegetation thus leads to a significant
amplification of the simulated climate signals. The amplification of the summer
monsoon leads to vegetation formation in the Sahara. Conditional on the darker
surface, the increase in monsoon intensifies significantly. Thus, a strong positive
feedback effect is also present here.
Even 10,000 years later, interactive vegetation causes an amplification of the
climate signal, in this case cooling. The northern forest boundary is shifting
southward in North America and over Eurasia. This makes the surface lighter
overall, further amplifying the cooling. The enhancement of the albedo effect by
vegetation is clearly evident. Between the two extreme experiments, the difference
in land albedo over North America is about 0.15. With prescribed vegetation, this
difference becomes only about half as large in summer and is almost negligible
in winter. With interactive vegetation, the simulated climate signal is about 1.5 .◦ C
to 2 .◦ C larger for summer between 125,000 and 115,000 years ago. Monsoon
precipitation over North Africa and South Asia is weaker before 115,000 years,
and drylands expand.
These first results clearly show that the evolution from a pure climate model
(ocean, sea ice, and atmosphere) to an Earth system model also leads to an
improvement in the simulation of climate signals. Coupling with the still missing
model components (ice sheets, marine carbon) will likely add further positive
feedback mechanisms. Through the ice sheet model, it is in principle possible
to maintain snow accumulation during periods of low summer insolation over
the next periods of high insolation. A sufficiently high ice sheet leads to colder
surface temperatures due to the higher surface area and thus also reduces summer
melting. In addition, by about 115,000 years before present, the atmospheric .CO2
108 8 Other Applications

Vostok Ice Core - Temperature and CO2

Temperature - difference from present CO2

4
290
2
270
0
Temperature °C

250

CO2 ppm
–2

230
–4

–6 210

–8 190

–10 170
–420000 –370000 –320000 –270000 –220000 –170000 –120000 –70000 –20000
Time before present

Fig. 8.3 Waves of warming and cooling

concentration is also reduced, further enhancing cooling. The reasons for this
reduction are still unclear. We hope to find the answer in the Earth system model.
There is no better way to show the usefulness of simulation models.
The curves and surfaces shown in Fig. 8.3 clearly indicate a fractal structure. The
same shapes are always repeated; sometimes, they are larger, sometimes smaller,
but they do not change.
Global warming or climate change is at present an extraordinarily hotly discussed
subject. Therefore, a second model is to be presented here, which goes back even
3 million years. This work [21] was published by Matteo Willeit, paleoclimatologist
at the Potsdam Institute for Climate Impact Research in Spectrum.de SciLogs under
the title “3 million years of climate change in computer simulation”.
The Quaternary is the youngest geological period; it covers the last 2.6 million
years. It is characterized by massive climate fluctuations, the ice age cycles, the
dramatic alternation of growth, and decay of giant continental ice sheets in the
northern hemisphere. Quaternary climate fluctuations are best identified by oxygen
isotopes measured in deep-sea sediment cores, which reflect variations in global
ice volume and ocean temperature. These data clearly show that there has been a
general trend toward larger ice sheets and cooler temperatures over the past 3 million
years, accompanied by an increase in the amplitude of fluctuations between ice ages
(glacials) and interglacials (interglacials), as well as a transition about a million
years ago from mostly symmetric cycles with a periodicity of 40,000 years to
strongly asymmetric 100,000-year cycles. However, the causes of these transitions
in the dynamics of the glacial cycles remain controversial among experts.
8.4 The Role of Simulation 109

Among other things, the role of .CO2 changes in Quaternary climate dynamics
is not fully understood, largely because of imprecise data of atmospheric .CO2
concentrations for the period before 800,000 years ago, the time span for which
we have high-quality ice core data.
In a recent study, we were able to reproduce natural climate variability through-
out the Quaternary using an Earth system model of intermediate complexity. In
addition to ocean and atmosphere, the model includes interactive ice sheets for
the Northern Hemisphere and a fully coupled global carbon cycle. The model was
driven only by changes in Earth’s orbital configuration (i.e., by the well-known
Milanković cycles), as well as by several scenarios for slowly changing boundary
conditions, namely, .CO2 outgassing from volcanoes as a geological .CO2 source
and changes in sediment distribution on continents.
These model simulations provide a consistent reconstruction of .CO2 , climate,
and ice sheets that is matched to available observational data. The fact that the model
can reproduce the main features of climate history, including data on Earth’s ice
mass and surface temperatures, gives us confidence in our general understanding of
how the climate system works and provides estimates for the contribution of external
forcing and internal feedbacks to climate variability.
Our results show a strong sensitivity of the Earth system to relatively small
variations in atmospheric .CO2 . A gradual decrease in .CO2 to values below
.≈350 ppm led to the onset of glaciation of Greenland and, more generally, the

northern hemisphere at the beginning of the Quaternary. The growth and melting
of the ice sheets subsequently led to the gradual erosion of the thick layer of loose
sediments that had previously formed on the continents through the undisturbed
action of weathering over millions of years. The erosion of this sedimentary layer—
it was essentially pushed away by the flowing glaciers—influenced the development
of the ice age cycles in several ways. First, ice sheets lying on soft sediments
are generally more mobile than ice sheets on hard bedrock because the ice slides
over the sediments more easily than over the bedrock. In addition, the transport
of sediments to the edges of the ice sheet generates significant amounts of dust,
which, once deposited on the ice sheet surface, enhances ice sheet melting as the
ice becomes dirtier and thus darker. Our results show that the gradual increase in
the area of exposed bedrock over time led to more stable ice sheets that were less
responsive to Earth’s orbital cycles, eventually paving the way for the transition to
100,000-year cycles about a million years ago.
The evolution of ice sheets is very sensitive to temperature, and the onset of
northern hemisphere glaciation about 3 million years ago would not have been
possible in the model if global temperature in the early Quaternary had been higher
than 2 .◦ C relative to pre-industrial times. Because the model has been shown to
correctly predict sea-level variations over the past 400,000 years and also the spatial
distribution of ice sheets at the peak of the last Ice Age, we are confident that the
sensitivity of ice sheets to climate is realistic in the model.
One can see in Fig. 8.3 very clearly that warming and cooling are periodic
functions. The present struggle to reduce .CO2 emissions makes cities cleaner, which
is very desirable, but a significant impact on global warming is not visible. It is
110 8 Other Applications

impossible to prove this, but simulation of these processes does allow fairly reliable
conclusions to be drawn.
Of course, the problem of pandemic simulation has also been turned to very
quickly [12]. Pandemic simulations became very popular in the 2000s. Biosafety
specialists and public health experts wanted to stress test the public health system
and look at where problems were occurring. In running the simulation, scientists,
business people, and government officials sit together and must make decisions in
real time to deal with a widening crisis presented to them in the style of television
news. Some striking examples:
• 2001: As part of Dark Winter, researchers simulate a smallpox attack on the
United States—a few months before a series of anthrax attacks in the United
States.
• 2003: Outbreak of severe acute respiratory syndrome (Sars) in Asia. Triggered
by a coronavirus, SARS spreads to a dozen countries.
• 2005: WHO revises its International Health Regulations (IHR). Countries pledge
to better monitor and promptly report disease outbreaks.
• 2009: H1N1 swine flu emerges in the United States.
• 2014: An Ebola outbreak is reported in West Africa.
• 2015: Zika virus spreads in Brazil.
• 2017: A pandemic simulation is conducted as part of the World Economic
Summit in Davos.
• 2018: Ebola breaks out twice independently in the Democratic Republic of
Congo.
• 2019: Event 201, a simulation of a novel coronavirus pandemic, takes place in
New York City.
• 2020: The SARS-CoV-2 pandemic kills 670,000 people in the first half of 2020
alone.
Efforts are also underway in Germany to create simulation programs for pan-
demics. The “Chair of Innovation Economics” at the University of Hohenheim
presents such a program [13] entitled “Agent-based Model of Health Policy
Measures in the Corona Epidemic,” with the following text:
Exit barriers and social distancing or contamination? How is the course of the
Corona pandemic affected by policy actions? Test for yourself in an online policy
lab what impact the various tools have on the course of the epidemic. What mortality
rates can be expected, and what will be the time course of the crisis? Please note:
Our policy lab does not provide forecasts, but serves to strengthen the understanding
of complex interrelationships.
The corona pandemic is unsettling us all. For many days, we have been following
the publications of Johns Hopkins University and the Robert Koch Institute with
the latest numbers of people infected with the coronavirus. Worldwide, politicians
are reacting with different instruments: While in most countries slowing down the
infection rates is declared the top goal, elsewhere, people are also discussing a
8.4 The Role of Simulation 111

possible strategy of contagion to establish a broad immunization of the population

as soon as possible. However, in view of the inevitable overloading of the healthcare
system that this would entail and the sharp rise in mortality rates, the latter strategy is
not a convincing way forward. But the slowdown strategy is not without controversy
either, as it is unclear how long measures that severely disrupt social and economic
life will have to be implemented.
Two of the key terms we keep hearing in this debate are exponential growth in
the number of infected people and critical thresholds that could be crossed in the
hospital system, risking a total breakdown in service delivery. These are the typical
ingredients of complex systems that are difficult to predict and good for unexpected
surprises.
With the help of modern computer simulations, one can describe the development
of complex systems well and develop an intuition for their seemingly unpredictable
development paths. We have developed an easy-to-use policy lab for you to consider
possible development paths. You see a typical European city with its various
activities (workplaces, supermarkets, schools, etc.) as well as residential areas. The
inhabitants of the city lead a quite normal life, which can be described quite well
with a calendar. In the morning, adults go to work in their offices and factories, and
children go to school, and in the afternoon, they go to sports. In the shopping malls,
business also picks up in the afternoon hours. In all places, numerous encounters
and social interactions take place. For a virus like the highly contagious coronavirus,
these are ideal spreading conditions. You can easily observe on the monitor how a
large percentage of people gradually become infected and, in some cases, become
seriously ill or even die. Hospital capacity in our city is limited, so deaths increase as
the capacity limit is exceeded. The virus disappears after a certain time even without
intervention, and the surviving city residents have developed immunity. There are
many deaths.
You can shape health policy in this city and watch the resulting changes in the
situation on the screen. For example, you can send the infected people to domestic
quarantine, improve the general hygiene conditions with health education, or close
the schools. You can also invest in hospital bed capacity. All measures have an
impact on the number of severe cases and deaths, as well as the length and course
of the epidemic. See for yourself in our policy lab what big differences can be made
by the different measures, and make your own, then better informed, conclusions.
It is important to point out that the policy lab focuses on social interactions.
Epidemiological and medical interactions are modeled very simply and make use
of publicly available sources of knowledge. This results in many opportunities for
interdisciplinary collaboration. The model offers many possibilities for expansion.
We are working on it. On the next Web page, one can then work directly with the
simulation model [14]. It is extremely interesting to change the different parameters
of the model and observe how these changes affect it.
112 8 Other Applications

References

1. Viktor Mayer-Schönberger, Kenneth Cukier. Big Data: The Essential Guide to Work, Life and
Learning in the Age of Insight (English Edition), John Murray (Publishers), 2017, 978-1-848-
54793-3
2. https://browse.dict.cc
3. https://www.stampworld.com/de
4. https://de.wikipedia.org/wiki/Amazon/Geschichte
5. https://de.wikipedia.org/wiki/Tripadvisor
6. https://statistikgrundlagen.de/ebook/chapter/korrelation
7. https://www.cnbc.com/2019/02/04/what-is-wechat-china-biggest-messaging-app.html
8. https://sourceforge.net/software/big-data/japan
9. https://www.thestar.com.my/tech/tech-news/2021/10/29/how-big-data-is-helping-make-
transport-cleaner-in-Dubai
10. https://www.gtai.de/gtai-de/trade/specials/special/chile/in-chile-bringen-fast-nur-start-ups-
die-kuenstliche-intelligenz-22598
11. https://www.softguide.de/software/simulation
12. https://www.spektrum.de/news/covid-19-pandemiesimulationen-konnten-uns-nicht-
vorbereiten/1761094
13. https://inno.uni-hohenheim.de/corona
14. https://inno.uni-hohenheim.de/corona-modell
15. https://www.gtai.de/gtai-de/trade/specials/special/chile/in-chile
16. http://www.agendadigital.gob.cl/
17. https://www.conicyt.cl/fondequip/sobre-fondequip/que-es-fondequip
18. https://www.softguide.de/software/simulation
19. https://de.wikipedia.org/wiki/Netflix
20. https://www.mpg.de/847908/forschungsSchwerpunkt
21. https://scilogs.spektrum.de/klimalounge/3-millionen-jahre-klimawandel-in-der-
computersimulation
22. https://teletimesinternational.com/2021/rta-digital-dubai-dubai-in-motion-gitex-2021
Chapter 9
Digitization

It is currently possible to capture all information that exists in digital form. A digital
camera is a camera that uses a digital storage medium as the recording medium
instead of a film; the image is previously digitized by means of an electronic image
converter. Its price ranges from 100 to 2000 euros. The first camera was designed by
Steven J. Sasson, a US engineer, at Kodak in 1975. In August 1981, Sony CEO Akio
Morita introduced the first filmless magnetic camera under the name Mavica and
expanded the product range in the following years. Figure 9.1 shows an example.
If the information is in analog form, it can also be digitized using suitable tools.
Perhaps the best known is the possibility of scanning written documents. It can be
assumed that any information is available in digital form. The possibility of further
processing on computer systems is a principle common to all manifestations of the
Digital Revolution and the digital transformation in economic,
social, working, and private life. In fact, technically networked digital commu-
nication significantly increases the diversity of technical-organizational solution
options. It therefore does not create structures that are stable in the long term, but
increases their flexibility and complexity and reduces their predictability through
the processes of disruptive change it triggers.

9.1 The Basics of Digitization

For mass storage and processing, increasingly powerful storage media have existed
since the 1960s and microprocessors since the 1970s. It is estimated that by 2007,
94 percent of the world’s technological information capacity was digital (up from
just 3 percent in 1993).
The quantity to be digitized can be anything that can be measured by sensors.
Typical examples are as follows:

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 113
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_9
114 9 Digitization

Fig. 9.1 A Sony digital

camera

• The sound pressure when recording sound with a microphone

• The brightness in image and video recording with an image sensor
• The texts of a scanned document
• The forces where the weight or the acceleration caused is measured
• The temperature
• The magnetic or electric field strength
A sensor measures the physical quantity and passes it on in the form of an
electrical voltage or current. This measured value is then converted into a digital
value using an analog-to-digital converter. This process can be done once or at
regular time intervals. From here on, the measured values are digitized and can be
further processed or stored by a computer system.
How the digitized values are subsequently represented internally in the system
depends on the system in question. Here, a distinction must first be made between
memory-independent coding and then the storage of information blocks. The coding
and the format depend on the kind of the information, the used programs, and
also the later use. The storage can take place in the volatile working memory or
permanently, for example, in database systems or directly as individual file, in a file
system.
Here, file formats that standardize both binary encoding and metadata are
essential. Examples are text files in ASCII or the Unicode encoding,
image formats, or formats for vector graphics, which, for example, describe
the coordinates of a curve within a surface or a space (Table 9.1).
With regard to process digitization, interfaces between the digital world and the
outside world are crucial. Digital information is output to analog devices or attached
to physical goods so that it can be time-shifted by humans or the same machine or
read again by other machines. This includes not only classical techniques such as
the output of digital information on carrier materials such as paper by means of
human-readable characters (and their reconversion by text recognition), but also
specialized techniques such as barcodes, 2D codes (e.g., the QR code) or
radio networks, which are used in the Internet of Things even without visual contact
and without electrical connection for communication between devices (e.g., via
9.1 The Basics of Digitization 115

Table 9.1 Examples for the use of the Unicode

Unicode Symbol Meaning
03E0 Greek letter sampi
0429 Cyrillic capital letter shcha
04BC Cyrillic capital letter Abkhazian che
05D8 Hebrew letter tet
060F Arabisches Zeichen misra
0F16 Tibetischer Schriftzug lhag rtags
1254 Äthiopische Silbe qhee
14D9 Kanadische Silben y-cree loo
189A Mongolischer Buchstabe manchu ali gali gha
4F86 Chinesisch-japanisch-koreanische Einheitsgrafik
A98B Javanesischer Buchstabe nga lelet raswadi

wireless local area networks (WLAN) or with radio frequency

identification (RFID)).
Digital twins can be modeled from real objects or processes, with which virtual
simulations can be carried out without affecting reality.
• Digital data allow the use, editing, distribution, indexing, and reproduction in
electronic data processing systems.
• Digital data can be processed, distributed, and duplicated by machine and thus
more quickly.
• They can be searched.
• The space required is significantly less than for other forms of archiving.
• Even with long transport routes and after multiple processing, errors and distor-
tions (e.g., noise overlays) are low or can be completely eliminated compared
to analog processing. Another reason for digitizing analog content is long-term
archiving. Assuming that there is no such thing as a data carrier that lasts forever,
constant migration is necessary. It is also a fact that analog content loses quality
with each copying process. Digital content, on the other hand, consists of discrete
values that are either readable and thus equivalent to the digital original or no
longer readable, which is prevented by redundant storage of the content or error
correction algorithms. Finally, analog originals can be preserved by creating
digital user copies. This is because many data carriers, including records, analog
feature films, and color slides, lose quality simply through playback or even
simple aging processes. Printed books or newspapers and archival materials also
suffer from use and can be preserved through digitization.
116 9 Digitization

9.2 Some Applications

To digitally represent a color image, a higher quantization is required. For digitized

images in the RGB color space, each color value of a pixel is decomposed into the
values red, green and blue, and these are stored individually with the same
quantization (maximum one byte/color value = 24 bits/pixel).

Today is Friday. That’s followed by the weekend.

.

Today is Friday. That’s followed by the weekend.

Today is Friday. That’s followed by the weekend.

All word processors provide the ability to color the background as well as the
text.
The digitization of audio data is often referred to as “sampling.” Sound waves
from a microphone, previously transformed into analog electronic oscillations, are
randomly measured and stored in rapid succession as digital values. Conversely,
these values can also be played back in quick succession and assembled into an
analog sound wave that can then be made audible again. The measured values
would actually result in an angular waveform during the reconversion: the lower the
sampling frequency, the more angular the waveform or signal. This can be reduced
by mathematical methods. In sampling, the bitdepth denotes the “space” for
values in bits, which are necessary, among other things, for the resolution of the
dynamic range. From a sampling frequency of 44.1 kilohertz and a resolution of 16
bits, one speaks of CD quality.
Records can be read and digitized contact-free with software support by scanning
a high-resolution optical image of the sound carrier by a program. This process
is used in the reconstruction of historical sound recordings. So you can take your
records, which you have collected over many years, to the flea market or give them
away without any problems if you have digitized them beforehand.
Digitization of archaeological objects is the digital recording of archaeological
objects in writing and images. All available information (classification, dating,
dimensions, properties, etc.) about an archaeological object (vessel, stone tool,
sword) is digitally recorded, supplemented by electronic images and drawings and
stored in a database. Subsequently, the objects can be integrated in the form of a
data import into an object portal such as museum-digital, where the objects can
be freely researched by anyone (Fig. 9.2). The reason for digitizing archaeological
objects is usually the recording of larger holdings such as archaeological collections
at museums or the offices responsible for the preservation of archaeological
monuments in order to present them to the public. Since in everyday museum life
it is never possible to show all the objects in a collection in the form of exhibitions
or publications, digitization is a way of nevertheless presenting the objects to the
general public and also to the scientific world. In addition, an electronic preservation
9.2 Some Applications 117

Fig. 9.2 A museum for digital art in Tokyo

Fig. 9.3 Gravitational waves

of the collection is carried out, an aspect that is not insignificant in view of the
collapse of the historical archive of the city of Cologne.
Already in 1916, Albert Einstein postulated “gravitational waves.” These dis-
tortions of spacetime arise as a direct consequence of his “General Theory of
Relativity”. On September 14, 2015, the first direct detection of gravitational waves
was achieved in the USA with the Advanced LIGO detectors. The measured signal
came from two black holes orbiting each other, getting closer and closer, and
eventually merging.
On August 17, 2017, astronomers received both electromagnetic radiation and
gravitational waves from an event for the first time: in the galaxy NGC 4993, 130
million light-years away, two neutron stars had collided and merged. This marked
the beginning of a new era of gravitational wave astronomy. The result of their
digitization is shown in Fig. 9.3.
118 9 Digitization

In healthcare, digital applications from telemedicine offer new opportunities

to increase the effectiveness and efficiency of examinations and treatments, improve
patient care, and increase the transparency of service and value-added processes.
The aim is to make medical knowledge and therapeutic options more widely and
easily available through intelligent electronic data use and to relieve doctors, nurses,
caregivers, and other service providers of administrative and routine tasks, thus
significantly improving the quality of healthcare in rural areas as well.
The basis of digitization is the insured person’s medical data, which is exchanged
between doctors and patients and between the individual service providers, thanks
to modern information and communication technologies. Digitization makes it
possible to modernize and significantly develop diagnostics and treatment. It helps
to facilitate communication between the individual players in the healthcare system
and enables individual patients to take greater control of their health, for example,
through apps and information on the Internet. These issues illustrate just how
expansive the transformation to which digitalization in medicine is subject is. The
respective role of the patient, the relationship with those treating him, and the
healthcare system as a whole are also subject to this influence. Physicians will
feel how their professional image and also their self-image are adapting to the
new circumstances. When it comes to digitization, the German healthcare system is
far behind in an international comparison. In other countries, telemonitoring, video
consultations, and electronic patient files have long been the order of the day—in
Germany, many of these innovations are not used or are used only in part [1].
The digitization of production technology includes “design and code creation
processes” (CAD, CAM), “manufacturing processes” (e.g., using CNC machines
or 3D printing), and assembly processes (e.g., using industrial robots). Increasing
networking requires the design of common standards so that increasingly complex
production systems can be controlled.
Digitally controlled warehouse technology, navigation systems, and digital traffic
guidance systems represent special branches of technical development.
Digitization in agriculture has been advancing since the advent of the personal
computer. While it was initially accounting and field documentation in the farm
office that could be done in a time-saving manner by means of agricultural
software, since the 1990s, various development thrusts such as precision
farming, smart farming, and most recently digital farming have
caused computer and sensor technology to become widespread in current agricul-
tural machinery. Autonomous vehicles, tractors, and field robots also now exist in
agriculture, and not just as prototypes (Fig. 9.4).
The agricultural community has spent several years creating a farm description
for digital agriculture. Agriculture is one of the most complex systems that can
be analyzed, and most of what has been proposed are solutions such as weather
and image data. Digital farming is the application of precision location methods
and decision-quality agronomic information to illuminate, predict, and influence
the full range of cropping problems on the farm. Here’s a look at each part of
the definition: Precision is about geolocation services associated with the Global
Positioning System (GPS) and its extensions. It is the overlay of these
9.2 Some Applications 119

Fig. 9.4 The job profile is

also changing in agriculture

geolocation services on a digital map for precision sensing, identification, predictive

decision-making, and action.
Information of high decision quality is provided in a timely manner and within
the decision loop of the event. It is provided by advanced sensors, descriptive
models, and predictive algorithms that provide the required insight into the agro-
nomic problem. Cultivation is an end-to-end, continuous, real-time decision-making
process that requires timely decisions and actions. It spans from seeding to harvest.
Today, cultivation is about the knowledgeable but subjective observational skills of
the individual. In the future, it will be about the objective, predictive capabilities
and precision of the digital ecosystem. Digital agriculture must also be a system
that works. For this reason, the following requirements ensure that the system is
scalable to millions of acres, can be used for multiple crops, provides an end-to-end
solution, exists in an ecosystem, and supports the diverse agronomic and economic
needs of hundreds or even thousands of stakeholders simultaneously.
Digital agriculture at the next level down must organize, analyze, and orchestrate
the timely delivery of information from the data assets that make up a field. It
must be about dividing the field into differentiable, geographically located, and
individually homogeneous units of production assets. This requirement is about
each unit as a productive asset. Precise localization ensures that the information
collected about that unit is measured, recorded, analyzed, and processed in the same
place and can be differentiated from all other surrounding assets. The point is to be
individually homogeneous and identical in size, extent, and depth so that the system
can analyze the same entity in a repeatable manner. This means that each variable
for a productive unit has the same variable value across the footprint and depth.
Forecasts and recipes can be created for each means of production, and produc-
tion is monitored individually. Each productive unit of the plant can be identified and
analyzed separately and precisely. From this, measures are derived that are taken to
achieve a predicted result. The sum of the results for all these homogeneous units
120 9 Digitization

Fig. 9.5 The future of

education

forms a field, and the sum of all predictions, regulations, costs, and yields is the
economics of the field. By viewing the field as a sum of productive assets, digital
agriculture can apply advanced data and analytics algorithms in real time to the
management of each asset and thus to the field as a whole.
Digital agriculture is not about genetics, weather forecasting, etc. While these
factors are extremely important for predictions, they are generated externally and
used as inputs. For example, seed selection would be tested based on seed profiles
developed and produced by agribusinesses. Soil conditions, hydration, and other
known information would be obtained from sensors or from historical records for
the region and field. The region and field data are historical and measured, and these
data are used as the basis for forecasting. Digital agriculture is about using known
field, crop, nutrient, protection, and hydration data to forecast
Digitization in education is very extensive. The use of calculators, cell phones,
tablets, and computers is part of everyday life, and the use of computers and
mathematics software has already been presented in an earlier chapter. Distance
learning became important since the COVID-19 pandemic and was part of everyday
life in most educational institutions for quite some time (Fig. 9.5).
The fundamental advantages of digitization lie in the speed and universality of
information dissemination. Due to low-cost hardware and software for digitization
and the ever-increasing networking via the Internet, new application possibilities
in business, administration, and everyday life are emerging at a rapid pace. When
the logic of production and business models, value chains, economic sectors,
administrative routines, consumption patterns, or even everyday interaction and
the culture of a society are profoundly changed as a result, we speak of digital
transformation. This entails opportunities, but also risks.
Digitization places new demands on the legal system, although legal science only
began to address this problem a few years ago. The theory of fuzzy law assumes that
law as a whole is fundamentally changing in a digitized environment. According
to it, the importance of law as a means of controlling society is significantly
Reference 121

relativized, since society’s claims are additionally oriented toward intangible goods
that transcend national boundaries.
Advances in digital techniques mean that better and better copies, reproductions,
and even counterfeits can be produced. Copyright infringement and counterfeiting
can lead to significant economic damage. In the early 1990s, for example, an
internationally active counterfeiting ring was discovered in New York that had
produced fake works of art with a retail value of $ 1.8 billion. The crisis in the
music industry that began in 1997 due to unauthorized music downloads from the
Internet, which can be burned onto blank CDs and thus also distributed illegally,
coincided with the same period. The number of blank CDs in Germany rose from
58 million to 303 million between 1999 and 2004, while the number of recorded
original CDs fell from 210 million to 146 million during the same period.
New types of copyright problems are raised by meta-media such as search
engines, which select content (texts, images, etc.) from primary media, grab it,
recombine it, and distribute it. A class action lawsuit has been filed against Google
Books by US publishers and authors, but criticism of the Google corporation’s
behavior has also come from European countries. Media report a “creeping decline
of the book industry.”
Another copyright problem can be found in 3D printing: with the creation of a
3D-printed template (usually in the form of a digital template or CAD design), a
work exists in the sense of copyright law and is protected. The production of such a
template by a 3D print constitutes a reproduction. Unlawful sale by third parties (or
without the consent of the author) may constitute a violation of copyright law.
The Digital Revolution has enabled great advances in science in a wide variety
of fields:
• Successes in genome decoding
• Predictions of climate research
• Complex models in physics and chemistry
• Nanotechnology
• Neurophysiological bases of language development and cognitive functions
• Economic simulations
• Comparative studies in linguistics and literature
Actually, every scientific discipline has benefited from developments in computer
technology.

Reference

1. https://4919758967.naturavitalis.de/Mediathek/Magazine/Wie-viel-Digitalisierung-vertraegt-
das-Gesundheitswesen.html
Chapter 10
Image Processing

10.1 Predicting Earthquakes

We start with a very interesting example: “Thanks to Artificial Intelligence to

Improved Earthquake Analysis”[2]. Worldwide, there are a few hundred earth-
quakes every day, and more than a million every year. Most are so weak, however,
with a magnitude of 1 to 2 on the Richter scale, that they can only be detected by
sensitive instruments. Only quakes of magnitude 4 cause noticeable tremors, and
at magnitude 5, there can be damage to buildings. Quakes of this magnitude occur
more than 10,000 times in a year. Even quakes greater than 7 in magnitude happen
more than once a month, and quakes stronger than 8 about once a year. The strongest
earthquake ever measured was the Valdivia quake and occurred in Chile on May 22,
1960. It had a magnitude of 9.5 and triggered a 25 meter high tsunami. Incidentally,
the earth also quakes almost daily in Germany. Between January 1 and May 30,
2019, 114 quakes with a magnitude between 1 and 2.8 were registered.
While quakes in Germany do not cause any damage, they cost people in many
countries not only their property but also their lives. However, it is not always
the main quakes that are the strongest and cause the greatest damage; it is often
the actually weaker aftershocks that have far more catastrophic effects. Yet much
damage could certainly be mitigated and lives saved if these quakes could be
predicted.
At a place threatened by an earthquake, two types of seismic waves arrive: space
waves and surface waves. The waves arriving first are called primary waves
or P-waves. These are followed by the secondary waves or S-waves. Both types
of waves pass through the whole globe and are therefore space waves. Finally, the
surface waves arrive, which are divided into Rayleigh waves and Love
waves.
• P-waves behave in rocks like sound waves in air. However, they are much faster
than sound waves, with a propagation speed of about 5 km per second. P-waves

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 123
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_10
124 10 Image Processing

propagate in solid matter as periodic compression and stretching of matter and

are, just like sound waves, so-called longitudinal waves. The particles of the rock
oscillate back and forth in the propagation direction of the wave.
• With slightly less than half the speed of P-waves, S-waves propagate. Because
in them the rocks oscillate in a plane perpendicular to the direction of propagation
and undergo a bending or shearing motion, they are called transverse or
shear waves. They resemble the transverse motions of light waves. Neither
in liquids nor in gases can they propagate, since shear motion is not possible in
these.
• Some P- and S-waves reach the surface and are reflected there. Rayleigh
waves are formed. They are surface waves and propagate along the Earth’s
surface. For their oscillations, they need a free surface, like the waves on the
sea. Their propagation occurs in ellipses on a vertical plane. They cause the most
damage and are named after the physicist John William Rayleigh (1842–1919).
• The particle motion of the seismic surface waves occurs on a horizontal surface
at right angles to the direction of travel. There is no vertical displacement. Since
they often have large amplitudes, that is, their largest deflections are very large,
they cause severe damage to buildings by horizontal shearing of the subsurface.
• The P-waves arriving first at a certain point on the surface reach this point at
a steep angle and cause vertical ground motions, which, however, do not cause
major destruction. They are followed by the S-waves with a relatively violent
lateral shaking of the ground. Simultaneously with them or shortly after them, the
Love waves arrive. The ground now begins to shake more strongly, at right angles
to the wave propagation. Finally, the Rayleigh waves run in, producing ground
motions both longitudinally and vertically. They cause the much-described
rolling motion of the subsurface in large earthquakes. The sequence of different
surface waves forms the essential and devastating part of an earthquake. Love
and Rayleigh waves last five times longer than P- and S-waves. Each earthquake
decreases in intensity the farther it travels from its starting point, the earthquake
focus. Friction between moving rock particles during an earthquake gradually
converts some of the wave energy into heat. Therefore, rock masses set into
oscillation by seismic waves will come to rest sooner or later, depending on the
amount of wave energy generated in the hearth.
The analogous record in Fig. 10.1 must be digitized so that it can be displayed as an
image and evaluated accordingly.
Scientists at the Geophysical Institute (GPI) of KIT, at the University of
Liverpool, and at the University of Granada have shown that artificial intelligence
can evaluate data as accurately as humans. To do so, they used a convolutional neural
network and trained it with a relatively small dataset on 411 earthquake events in
northern Chile. The CNN then determined the onset times of unknown P-phases
and S-phases at least as accurately as an experienced seismologist doing manual
picking; it was far more accurate than a classical picking algorithm.
10.2 Tumors, Alzheimer’s Disease, and Heart Disease 125

Fig. 10.1 Magnitude 3.6

earthquake in Switzerland

10.2 Tumors, Alzheimer’s Disease, and Heart Disease

Every year in Germany, approximately 4,500 people develop glioma, the most
common and malignant brain tumor in adults. Because this type of tumor is
extremely resistant, chemotherapy or radiation therapy has limited effectiveness,
and the tumors often cannot be completely removed even by surgery. For this reason,
scientists have been researching new treatment approaches at full speed for a long
time.
One of the most important factors in assessing the efficacy of a new therapy
for brain tumors is growth dynamics, which are determined via standard magnetic
resonance imaging (MRI) scans. However, these scans are prone to error and easily
lead to discrepant results when tumor extent is measured manually. “This can nega-
tively affect the assessment of therapy response and subsequently the reproducibility
and precision of scientific statements based on imaging,” explains Martin Bendszus,
medical director of the Department of Neuroradiology at Heidelberg University
Hospital.
A team from Heidelberg University Hospital (UKHD) and the German Can-
cer Research Center (DKFZ) has developed a new, machine-based method for
automated image analysis of brain tumors. For this purpose, the scientists have
designed algorithms and neuronal networks that can reproduce the therapy response
in brain tumors based on MRI images more reliably and precisely than traditional
radiological methods (Fig. 10.2).
Using a reference database of MRI scans from nearly 500 brain tumor patients
at Heidelberg University Hospital, the algorithms learned to automatically detect
and localize the brain tumors. In addition, the algorithms were able to volumetri-
cally measure the individual areas (contrast-enhancing tumor fraction, peritumoral
edema) and precisely assess the response to therapy.
The results of the research project were extensively validated in cooperation
with the European Organisation for Research and Treatment
of Cancer (EORTC). “The evaluation of more than 2,000 MRI examinations
126 10 Image Processing

Fig. 10.2 Magnetic

resonance tomography

of 534 glioblastoma patients from all over Europe shows that our computer-based
approach enables a more reliable assessment of therapy response than would be
possible with the conventional method of manual measurement,” explains Philipp
Kickingereder from the Department of Neuroradiology at Heidelberg University
Hospital. “We were able to improve the reliability of the assessment by 36 percent.
This may be critical for imaging-based assessment of therapy efficacy in clinical
trials. Prediction of overall survival was also more accurate with our new method.”
Here another problem becomes apparent. It is not so easy to obtain larger amounts
of relevant images. Here, many organizations have to work together!
The researchers hope that this technique can soon be used in clinical trials and,
in the future, also in routine clinical practice for standardized and fully automated
assessment of the therapy response of brain tumors. To this end, they designed
and evaluated a software infrastructure so that the technique can be integrated into
existing radiological infrastructure. “With this, we are creating the conditions for
widespread use and fully automated processing and analysis of MRI examinations
of brain tumors within a few minutes,” explains Klaus Maier-Hein from the Medical
Image Processing Division at the German Cancer Research Center.
Currently, the new technology is being re-evaluated at the NCT Heidelberg
as part of a clinical trial to improve the treatment of glioblastoma patients. “For
precision therapies, a standardized and reliable assessment of the effectiveness of
new treatment approaches is of paramount importance. This is where the technology
we have developed can potentially make a decisive contribution,” explains Wolfgang
Wick, head of the Neurological Clinic at Heidelberg University Hospital.
Philipp Kickingereder says that with this work, he and his colleagues have
been able to demonstrate the great potential of artificial neural networks in
radiological diagnostics. Klaus Maier-Hein hopes to further advance the technology
for automated high-throughput analysis of medical image data and to “send it to
10.2 Tumors, Alzheimer’s Disease, and Heart Disease 127

other diseases, such as brain metastases or multiple sclerosis, in addition to its use
in brain tumors.”
Usually, mammography is used to diagnose breast cancer. Radiologists detect
about 80 percent of the tumors present—the remaining 20 percent are overlooked
for various reasons. Tests with DeepMind have shown that even the previously
overlooked tumors can be detected accurately. Comparative tests have shown that
thanks to DeepMind, around 9.4 percent more positive findings could be made in
the USA and as much as 2.7 percent in the UK. The number of false positives has
also been reduced. The number fell by 5.7 percent in the USA and by 1.2 percent in
the UK [6].
Researchers have trained an adaptive artificial intelligence to recognize symp-
toms of Alzheimer’s disease . These signs are visible in special brain scans, but can
hardly be spotted by medical professionals. In the pilot project, the AI proved highly
successful, with a hit rate of 100 percent [5].
Scientists at Boston Children’s Hospital have developed a catheter that
autonomously moves through blood vessels and the heart. Once in the heart,
the robotic catheter autonomously searches for the target previously defined by
physicians. To do this, the tip of the catheter has tactile sensors and a miniature
camera that provide orientation similar to that of rodents at night. The sensor
technology, which moves mainly by touch through the beating heart, is controlled
by self-learning algorithms.
During the movement to the target, the tip of the catheter regularly touches the
vessel walls with its tactile sensors. The data obtained in this way is evaluated by an
artificial intelligence system, which recognizes where the robot is currently located
on the basis of the tissue being scanned. This information is then used to calculate
the next movement until the target in the heart is reached. To prevent injuries, the
catheter tip also has pressure sensors that regulate the contact pressure and thus
avoid puncturing the vessel walls. The camera, which is also present, also enables
monitoring by physicians, who do not normally have to become active during the
surgical procedure [6].
According to the study published in the journal Science Robotics, the first animal
experiments with the surgical robot were very successful. Pigs, whose hearts are
very similar to the human organ, were used. During the tests, the catheter had
to navigate autonomously from the tip of the heart through the left ventricle to a
defective aortic valve and implant an artificial heart valve there. This application
option was chosen because defective aortic valves are also a common problem in
human medicine.
In order to be able to compare the results of the robot, the same operation was
performed in parallel by surgeons completely manually and via a catheter controlled
externally via joystick.
The results of the robotic catheter show great potential. Eighty-nine out of
90 surgeries were completed successfully and even faster than the surgeons who
navigated their catheter through the heart using a joystick. Only the surgeons who
performed the operation completely manually were even faster.
128 10 Image Processing

Pierre Dupont, professor of surgical procedures at Harvard Medical

School, explains that the results of the experiments are quite impressive, consid-
ering that they involve the robot moving around inside the blood-filled, beating heart
and having to reach a target only millimeters smaller.
According to the authors of the studies, the transforming autonomous navigation
can also be used in other minimally invasive procedures, for example, in vessels,
airways, the digestive tract, or the brain. In practice, the robot could navigate the
endoscopic surgical tool to the target, where surgeons would then take over and
perform the actual surgery. This should help ensure that, similar to the autopilot in
an airplane, routine tasks are automated so that humans have enough energy for the
really difficult parts of the procedure.

10.3 Recognition of Viruses

A DeepBreath algorithm detects Covid-19 based on breath, even in patients without

symptoms. Scientists at the Swiss Federal Institute of Technology Lausanne (EPFL)
have unveiled two artificial intelligences that can detect Covid-19. According to
the team led by Mary-Anne Hartley, DeepChest uses ultrasound images
of the lungs to do so, similar to a recently unveiled AI that detects early-stage
Parkinson’s using images of the retina. DeepChest is based on lung ultrasound
images from thousands of Covid-19 patients treated at the University Hospital of
Lausanne (CHUV) [7].
DeepBreath, on the other hand, detects Covid-19 based on breath sounds listened
to with a digital stethoscope. Alain Gervaix, a pediatrician and infectiologist,
has already been analyzing breathing sounds from his patients since 2017 to develop
the so-called Pneumoscope, an intelligent digital stethoscope. Gervaix got the idea
for it because he noticed that by listening to the lungs, he could detect asthma,
bronchitis, or pneumonia based on the sounds.
Initial results from the DeepBreath algorithm suggest that changes in the lung
tissue occur before the actual onset of Covid-19. This would particularly help
identify people with asymptotic Covid-19 infection, which is also infectious. Studies
show that up to 80 percent of SARS-CoV-2 infections are asymptomatic. According
to Hartley, “Scientists ‘want to create robust and trustworthy tools that will still be
useful after the pandemic’.” Ready for use, the DeepBreath application should be
available by the end of the year.

10.4 Medical Applications

Artificial intelligence is also making increasing inroads in the field of medicine.

Its purpose is to support the physician in diagnosis and therapy for the benefit of
the patient without losing touch with him. During colonoscopy with AI support,
10.4 Medical Applications 129

Fig. 10.3 Human internal

organs

nothing changes for the patient. As before, the flexible colonoscope is inserted into
the intestine after bowel cleansing. The images from inside the intestine are output to
a monitor and examined by the physician for changes. However, the software, which
is simply switched on at the touch of a button, additionally detects problematic areas.
In this way, up to 15 percent of changes not noticed by the doctor can be further
examined and, in case of doubt, removed. However, there are already systems that
do not require a colonoscope and move a small robot with camera and surgical tools
in the clean intestine [13].
Figure 10.3 shows the application of this approach—the picture is taken from the
“Neue Züricher Zeitung.” An important feature of image processing by computer
programs is that the images can be greatly enlarged; they can reach the level of a
microscope.
The combination of 3D printing and image processing has now been extended to
the production of artificial hearts. Scientists from Tel Aviv University have presented
a prototype of an artificial heart in the journal Advanced Science. The heart from
the 3D printer is made of human tissue, but at present, the cells cannot contract
synchronously. The organ is the size of a rabbit’s heart and otherwise has all the
characteristics of a human heart and already has individual chambers. It is the first
complete heart that scientists have been able to 3D print from human tissue. In the
future, organs made in this way could help people who currently have to wait for
donor organs that are hard to come by.
According to Tal Dvir, biotechnologist and head of the study, the heart is
currently comparable to the heart of a human fetus. In addition to greater availability,
the artificial heart also has the advantage of being completely compatible with the
recipient. In many cases with donor organs, the organ is rejected by the body and
the immune system function in the recipient must be suppressed. Since the artificial
heart is made of fat cells from the future recipient, complete biocompatibility is
ensured.
In the course of the study, the scientists biopsied fat tissue from a patient’s
abdomen and then separated the cellular and acellular components. The fat cells
130 10 Image Processing

were then reprogrammed into pluripotent stem cells containing various types of
somatic cells, including heart cells. From the acellular material, such as glycopro-
teins and collagen, the scientists created a hydrogel that was used as a support tissue
for the artificial heart.
The scientists then created a bio-ink from the endothelial and cardiac muscle
cells and the hydrogel, from which the 3D printer created tissue patches (cardiac
patches). In the past, these tissue patches have been used after heart attacks to
supply dead areas of human hearts with new cells. Now, for the first time, the patches
have been successfully used to create a completely new organ.
According to Dvir, it is “principally possible to use the technology to make a
larger heart for humans as well.” However, several years of research will be needed
before hearts from 3D printers can help sick people. The currently manufactured
heart already looks like a natural heart, but its function is not yet complete.
In principle, the artificial heart muscle is already contracting, but the coordination
between the individual heart chambers required for pumping does not work. Should
the scientists solve this problem, hearts from the 3D printer could also be used in
practice. The Israeli scientists expect that animal trials with rabbits and rats can take
place within a year. Clinical trials with humans are not expected for several years.
Currently, heart disease is one of the leading causes of death worldwide and often
cannot be treated due to a lack of donor organs, even when detected.
Researchers at the Massachusetts Institute of Technology (MIT) have used artifi-
cial intelligence to discover a very special antibiotic that, unlike most antibiotics, is
not a conventional variation of an already known drug. The researchers named the
highly effective antibiotic halicin. The name is derived from the computer HAL
9000 from the cult film 2001 [8].
The antibiotic kills bacteria by limiting their ability to maintain an electrochem-
ical potential that is used to produce molecules for energy storage. The results of
tests with the developed agent are astonishing. For example, E. coli bacteria did not
develop resistance to the new antibiotic even after 30 days. By comparison, with the
conventional antibiotic ciprofloxacin, the bacteria adapted to the drug after
just 3 days.
The algorithms are able to find molecular structures with desired properties, such
as killing bacteria in this case. The system’s neural networks automatically learn the
molecules’ representations and classify them into specific domains, which makes it
possible to predict behavior very accurately.
The researchers then trained the system with 2,500 molecules, 1,700 of which
corresponded to those of already known drugs and the remaining 800 to natural
products. With this basis, the computer developed a library of around 6,000 possible
combinations, including the newly discovered antibiotic halicin, which actually
proved to be highly effective and efficient in the test.
Experience shows that it will be some time before the new antibiotic can be used
in human medicine. In mice, however, halicin has already proven that it can easily
combat a problematic strain of bacteria with which US soldiers frequently become
infected in the Middle East.
10.4 Medical Applications 131

The researchers are very confident that the new system will have a very positive
impact on future drug development. Just to see the potential of the system, the
researchers fed the neural network with a database of more than 100 million
molecules. After only a short time, the program was already suggesting 23
potentially highly effective new drugs. In the future, however, the system should
not only develop completely new drugs but also modify existing drugs so that they
have fewer side effects and increased efficacy.
A new calculation model precisely predicts the loss of performance of athletes
into old age. A research group led by Bergita Ganse, endowed professor of inno-
vative implant development at Saarland University, has developed a computational
model to determine the physical decline of people as they age. “We asked ourselves
whether we could succeed in predicting an athlete’s performance into old age, using
a single measurement,” Ganse explains.
According to the publication in the journal Geroscience, to do this, the scientists
trained a program with data from nearly 5,000 Swedish track and field athletes
whose performances were documented in the Swedish Veteran Athletics
database between 1901 and 2021. In total, the scientists were able to use 21,000
data points for their study. On average, there were four measurement data points
from each athlete. Only running disciplines were taken into account, because in
other disciplines such as javelin or discus throwing, the weight of the throwing
equipment changes depending on the age of the athletes. This makes comparability
difficult and would make a prediction of performance decline with age less accurate.
“Runners, on the other hand, run 100, 200, 800 meters whether they are 23, 40, or
70 years old” [9].
As Ganse explains, as a central result of the study, the scientists succeeded in
developing a model that predicts the decline in performance of athletes into senior
age with high accuracy. The precision achieved is significantly higher than that of
older models, which in most cases are based on a linear decline in performance.
“We were surprised to find that athletes who were very powerful and young had the
greatest relative decline in performance, which was also true for older athletes with
lower baseline performance. We found the lowest rate of decline in high-performing
athletes with high baseline age.” Moreover, the study shows that people should still
start exercising at an older age. Individuals who ran a good time in their discipline
at an advanced age were very likely to perform significantly better than their age
average, even at an even older age.
Image processing currently has perhaps the greatest significance for daily life,
for medicine and the use of robots, and much more. There are excellent textbooks,
software packages, equipment systems, and opportunities for comprehensive and
solid continuing education.
132 10 Image Processing

10.5 Further Applications

In many cities and countries, important streets and squares are secured by cameras.
With the aid of a learning program, the control of traffic lights can be adjusted to
reduce the waiting time of vehicles; in some cases, this significantly reduces the
amount of exhaust gas. Traffic jams can also be avoided or reduced in this way
(Fig. 10.4).
Singapore lives up to its reputation as an extremely clean city. Even skeptical
natures will not encounter litter or cigarette butts on the streets or in the subways,
not even in public restrooms. The reason for this is the rigorous policy that the
city-state has consistently pursued for years in matters of cleanliness: violations of
regulations that make carelessly throwing away cigarette butts or trash in public
a punishable offense are followed by draconian fines. The principle is effective:
anyone who throws trash on the street pays 1,000 Singapore dollars. Foreigners
have to leave the country the next day.
• Due to the level achieved in image processing, blind people can be greatly
assisted. They can be given a camera that captures the whole environment. This
recognizes when roads can be crossed or when obstacles appear on the way.
It guides its wearer safely to the entrance of a bus or a streetcar. In addition,
the camera is equipped with a navigation system otherwise used in cars. In this
way, the disabled person can reach their destination relatively easily. Their life is
arduous, yet relatively safe. Some such systems even respond to acoustic input.
• A modern car today has a greater number of cameras and microprocessors.
In autonomous driving, this equipment is fully responsible for safe driving. In
particular, it recognizes all traffic signs and follows them. The same equipment

Fig. 10.4 Road traffic control

10.6 The Basics of Image Processing 133

can be installed in a normal vehicle. The driver thinks that he determines his
own driving; however, the observance of the mandatory and prohibitive signs is
enforced by the existing computer system.

10.6 The Basics of Image Processing

One can always start from a raster graphic; this is an image that is already available
in digital form. The individual color values of the raster graphics are referred to as
pixels as well as the area elements necessary for the acquisition or representation
of a color value in the case of an image sensor or a screen with raster control
(Fig. 10.5) [1].
Each pixel is characterized by its shape, its coordinates, and its color value. The
encoding of the color used in the pixel is defined, among other things, by the color
space and the color depth. The simplest case is a binary image, where a pixel
stores a black and white value. The screen resolution specifies how many pixels
make up the image representation. The larger the resolution, the more detail
can be seen on the screen or the more space is available to display objects.
The resolution is specified in different notations, values, and units. For a digital
camera, for example, 12 megapixels is used. This is the product of the side lengths
(width .× height), for example, 4,256 .× 2,848 pixels. For screen resolution,
however, it is not the product that is interesting, but the number of pixels per page
length. Since we prefer widescreen display, our screens are wider than they are taller.
Known screen resolutions are VGA (640 .× 480), Super VGA (800 .× 600),
and Full HD (1,920 .× 1,080).

Fig. 10.5 A typical pixel

image
134 10 Image Processing

Both the pixel density (and thus the local sampling rate) and the size of the
information stored in the pixel (color depth, color channels, position, shape ) are
limited in practice, which is why a pixel can only represent an approximation of
reality.
The pixels of an image sensor or display usually consist of areas of one basic
color each (red, green, and blue). In the case of liquid crystal displays (LCD),
each visible picture element is controlled by a color value. The areas responsible
for the basic colors of the pixel (subpixel) are often arranged adjacent to
each other. The finer subpixel structure compared to the pixel can be used to
increase the horizontal resolution during rasterization (subpixel rendering).
Furthermore, triangular, irregularly sized, alternately arranged, or additional white
subpixels are also possible.
The physical size of a pixel depends on the device. The pixel density of a
screen or scanner is given in pixels per inch (ppi) or dots per inch
(dpi). Commercially available computer screens achieve a pixel density of about
100 ppi, corresponding to 0.3 millimeters per pixel. For televisions, the pixel
density is usually lower and several times higher for newer smartphones, while
the sensors of scanners and digital cameras can reach several thousand ppi. The
maximum number of pixels that can be used in image sensors is often specified in
megapixels, but this usually refers only to the color dots of a Bayer sensor and
not the pixels. The aspect ratio of a pixel on the screen (pixel aspect ratio)
does not necessarily have to be 1:1; most video standards prescribe irregular pixel
aspect ratios. The pixel size as well as the pixel pitch in relation to the image
resolution have a decisive influence on the readability and recognizability of texts
and graphics on computer monitors and televisions.

10.7 The Processing of Images

After scanning an image, a digital form is available for the image, which can
be further processed with the help of suitable software. Scanners are usually
integrated with a printer. Their quality is so good that many authorities accept
scanned documents. We recall the representation (red, green, blue) by three bytes.
Their content is between (0,0,0) and (256,256,256). (0,0,0) returns black, and
(256,256,256) corresponds to white. If the byte contents are the same, then different
gray levels are obtained. (127,127,127) is in the middle between all black and all
white.
In Fig. 10.7, red (upper circle), green (left circle), and blue (right circle) were
each used with the value 30. The superimposition of the different colors already
results in quite different color levels. Figure 10.6 focuses on white and black. These
images require less memory (Fig. 10.7).
In Fig. 10.8, red, green, and blue were each used with the value 60.
10.7 The Processing of Images 135

Fig. 10.6 A black and white

representation

Fig. 10.7 The use of colors

with the value 30
Typography

LaTeX
Design Coding

Fig. 10.8 The use of colors

with the value 60
Typography

LaTeX
Design Coding

Fig. 10.9 The use of colors

with the value 90
Typography

LaTeX
Design Coding

Figure 10.9 uses the three colors with the value 90. With this value, no shading
appears, and the respective colors are distinct. You can see that the contrast of the
individual colors increases.
136 10 Image Processing

The PixelMath program used in [1] and any similar program allows different
operations. One can enlarge or reduce a pixel with the screen coordinates .(x, y).
For example, one replaces the values .(x, y) with .(2 · x, 2 · y) to enlarge the image,
or one uses .(x/2,.y/2) to reduce the image. You can also do this for the values
individually, so replace .(x, y) with .(x/2, y) and so on. For example, you can make
an image gallery where all images have the same size. But you have to keep in
mind that the number of pixels does not increase, only the distance between them
becomes smaller or larger. Of course, this can lead to effects that change the image
quality.
Since a Cartesian coordinate system is present, one can carry out all transforma-
tions, which can be expressed by formulas of the analytic geometry . One can also
use polar coordinates , there one can express very nicely effects that start circularly
from a point. As an example, we look at the fish-eye projection .
If one uses spherical coordinates, then each point in space is determined by
.(r, ϕ, θ ). The spherical radius r has theoretically no influence on the mapping

function. Thus, all objects can be scaled to exactly one distance, and an ambient
sphere is created. For the radius of the surrounding sphere, one chooses the focal
length of the lens (.r = f ).
With Cartesian coordinates, the image plane is located at .z = f and touches
the surrounding sphere at the pole (.θ = 0). Plane polar coordinates are applied
to the image. The position of an image detail is described by .(r, φ). In this
geometric arrangement, the imaging functions can be well illustrated and usually
constructed. The object coordinates .ϕ (azimuth) and .θ (polar angle) map into the
image coordinates .φ (azimuth) and r (radius).
The principle is the same as in the azimuthal map mesh of the earth. There .φ = ϕ
is valid. However, in contrast to the earth sphere, the surrounding sphere is viewed
from the inside. The front side of the image plane then shows the environment
as a mirror image. For the backward view required here, the image coordinates
(polar and 2D Cartesian) are oriented differently, so the azimuth .−0.◦ direction and
azimuth-orbit sense may differ between the environment and the image. Azimuth
angular distances are imaged unaltered, and imaging function and azimuth do not
affect each other. Thus, an azimuthal imaging is present.
The imaging function .g describes how an object in the polar angle .θ appears on
the image displaced from the center by radius r. The lens has the (central) focal
length f . With .f = 1, .g becomes the normalized imaging function .h:

r = g(θ, f ) = f · h(θ ).
. (10.1)

Lateral objects appear, on the image compared to the central position, in a

different size, which usually also differs meridionally and sagittally:

dh
meridional scaling:
. SM = . (10.2)
dθ
10.7 The Processing of Images 137

h(θ )
sagittal scaling: Sθ = . (10.3)
sin(θ )

From this, solid angle scaling .SΩ , linear (effective) scaling .S, and deformation
D can be derived:
.

SΩ = Sm · Sθ .
. (10.4)

S = SΩ. (10.5)
Sm
D= . (10.6)
Sθ

The fish-eye transformation is defined by the following system of equations:

θ
.angularly true: r = 2 · f · tan . (10.7)
2
equidistant : r = f · θ. (10.8)
θ
true to area: r = 2 · f · sin . (10.9)
2
orthographically: r = f · sin(θ ). (10.10)

There are a variety of ways to change the parameters and angle functions used,
creating a variety of interesting projections, for many different applications.
• Scientists and resource managers (e.g., biologists, foresters, geographers, and
meteorologists) use fish-eye lenses to image a hemispherical region of plant
vegetation in ecophysiology or to predict potential shortwave irradiance from
horizon elevation (sky view factor) and the longwave irradiance to be
derived from it. Relevant structural parameters of tree canopies such as leaf area
index (LAI), leaf angle distribution, and ground-level light availability can be
derived from the analysis of these images.
• Fish-eye lenses also help in evaluating forest health, detecting overwintering sites
of monarch butterflies, and managing vineyards.
• In topoclimatology, horizon elevation of ground depressions can be used to infer
the origin of deep frosts during inversion weather, as well as to make statements
about the causes of cold air lake phenomena.
• From data of the sky view factors parameters obtained from georefer-
enced fish-eye images, meteorological correlations of the radiation balance of
urban heat islands are studied in urban climatology.
• Meteorologists determine cloud cover (degree of coverage of the sky).
• Astronomers image a large portion of the sky to record constellations, the Milky
Way, meteors, auroras, and light pollution.
• Many planetariums use fish-eye projection lenses to project the night sky or other
digital content onto the inside of a dome.
138 10 Image Processing

• Surveillance cameras with fish-eye lenses can capture an entire room at once.
Unlike swiveling cameras, there is no intermittent dead zone and no vulnerable
drive.
• Video door phones with extra wide angle of view (door viewer function).
• In the IMAX dome system (previously “OMNIMAX”), film capture is through
a circular fish-eye lens, and projection of the motion picture is through similar
optics onto a hemispherical screen.
• Photographers and videographers use fish-eye lenses to get the camera as close
as possible to the decisive point for action shots while also capturing the overall
context. For example, in skateboarding, the focus is on the board, and the skater
is still in the frame.
• The first music video shot entirely with a fish-eye lens was the song “Shake Your
Rump” by the Beastie Boys in 1989.
• Flight simulators and visual combat simulators use fish-eye—projection lenses
to create a seamless environment for pilots, air traffic controllers, or military
personnel to train in.
• In computer graphics, circular fish-eyes can be used to create an environment
mapping of the physical world. A complete 180-degree fish-eye image can fill
half of a cubic environment mapping with an appropriate algorithm. Environment
maps can be used to render 3D objects embedded in virtual panoramas.
Many image transformations are used in painting, for example, the cylinder
anamorphosis. An image transformed in this way looks normal when reflected
in a cylindrical mirror. A famous example is Holbein’s painting “The Envoys”
(Fig. 10.10).
• The seventeenth century saw a revival of fantastic anamorphic imagery. Mag-
ical and religious connotations were largely abandoned, and the images were
understood as scientific curiosities. Two important works on perspective were
published: Perspective (1612) by Salomon de Caus and Curious
Perspective (1638) by Jean-Francois Niceron. Both contained extensive
scientific and practical information about anamorphic images. Niceron’s work
explains three types of large-scale anamorphism: optical (horizontal view),
anoptical (upward view), and catoptical (downward view, e.g., from a
mezzanine). A conic perspective is also described. Toward the end of the
century, the techniques for creating anamorphic images became widely known
through Charles Ozanam’s Mathematical Recreations.
Between 1669 and 1685, both perspective and mirror anamorphosis were
introduced by the Jesuits to Emperor K’ang-hi and the monks of the Beijing
Mission in China. However, Chinese production of anamorphic images was
already taking place on a large scale during the late Ming Dynasty. Unlike the
halftone system used in the West, the images were usually created freehand.
Since Chinese anamorphoses focused primarily on erotic subjects, a Jesuit
influence is unlikely. It is considered likely that Chinese catoptic techniques,
which are technically unrelated to geometric anamorphosis, influenced European
mirror anamorphosis and not vice versa.
10.7 The Processing of Images 139

Fig. 10.10 Diagonally below

is an anamorphic image of the
standing man

Baroque wall paintings often used anamorphism to combine actual archi-

tectural elements with illusory painted elements to create a seamless effect
when viewed from a particular location. The dome and vault of the Church of
St. Ignatius in Rome, painted by Andrea Pozzo, represented the pinnacle of
illusion. Because neighboring monks complained about light pollution, Pozzo
was commissioned to paint the ceiling to look like the inside of a dome, rather
than a real dome. Since the ceiling is flat, there is only one place where the
illusion is perfect and a dome looks undistorted.
• In the eighteenth and nineteenth centuries, anamorphism entered the realm of
entertainment and pastime and experienced the greatest spread of the tech-
nique. The nineteenth century saw a resurgence of interest in anamorphism for
architectural illusions, and classical subjects also came into vogue. Reprints of
Renaissance engravings became popular, as did political, obscene, and popular
subjects. Edgar Allan Poe’s short story “Ligeia” describes a room filled with
“simple monstrosities,” which are reduced to “an endless succession of ... ghastly
forms” as the narrator walks through the room. This mass popularization would
later have an impact on the Surrealists (Fig. 10.11).
• In the twentieth century, some artists wanted to renew the technique of anamor-
phosis for aesthetic and conceptual reasons. During World War I, Arthur Mole,
an American commercial photographer, used anamorphic techniques to create
patriotic images of huge, assembled groups of soldiers and reservists. Viewed
140 10 Image Processing

Fig. 10.11 Monstrosities

from a tower at their base, the assembled people dissolved into recognizable
images.
Surrealist artist Salvador Dali used extreme foreshortening and anamor-
phosis in his paintings and works. A glass floor in a room next to his studio
allowed for radical perspective studies from above and below. In the Dali Theater
and Museum, there is a three-dimensional anamorphic living room installation,
Mae West Lips Sofa (Fig. 10.12), which looks like the movie star’s face
from a certain angle.
• In the twentieth century, artists began to play with perspective by drawing
impossible objects. These objects included stairs that always lead upward or
cubes where the back meets the front. Such works were popularized by the artist
M. C. Escher and the mathematician Roger Penrose. Although they are called
impossible objects, such objects as the Necker cube and the Penrose triangle can
be modeled in 3D by anamorphic illusion. When viewed from a certain angle,
such sculptures appear as the so-called impossible objects (Fig. 10.13).
• The Ames room was invented in 1946 by the American scientist Adelbert
Ames Jr.. When viewed through a peephole, the room appears to have a
normal perspective. From all other angles, however, one can see that the room
is made up of irregular trapezoids. Similar effects were achieved as early as
the Renaissance through the use of accelerated perspective in stage
design. These included productions by Scamozzi (1588–1589), Furtenbach
(1625), Sabbattini (1637), and Troili (1672).
10.7 The Processing of Images 141

Fig. 10.12 Mae West Lips

Sofa

Fig. 10.13 An impossible

geometric construction

One of the most interesting effects of an Ames space is that the distorted
perspective can make people and objects appear much larger or smaller than they
actually are. For this reason, Ames rooms are often used in cinema for practical
special effects. A well-known example is the houses in the Shire in the films
The Lord of the Rings and The Hobbit. By using forced perspective,
the character of Gandalf appeared much larger than the characters of Frodo and
Bilbo, without the use of digital effects (Fig. 10.14).
• Cinemascope, Panavision, Technirama and other widescreen formats
use anamorphosis to project a wider image from a narrower film frame. The
IMAX company uses even more extreme anamorphic transformations to project
moving images from a flat film image onto the inside of a hemispherical dome,
in its Omnimax or IMAX Dome process.
142 10 Image Processing

Fig. 10.14 Distortions in all directions

The technique of anamorphic projection is often seen in text written at a very

shallow angle on the roadway, such as bus lanes or children crossing, so that it
can be easily read by drivers who would otherwise have difficulty reading at an
angle as the vehicle approaches the text; when the vehicle is almost over the text,
its true, unusually elongated shape can be seen. Similarly, many sports stadiums,
especially rugby soccer in Australia, advertise company brands painted on the
playing surface; from the point of view of the television camera, the lettering
appears as characters standing vertically in the field. Many letterings on shop
windows are in principle anamorphic, having been written in mirror image on
the inside of the window glass [1, 3].
Representations of mathematical formulas do not cause any difficulties: one
calculates respectively .x ⇒ f (x), .(x, y) ⇒ f (x, y) and .(x, y, z) ⇒ f (x, y, z).
Image processing is in itself a boundless field. Here are some more concepts to
get familiar with:
• stereograms
• autostereograms
• Steganography: The use of steganography aims at secrecy and confiden-
tiality. Information is hidden in such a way that a third party does not suspect
anything when looking at the carrier medium. Thus, it is at the same time reached
that the hidden information does not become third admits, i.e., the secrecy
is ensured (as with the cryptography) [4]. Steganography is usually classified
in two possible ways: it is regarded as either a subchapter of cryptography
or as an independent scientific field. In favor of the latter is the fact that the
10.7 The Processing of Images 143

Fig. 10.15 A message

hidden in the picture

objective of cryptography (secrecy) is not the same as the objective of steganog-

raphy (confidential secrecy by concealing secrecy). In practice, cryptography
and steganography are often combined because, for example, ciphertexts have
interesting statistical features for steganography (Fig. 10.15).
Inserting the message is not difficult. [1] The pixels of the message to be
hidden are inserted into the second image according to a certain rule, usually
defined by a mathematical function, and can then be reassembled on the
receiver’s side.
• Very attractive images are created by the different fractals, and also the formation
of mosaic images is an interesting art. Designers today hardly work with the
drawing board, but on the screen. A significant advantage is that you do not have
to start drawing new designs from scratch, but can continue to process unfinished
designs that still exist.
Image processing has also become very significant for the work of the police.
Pattern recognition, comparison of faces, and fingerprints are standard. Surveillance
cameras and the analysis of their images have almost eliminated petty crime in
important business areas. For longer absences, surveillance cameras can also be
installed at home.
Clothing of all kinds is also designed on the computer. Graphics Suite X5
by CorelDRAW is a program that facilitates working with hand-drawn graphics
as well as creating virtual textiles. Various functions such as recoloring scanned
clothing drawings or photographs, as well as a construction kit function, make it
easier to work as a fashion designer.
London researchers have trained a program on the style of artists. This enables
them to reconstruct works that were long thought to be irretrievably lost. In the early
stages of his career, Pablo Picasso (1881–1973) was apparently so strapped for cash
that he couldn’t afford new canvases. So he painted over existing works. This is
what happened with the work “The Lonely Crouching Nude,” which disappeared
144 10 Image Processing

Fig. 10.16 Pablo Picasso:

The Lonely Crouching Nude

Fig. 10.17 A painting in the

style of Modigliani
(1884–1920)

behind the “Meal of the Blind” in 1903. In the background of Picasso’s famous
painting “The Life,” the crouching woman can still be seen, but it was to take more
than a hundred years for the original to reappear (Fig. 10.16). The starting point
was a fluoroscopy of the original by X-ray fluorescence spectroscopy. But since
only the outlines of the crouching woman could be revealed in this way, a further
step was necessary. George Cann and Anthony Bourached, researchers at University
College London and founders of Oxia Palus, trained a convolutional neural network
with all the images from Picasso’s “Blue Period.” From this, the AI generated
Picasso’s typical style, brushstrokes, and color choices during this creative phase
and transferred them to the outlines. The result was then printed out in original size
on a 3D printer—and the new old Picasso was ready. Now, London-based start-up
Oxia Palus has achieved another step in reconstruction: it has reconstructed the
painted-over image and restored it using 3D printing.
References 145

Back in 2019, Cann and Bourached discovered and reconstructed a female

nude under Picasso’s “old guitar player”—but at the time, without printing it out.
Oxia Palus has since reconstructed overpainted images by Leonardo da Vinci and
Amedeo Modigliani using the same principle. The images were printed in Los
Angeles by the service provider Hero Graphics. Modigliani’s work can be purchased
through London’s Morf Gallery in a limited edition for 22,222.22 euros. Buyers
receive not only a framed piece of canvas but also digital documentation of the
production and an NFT as a certificate of authenticity (Fig. 10.17). In a similar way,
however, completely new works by old masters can also be created. In 2016, a Dutch
advertising agency had statistically analyzed all 346 known works by Rembrandt
and distilled from these data something like a typical Rembrandt—a portrait
of a middle-aged man looking to the right [10].
Of course, it is not at all necessary to orient to earlier paintings. At present,
it is possible to see that a new direction “painting on the screen” arises [11][12]
and affects all other arts (music, poetry) in the same way. Immediately apparent are
many advantages: digital art requires no other material, no canvas, and no paints. All
intermediate results can be saved and later perhaps serve as a template for another
version and much more.

References

1. Steven L. Tanimoto. An Interdisciplinary Introduction to Image Processing - Pixels, Numbers,

and Programs, The MIT Press, Cambridge, Massachusetts, 2012, 978-0-262-01716-9
2. https://innovationorigins.com/de/erdbeben-analyse-kuenstliche-intelligenz
3. https://en.wikipedia.org/wiki/Anamorphosis
4. https://de.wikipedia.org/wiki/Steganographie
5. https://www.forschung-und-wissen.de/nachrichten/medizin/google-ki-schlaegt-mediziner-
bei-der-frueherkennung-von-brustkrebs-13373664
6. https://www.forschung-und-wissen.de/nachrichten/medizin/kuenstliche-intelligenz-bewegt-
roboter-katheter-autonom-zum-herzen-13372909
7. https://www.forschung-und-wissen.de/nachrichten/medizin/kuenstliche-intelligenz-erkennt-
covid-19-fruehzeitig-am-atem-13374406
8. https://www.forschung-und-wissen.de/nachrichten/medizin/kuenstliche-intelligenz-
entwickelt-hocheffektives-antibiotikum-13373708
9. https://www.forschung-und-wissen.de/nachrichten/medizin/kuenstliche-intelligenz-
prognostiziert-leistungsverlust-von-sportlern-13375408
10. https://www.heise.de/hintergrund/Uebermaltes-Picasso-Werk-durch-3D-Druck-
wiederhergestellt-6233298.html
11. https://www.blitzgedanken.de/kunst/digitale-kunst/was-ist-computerkunst
12. https://de.wikipedia.org/wiki/DigitaleKunst
13. Volker Watschounek. Darmspiegelung, unterstützt von Künstlicher Intelligenz https://
wiesbaden-lebt.de/2021
Chapter 11
Robotics

Robots are a core issue of artificial intelligence for many; especially the topic self-
driving cars is the basis of discussions about autonomous systems independent of
humans. Opinions vary widely. The subject area of robotics deals with the attempt to
reduce the concept of interaction of a system with the physical world to principles of
information technology as well as to technically feasible kinetics. The term “robot”
describes a system that combines both concepts by implementing interaction with
the physical world on the basis of sensors, actuators, and information processing.
The core area of robotics is the development and control of such robots. It
encompasses subfields of computer science, electrical engineering, and mechanical
engineering. The goal of robotics is to establish a controlled cooperation of robot
electronics and robot mechanics through programming.
The term was invented by science fiction author Isaac Asimov, first men-
tioned in his short story Runaround in March 1942 in Astounding magazine.
According to Asimov’s definition, robotics refers to the study of robots or even
machines.
• Already in the antiquity, first experiments with automata were carried out
(Fig. 11.1). Known are automatic theaters and music machines of Heron of
Alexandria. With the decline of ancient civilizations, the scientific knowledge
of that time disappeared. Around 1205, Al-Jazari, a twelfth-century Arab
engineer and author, wrote his work on mechanical devices, the “Book of
Knowledge of Sensible Mechanical Devices” which also became known as
“Automata” in Western culture. Leonardo da Vinci is said to have been
influenced by the classical automata of Al-Jazari. Thus, records and sketches
from the fifteenth century are known from him, which can be interpreted as plans
for androids. However, the level of technical knowledge was not sufficient to
realize such plans. Around 1740, Jacques de Vaucanson designed and
built a flute-playing automaton, an automatic duck and the first programmable
fully automatic loom.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 147
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_11
148 11 Robotics

Fig. 11.1 A water organ

from ancient times

• At the end of the nineteenth century, increased efforts were made in the military
(remotely operated boats, torpedo controls). The writer Jules Verne (1828–
1905) wrote a story about a human machine. In 1920, the writer Karel Čapek
introduced the term robot for an android. After the end of World War II, the
field of robotics underwent rapid progress. Key factors were the invention of the
transistor in 1947 at Bell Laboratories and integrated circuits and, subsequently,
the development of powerful and space-saving computers.
• From around 1955, the first NC machines (devices for controlling machines)
appeared on the market, and in 1954 George Devol (1912–2011) applied for
a patent for a programmable manipulator in the USA. This date is considered
the birth of the development of industrial robots. Devol was also a co-founder of
Unimation, which introduced the first hydraulically operated industrial robot in
1960. In 1968, the first mobile robot was developed at MIT. In Germany, robot
technology was not used productively until the early 1970s. Around 1970, the
first autonomous mobile robot Shakey (the shaky one) was also developed
at Stanford Research Institute.
• In 1973, development of the WABOT-1 humanoid robot was launched at Waseda
University in Tokyo. That same year, German robotics pioneer KUKA built the
world’s first industrial robot with six electromechanically driven axes, known as
FAMULUS. A year later (1974), Sweden’s ASEA introduced its fully electrically
driven robot (IRb6).
• In 1986, Honda launched the “Humanoid Robot Research and Development Pro-
gram.” The result was the humanoid robot versions P1 to P3. Honda introduced a
11 Robotics 149

Fig. 11.2 A drone

further development in 2004 in the form of the ASIMO humanoid robot. In 1997,
the first mobile robot landed on Mars (Sojourner).
• The toy industry has not closed its mind to robotics either. Examples of
such products are Lego Mindstorms, iPitara, Robonova, or Sony’s robot dog
Aibo (Fig. 11.2).
According to futurologists and philosophers, the ever-increasing automation and
digitization, combined with likewise growing collection and exchange of data,
requires fundamental questions about the role of humans in this process and in these
contexts. As early as 1942, Asimov, for example, formulated a corresponding code,
the “robot laws.”
• Alternative techniques to the wheel as a means of locomotion in the human
environment are now being researched, such as walking on six, four, two, or
even one leg. While industrial robots usually perform manual or handling tasks
in an environment adapted to them, such service robots are intended to provide
services for and to humans. To do this, they must be able to move and find their
way around in the human environment.
• Like a game, but with serious scientific research as a background are robot soccer
games between teams of similar robots. The researchers’ goal is to develop a
soccer team of autonomous bipedal robots by 2050 that can compete against the
soccer world champion.
• Industrial robots are mostly used in environments that are too dangerous or
unreasonable for humans. Today, modern robots do mindless assembly line work
faster and much more accurately than a human and can replace them in more
and more areas. Cars are now built with heavy robotic involvement, and even
a modern microprocessor would be impossible to manufacture without a robot.
Service robots have been used for some time to make people’s daily lives easier or
to entertain them, such as the RoboSapien. There are household robots that are
able to vacuum, mop the floor, or mow the lawn. They may specialize in only one
task, but they can perform it relatively autonomously. Research robots explore
distant planets or disaster zones, among other things, and penetrate volcanoes
150 11 Robotics

or sewage pipes. They are also used for a wide variety of detection missions in
the marine environment. There are concepts and first prototypes for cryobots
and hydrobots that will be used in space travel in the future. There are also
considerations to use robots for sample retrieval missions and asteroid mining.
• In medicine, robots are used for examinations, operations, and rehabilitation and
perform simple tasks in everyday hospital life.
A prototype for tiny nanorobots that can move in the bloodstream was already
tested on an eye at ETH Zurich in 2004. They are controlled by magnetic fields
from the outside. The FRIEND assistance robot, developed at the Institute of
Automation Technology at the University of Bremen, is designed to support
disabled and elderly people in the activities of daily living (e.g., preparing a meal)
and enable them to reintegrate into working life.
• First entertainment robots such as Sony’s robotic dog Aibo are a step toward
electronic pets. In addition to Aibo, there are other robot products from the toy
and entertainment industry that can be programmed with a computer in a mostly
simple language to, for example, follow a light source or a line on the floor or
sort colored building blocks.
• An interesting hobby is to build your own robots. This can be done supported by
prepared robot kits or by free imagination. In this case, for example, you have
to construct a car-like vehicle yourself, determine distances to the target or the
color of the ground with suitable sensors, and determine a course for the vehicle
to follow from these measurement results. The actual task is to link the sensor
data with the speed and direction of the vehicle. This is done in a microcontroller
that has to be programmed by the user. The required electronics are offered in
different designs. Well-known but also very elaborate role models are the Rovers.
• Robots are also a popular science fiction subject. There, there are humanlike
robots that have artificial intelligence.
An additional variation of the robot, already realized in a very simple form, is
the cyborg as a fusion of robotic technology with human anatomy. Androids,
artificial humanlike beings, can be robots, but robots need not be androids. A first
widely developed approach is Honda’s ASIMO robot.
• Robots are also increasingly an important topic in education. There are robots for
elementary school, robots for secondary school or for secondary schools, robots
for universities, and robots for vocational training. A special form of robots for
education are rovers, which are developed and tested, for example, as part of
space education at institutions in Germany. Usually, these specialized robots are
intended as rovers for a specific goal or competition. At the Maker Faire 2016
in Berlin, a rover called “EntdeckerRover” ER2 was presented, which is suitable
for education and leisure and can also be adapted for different educational areas.
Other systems are mostly available in plastic from other manufacturers and
projects.
• Robots and the special form rover mostly support education in the area of STEM
subjects in Germany and Austria. So it is also about the promotion of science and
technology education or technology knowledge as well as the topics of computer
11 Robotics 151

science and mathematics. Mathematics is particularly important for sophisticated

robots and rovers, for example, in the aerospace and aviation sector.
• Safety guidelines for robots result from the respective area of application and
the robot type. Industrial robots are safeguarded by legally prescribed safety
precautions such as cages, grids, light barriers, or other barriers. However, with
increasing autonomy, current or future more complex robot systems require
safety precautions adapted to the circumstances. However, the diverse use of
robots makes it difficult to establish universal safety rules for all robots. Even
the “three (or four) rules of robotics” (robot laws) established by science fiction
author Isaac Asimov in his novels can only be understood as ethical guidelines
for possible programming, since unpredictable situations cannot be calculated by
the robot. The more autonomously a robot acts in the human environment, the
greater the probability that living beings or objects will come to harm. Likewise,
the idea that robots can offer protection to humans is controversial—not least
because of the fuzziness of the term protection. The fact that no absolute values
can be programmed here is shown in parallel in the discussion about the tension
between protection and paternalism. This problem is addressed, for example, in
the film I, Robot, where, on the basis of a calculated “probability of survival,”
a man is rescued by a robot from a car that has fallen into the water, while a child
drowns in a car that is also sinking.
Lawyers are in high demand around these new, rapidly evolving technologies. It
is astonishing that laws concerning robotics were passed only from 2012, but then
quite a lot.
“A robot is a technical system with an embedded computer system; the systems
interact with each other. In this context, the computer system has the task of
controlling, regulating or monitoring the technical system in which it is embedded”
(ECJ, July 3, 2012 - C-128/11 = NJW 2012, 2565). An embedded system always
contains “embedded software.” Without this software, a robot would not be usable.
Even before the ECJ decision (ECJ, July 3, 2012 - C-128/11 = NJW 2012,
2565) on the resale of second-hand software, the TRIPS Agreement and WIPO
Copyright Treaty (WCT) stipulated that hardware with embedded software may be
freely traded (Vander, CR 2011, 77 (78–79)). There is also agreement that embedded
software also does not count as an essential element of a rental and thus for rental of
hardware (e.g., robots) controlled by embedded software, no rental right within the
meaning of section 69 c (3) UrhG needs to be explicitly transferred, even if some
authors refer to a case-by-case consideration (Grützmacher in Wandtke/Bullinger,
UrhR, 3rd edition 2009, section 69 c marginal no. 48). As a result, it can be stated
that robots may be sold and rented out without the need for additional rights.
In Germany, patents can be protected by the Patent Act (PatG); in the EU the
European Patent Convention (EPC) protects patents. The PatG defines a patent in
the first section (§§1–25 PatG). According to section 1 (1) PatG, patents are granted
for inventions in all fields of technology, provided that they are new, involve an
inventive step, and are susceptible to industrial application. According to section 3
(1) Patent Act and Art. 54 EPC, an invention is considered new if it does not belong
152 11 Robotics

to the state of the art. The state of the art comprises all knowledge which has been
made available to the public by written or oral description, by use, or in any other
way before the date relevant for the priority of the application; cf. section 3 (1), 2nd
sentence, Patent Law. In the case of robots, the patent applicant must therefore show
that his robot has new functions that do not belong to the state of the art (e.g., on the
ability of robots to run).
Furthermore, it must be an invention. Patentable inventions are technical teach-
ings for planned action which reproducibly bring about a causally foreseeable
success using controllable natural forces without the interposition of intellectual
activities (BGH, March 27, 1969 - X ZB 15/67 = BGHZ 52, 74; NJW 1969, 1713;
GRUR 1969, 672). A technical further development of a robot is a patentable
invention only if it is not obvious from the prior art to “the average person skilled in
the art who knows the entire prior art” (a legal fiction, not a real person); cf. section 4
p. 1 Patent Act, Art. 56 p. 1 EPC. That is, there is a lack of inventive step if the
person skilled in the art can be expected to have come up with this solution promptly
and with a reasonable effort, based on the prior art, without being inventive. Thus,
in the field of robotics, only inventions that represent a significant advance in the
development of robot technologies are patentable. However, this does not have to
refer to the robot as a whole, but can also refer to individual components, such as a
robot arm or a mode of operation for locomotion.
In addition, the invention must be applicable in some industrial field according
to section 5 (1) Patent Act, Art. 57 EPC. The term “industrial applicability” is
interpreted broadly by the European Patent Office and is of secondary importance
in practice. It is sufficient that the invention can be manufactured or otherwise used
in a technical business. It is also not important whether one can “make money” with
the device or process; the only decisive factor is that the claimed subject matter
can be used outside the private sphere. Most inventions in the field of robotics are
aimed at commercial success, for example, in the creation of household helpers or
robots for operations. This is already in the nature of things, since the inventions of
robotic technologies require enormous investments and these are reclaimed by the
investment providers at a profit.
The maximum term of a patent is 20 years from the day after filing, according
to §16 PatG and Art. 63(1) EPC. However, according to section 16a PatG, Art. 63
(2) (b) EPC in conjunction with Regulation (EEC) No. 1768/92, a supplementary
protection certificate can be granted for inventions that can only be commercially
exploited after extensive approval procedures, which then extends the patent term by
a maximum of 5 years. Due to the long development cycles in robotics, this should
be applied regularly.
According to section 1 (2) and (3) Patent Act and Art. 52 (2) and (3) EPC,
scientific theories and mathematical methods, such as construction plans for a robot,
cannot be protected as patents. The same applies to the design and appearance of a
robot, since aesthetic creations of form cannot be protected by patent.
A robot’s misbehavior, whether it stems from the desire for autonomy or any
other reason, always entails a number of liability issues. These can arise, on the one
hand, from a contractual breach of duty pursuant to section 280 (1) of the German
11 Robotics 153

Civil Code, on the other hand from tort law pursuant to section 823 of the German
Civil Code vis-à-vis third parties, or also from the Product Liability Act. If a robot is
used in the context of a contractual relationship (e.g., rental) by another contractual
party and the robot causes damage to this party, this is certainly a breach of duty
in the sense of §280 BGB. A case that has become known through the media is the
use of the ROBODOC from Integrated Surgical System, which has led to numerous
claims for damages (BGH, June 13, 2006 - VI ZR 323/04 = BGHZ 168, 103; NJW
2006, 2477).
According to §249 p. 1 BGB, the debtor who is obligated to pay damages must
restore the condition that would exist if the circumstance obligating him to pay
damages had not occurred. In doing so, the tortfeasor shall compensate for all
damage caused by the result obligating to compensation (so-called total reparation).
Apart from the rule of total reparation, another principle of the law of damages is
expressed in §249 p. 1 of the Civil Code, namely, the principle of production
or natural compensation. Here, the tortfeasor is to produce the condition in
money, which would exist without the damaging event.
A question that will certainly become increasingly important in the future is who
is liable for the decision made by a robot based on artificial intelligence. Thus, it
is certainly arguable that the person who uses the robots must be liable, since he is
responsible for the road safety of the robot used and must ensure appropriate safety
measures. In a contractual relationship, this certainly arises from the general duty
of care of the obligation relationship (cf. §280 Para. 1 BGB) toward third parties
certainly from the tort law, §§823 ff BGB. In principle, the manufacturer could be
liable under the Product Liability Act (ProdHaftG). The prerequisite for product
liability according to §1 para. 1 p. 1 ProdHaftG is, among other things, that there
was a defect in the object causing the damage (i.e., in the robot). Such a defect could
possibly exist if the manufacturer has not incorporated suitable safety measures in
the programming of the robot’s control software. In any case, the manufacturer is
not liable if the robot did not have the defect causing the damage at the time it
was placed on the market (Palandt Sprau Kommentar zum BGB 69. Auflage 2009
§1 ProdHaftG Rn. 17) and if the defect could not have been detected according
to the state of the art in science and technology at the time the manufacturer
placed the product on the market; cf. §1 para. 2 no. 5 ProdHaftG. Nevertheless,
the manufacturer of robots must build safeguards into a robot (and especially into
the software) so that no damage can occur, even after an AI learning process. In
science fiction literature, for example, the three laws of robotics were developed
by Isaac Asimov for this purpose (Asimov All Robot Stories 3rd edition 2011,
short story Herumtreiber (English Runaround), pp. 276–295). Whether such rather
philosophical laws are sufficient cannot be judged today, but it is certain that the
manufacturer and developer of robots have corresponding duties to ensure traffic
safety. However, the maintenance of these traffic safety obligations is then no longer
the responsibility of the manufacturer, but of the keeper or owner of the robot.
The principles for dealing with dangerous goods apply here. A dangerous object is,
for example, a motor vehicle that poses a certain operating risk. The manufacturer
produces a car that meets the relevant requirements for the registration of a motor
154 11 Robotics

vehicle, while the owner must ensure that the vehicle is constantly in a roadworthy
condition (BGH, October 14, 1997 - VI ZR 404/96 = NJW 1998, 311). In particular,
this applies in the case of a guarantor position vis-à-vis third parties (BGH, April
24, 1979 - VI ZR 73/78 = NJW 1979, 2309). The same should also apply to the
manufacture and use of robots.
So, as a manufacturer of robots and as their operator, you have to consult a lawyer
every now and then and discuss any problems that arise.
Employees of the Scientific Center in the Republic of Kabardino-Balkaria have
developed a caterpillar robot in the shape of a turtle shell, reports the news agency
RIA Novosti. This is primarily intended to help the Russian National Guard
Rosgvardiya disperse crowds at rallies. According to the developers, the vehicle
should move faster than pedestrians or mounted police.
The robot is about half the size of a human and armed with stun guns, net
throwers, and fast-setting foam. Its streamlined shape is designed to avoid injury
to humans while preventing protesters from tipping it over.
The scientists propose two options. The first envisions the robot moving at 60
kilometers per hour and cutting through crowds. In the second option, the robot acts
as a shield bearer to erect barriers and divide the rowdy crowd. The invention is
also said to be able to network and act together based on an algorithm modeled on
state-forming insects such as waspseren.
Most accidents with robots occur during maintenance or programming of the
robot, not during controlled operation. On July 21, 1984, the first human was killed
by an industrial robot in Michigan, USA. The robot was moving workpieces on a
die-casting machine. The 34-year-old factory worker already had 15 years of die
casting work experience and had completed a 1-week robot training course only
3 weeks before the accident. He was crushed to death between the supposedly
safe back of the robot and a steel post when, against all warning, he climbed into
the robot’s danger zone to remove scattered production debris. The US “National
Institute for Occupational Safety and Health (NIOSH)” provides guidelines for
robot design, training, and guidance for workers.
It can be seen from this that the manufacture and use of robots has become an
integral part of the engineering world. This is where the benefits of using these
systems and the problems associated with them become apparent.
• In many countries, children, young people, and students have the opportunity
to participate in robotics programs. They form teams, each of which is faced
with the task of programming a robot equipped with motors and sensors so that
it can solve predetermined tasks, such as sorting objects and moving them to
specific locations, on a playing field in a given time frame, either autonomously
or by remote control. In some of the programs, the task includes designing and
building the robot; in others, prefabricated robots are used. The teamwork results
in competitions, many of which are held at an international level.
• Man-made artificial beings have a long tradition in literature, well-known exam-
ples being the Golem created by magic or alchemically generated Homunculi.
11 Robotics 155

A robot-like manikin mistaken for the daughter of Emperor Maurikios of the

Iranian court is mentioned in the Persian epic Shahnameh by the poet Firdausi.
However, the idea of robots in the sense of machine people or autonomous
machine beings realized by human science and technology does not develop
until modern times. In the literature of the eighteenth century, deceptively
humanlike automata are described, which are brought to (apparent) life by an
elaborate mechanism. E. T. A. Hoffmann tells in “Die Automate (1819),” about
a mechanical oracle, which is based on the so-called Schachtürken. The chess
machine consisted of a figure of a man dressed in Turkish costume, sitting behind
a table with a chessboard. The figure played with many famous chess players of
the time and usually won. The Turk always started the game, raised his left arm,
moved the chess piece, and then put his arm back on a pad. At each move of
the opponent, he looked around the board. If the move was wrong, he shook his
head and corrected the position of the piece. At gardez (threat of the queen), he
nodded his head twice, at chess three times. All movements were accompanied
by a sound similar to that of clockwork running down.
Kempelen, the inventor, who was happy to show the inside of the machine
and its mechanics to anyone who wanted to see it, stood a little apart during the
game, looking into a small box that stood on a table. He left open, unspoken,
the possibility that a transmission was made to the machine by a human, but
always declined to give any hint as to the underlying principle of operation. The
possibility of magnetic transmission of train commands was puzzled over by
observers, as was the possibility that the machine could perform the calculations
independently, or at least for a section of several trains, without any human
intervention. Since the machine played very strongly, a good player must have
sat in it.
• In the penny dreadfuls of the late nineteenth century, for example, there is a
fiction of a steam-powered machine man “The Steam Man of the Prairies” by
Edward S. Ellis, 1868. Since the word robot has not yet become common
for artificial humans, such figures are still referred to as automaton or
imechanical man. In 1886, the writer Auguste de Villiers de L’Isle-Adam
published the novel “L’ Eve future,” in which a female automaton is supposed to
serve as the companion of a distinguished lord.
For the first time, the term “robot” coined by Josef Čapek was used in
1921 in the drama “R.U.R.”. The Roboti presented in the play are artificial
humans based on synthetic protoplasm, but the term was soon applied to the
contemporary notion of mechanically constructed automata. The term robot
became known to a wider audience through the stories of Isaac Asimov. In 1942,
in “Runaround,” he described for the first time the three robot laws, which are
still frequently received in science fiction literature today. In 1950, he published
“I, the Robot,” a collection of short stories on the subject. Isaac Asimov was a
believer in progress and saw in the robot mostly a helper in everyday tasks and an
assistant to help humans populate the universe, for example, in his short stories
“Reason” (“Reason,” 1941) and “Runaround” (1942).
156 11 Robotics

• Jack Williamson described in 1939 with “After World’s End” (“Beyond Space
and Time”) a humanoid robot that energetically turns against humanity. Jack
Williamson then began in 1947 with “The Humanoids” or “With Folded Hands”
and in 1952 with “Wing 4” a series of novels about humanoid robots that carry
out their task of serving and protecting humans a little too thoroughly, severely
limiting human freedom.
Philip K. Dick described non-humanoid and humanoid killer robots in his
1953 short story “Second Variety” (Variant Two, filmed as Screamers) and in his
1955 short story “Autofac” (Autofab or War of the Automata), a self-sustained
evolution of robot factories competing for raw materials into nanorobots unde-
sired by humans.
The Polish author Stanislaw Lem published the story cycle “Kyberiade”
starting in the 1950s. The main characters are Klapaucius and Trurl, two
humanlike robotic beings capable of thinking and feeling, who are themselves
constructors of machines and exist in a world populated predominantly by robots.
• Robot characters also found their way into children’s literature. In 1967, the
book “Robbi, Tobbi und das Fliewatüüt” was published, in which the robot ROB
344–66/IIIa and the inventor Tobias Findteisen experience adventures together.
Ellis Kaut published the children’s book “Schlupp vom grünen Stern” in
1974 about a little robot from the planet Balda 7/3, which unwantedly develops
a soul and feelings and is therefore supposed to be shot to a garbage planet,
but ends up on Earth instead. The Augsburger Puppenkiste filmed Schlupp as a
puppet play in 1986.
• In the novel “Machines Like Me” by British author Ian McEwan, published by
Diogenes Verlag in May 2019, a pair of young lovers enter into a tangled love
triangle with an android named Adam. In the novel “The Night Was Pale, the
Lights Flashed” by author Emma Braslavsky, published by Suhrkamp in August
2019, a robotics company sells humanlike robots as artificial life partners to
combat civilizational problems such as social loneliness and lack of relationships.
In the novel “Klara and the Sun,” a terminally ill girl makes friends with
a female android named Klara. The book, published by Blessing Verlag in
March 2021, was written by the author and Nobel Prize winner for literature
Kazuo Ishiguro. The novel is written from the first-person perspective of
the robot, as the android Klara sees the world of humans. The AI novel “The
Invention of Disobedience” (Unionsverlag) by Martina Clavadetscher,
which won the 2021 Swiss Book Prize, features a woman who works in a
sex doll factory in China where artificial female bodies are produced. For her
novel, the writer was inspired by the life’s work of the mathematician Ada
Lovelace, whom she sees as the foremother of programming and the computer.
The novel “Dave” by the writer Raphaela Edelbauer, which was awarded
the Austrian Book Prize 2021, is about a computer that scientists want to give a
human consciousness.
There is still a lot of scope here for the further creation of fantastic figures
(Fig. 11.3).
11 Robotics 157

Fig. 11.3 Star Trek

• The 1897 short film “Gugusse et l’Automate” by film pioneer Georges Méliès
was probably the first to show a robot on the cinema screen. The film, now
lost, was about a confrontation between the clown Gugusse and an “automaton.”
In 1911, the 10-minute short film “The Automatic Motorist” by Walter R.
Booth was released, featuring a robot chauffeur. In the episodic film “The
Master Mystery” released in 1919 with circus performer Harry Houdini,
there is another early example of the appearance of a robot. The 1921 Italian
film “L’uomo meccanico” by André Deed is about crooks who take control
of a remote-controlled robot in order to commit crimes with its help. Among
the most famous robot portrayals of the silent film era is the female machine
man in Fritz Lang’s “Metropolis” from 1927. The first robot to appear on
television was I Tobor in the science fiction series “Captain Video and His
Video Rangers” (from 1949).
• Well-known robot characters of the 1950s are the giant guardian Gort from the
film “The Day the Earth Stood Still” (1951) and the robot Robby from “Alarm
in Space” (1956). Robby subsequently appeared in a number of movies and
television shows and also served as the model for many toy robots. The Star Wars
saga (1977–2005) features robots in a comedic role with R2D2 and C3PO. The
two characters are at the same time characteristic examples of different types
of robots: while the “protocol droid” C3PO is a humanoid robot, R2D2 is an
“astromech droid” without humanlike appearance. In the series “2Star Trek: The
Next Generation” (1987–1994), the android Data is a command officer who often
grapples with the question of his own human-likeness. In “Number 5 Lives!”,
a military robot develops a naive personality. Films such as “Terminator” and
“I, Robot” offer other well-known examples of the portrayal of androids and
humanoid robots that are hostile to humans or assist them.
• Industrial robots (non-humanoid robots) are also finding their way onto the big
screen. KUKA industrial robots have been featured in films such as “James Bond
007: Die Another Day,” “Thunderbirds,” “Tomb Raider – The Cradle of Life,”
and “Sacrilege.” The documentary film “Plug & Pray” addresses the effects of
the increasing use of robots on the self-image of humans. It presents robots such
158 11 Robotics

as the Japanese Geminoid by Hiroshi Ishiguro, the Italian ICub, or the

German MuCar-3.
The 2012 feature film “Robot & Frank,” directed by Jake Schreier, tells of the
developing friendship between an elderly dementia patient and a caregiver robot.
In the Japanese city of Yokohama, a giant robot named Gundam, which stands
18 meters tall and weighs 25 tons, has been watching over the port area there as a
tourist attraction since 2020 after 6 years of development. The giant robot, which
has a cockpit inside and hands that are each 2 meters long, is based as a character
on a science fiction television series and can move and sink to its knees. The giant
robot was manufactured by the company “Gundam Factory Yokohama” under the
managing director Shin Sasaki.
• Probably the most prominent example of the use of the robot in the visual arts
is the group of “Family of Robots” by the Korean video artist Nam June
Paik. As early as 1964, Paik, together with the Japanese engineer Shuya
Abe, developed the robot K456, which was to represent Paik in performances
from then on—until K456 was involved in the first accident of the
twenty-first century. In the 1980s, the Family of Robots was
created, initially with “Mother” and “Father” (1983/86), which were expanded
by several figures. These were mainly figures from history or literary fiction,
such as Albert Einstein, Attila or Edgar Allan Poe, but also friends of Paik: John
Cage, Merce Cunningham, or Joseph Beuys.
SHIFZ is the abbreviated self-designation of the Austrian artists’ associ-
ation Syntharturalist Art Association. It was founded in 1996
and mainly focuses on the relationship between man and machine. The art
groups monochrom and Shifz organize events that deal artistically with
robot technology. They co-host Roboexotica, the festival of cocktail robotics.
monochrom hosts Arse Elektronika, the festival that has the creative use
of sex and technology as its core theme and presents artistic sex robots there.
The Japanese illustrator Hajime Sorayama draws female, chrome-shining
robots in erotic poses. For example, the American rock band Aerosmith chose a
drawing by Hajime Sorayama as the album cover for their 2001 album Just Push
Play. For their records Sign In Please (1984) and That’s the Stuff (1985),
the glam metal band Autograph also used motifs drawn by Sorayama.
The work of Swiss painter H. R. Giger, famous for his film design
for Ridley Scott’s Alien, is dominated by the concept he coined of
biomechanoids, a fusion of technology and mechanics with the creaturely
influences of fantastic realism.
From June 12 to July 6, 2019, The Barn Gallery of St. John’s College,
University of Oxford, exhibited various drawings and paintings of the female
robot Ai-Da under the title “Unsecured Futures.” The humanoid robot (drawing
robot artist) with the face of a woman, supported by a team of computer scientists,
robotics experts, and designers, was developed by the gallery owner Aidan
Meller. Ai-Da draws with chalk and paints in acrylic.
• The German electronic music band Kraftwerk reached number 18 in the German
charts in 1978 with their single “Die Roboter” from the album Die Mensch-
11.1 Robots in Surgery 159

Maschine. For their stage show, the musicians had special remote-controlled
robots developed with the faces of the band members replicated in plastic. In
1983, the song “Mr. Roboto” by the band Styx reached number 8 in Germany
and number 3 in the US charts. It is about a prison guarded by robots.
Music-making and dancing robots, under the direction of a mad scientist, can
be seen in the video clip for the 1984 track “Zoolook” by French musician
Jean-Michel Jarre. In Herbie Hancock’s video clip for his 1983
electronic instrumental Rockit, robotic legs dance without trunks, mechanical
dolls gather around a breakfast table, and a humanoid being twitches restlessly
under a bedspread. It was directed by the duo Godley & Creme, and the
puppets and robots were designed by artist Jim Whiting, who enjoyed great
success with his machine theater Unnatural Bodies beginning in 1988. The
instrumental Rockit is on Hancock’s album Future Shock.
• In October 2018, the writer Thomas Melle staged the play “Unheimliches
Tal/Uncanny Valley” at the Münchner Kammerspiele, in collaboration with
Stefan Kaegi from the theater group Rimini Protokoll, in which the author
converses about existential questions with a robot sitting on stage, a mechanical
replica of himself.
• Robots in sports As part of the 2020 Summer Olympics, automaker Toyota
unveiled Cue3, a basketball robot that has a 100 percent scoring rate.
You certainly can’t read, watch, or listen to it all here. But it shows that there are
no limits to imagination and that robotics technology is evolving in all directions.

11.1 Robots in Surgery

This is certainly one of the most interesting and important fields of application of
robotics, which is developing extraordinarily fast and can satisfy more and more
requirements.
An essential requirement for the introduction of new technologies is always
that the costs for the acquisition of such devices are covered and that appropriate
training opportunities are provided. The possibility of manufacturing such devices
should be available, as well as the IT specialists for programming and maintaining
these systems. It should also be pointed out that there are already courses of study:
“medical informatics” or “digital medicine.”
For example, the Faculty of Informatics at the University of Augsburg announces
the following program:
“The interprofessional degree program Medical Informatics (Medical
Information Sciences) builds on the bachelor’s degree program of the same name
and is located at the interface between informatics, medicine, and the application of
these two disciplines in the field of medical informatics.”
Medical informatics uses modern information technologies to provide organi-
zational and technical support for the healthcare system. To this end, it develops
160 11 Robotics

models and simulations for individualized therapy and diagnosis. Therefore, the
study program provides the necessary skills to develop, select, and apply suitable
methods, systems, and information technology tools for concrete medical issues.
At most universities, the course of study has a strong practical orientation in order
to optimally prepare students for their later work [1]. Important study contents of
medical informatics are:
• Medical documentation
• Imaging techniques such as computed tomography
• Scientific modeling
• Computer-assisted detection
• E-health
• Biosignal processing
• Biometrics
Depending on the university, the study of medical informatics also special-
izes for certain fields, for example, dental technology. The aim of dental
informatics is to develop IT applications to actively support the dentist during
treatment. Dental informatics is the application of computer and information sci-
ences to improve dental practice, research, education, and administration. Numerous
applications have been developed to support clinical care, education, and research.
Dental informatics is beginning to show the characteristics of a discipline: a core
literature exists, and trained specialists and educational programs are already in
place.
• Structure and content of the bachelor’s degree in medical informatics: The spe-
cialty enables graduates to independently develop IT—solutions in the medical
field—to realize imaging in connection with diagnostic procedures or to organize
the management of patients’ data. For this, the basics of computer science
are taught. These include programming, mathematics and statistics, algorithms
and data structures, or the development of databases. Planning skills for the
development of customized programs, software engineering, are also taught. The
basic education also includes fundamental knowledge from medicine, such as its
most important sub-areas, but also diagnostic and therapeutic systems.
• The bachelor study analytical and digital forensics: Tragic fate or crime? Natural
death or murder?—As early as the nineteenth century, Carl Remigius
Fresenius solved a sensational murder case that didn’t look like one at all
at first, by determining poisoning with arsenic. The founder of the Fresenius
University of Applied Sciences felt obliged to “stand up for law, truth, and
science”—and his work had a decisive influence on forensic analysis.
Today, the questions have become more diverse and analytics is also much
more diverse than it was back then: The modern analyst is not only familiar
with scientific basics, but is especially at home in the digital world. Whether
it’s designer drugs or cybercrime, environmental pollution or hacker attacks,
or genetic analysis or encryption technologies, modern analytics encompasses
11.2 Robots Help in Care 161

all steps from the right question to data collection, processing, and analysis to
interpreting the results in light of various issues.
Analytical thinking and action, embedded in a scientific approach, are still
the basis for fighting crime today. In the process, the real and digital worlds
are growing ever closer together and require investigators who are at home in
both worlds and can move between them. You can become that investigator:
Study Analytical and Digital Forensics (B.Sc.) at Fresenius University of Applied
Sciences—where forensics has been at home since 1848!
In [2], 28 study programs are displayed.
When you talk to interested people about this problem, you often meet with
skepticism or rejection. But this often disappears when you explain that a doctor
is always present. And of course, it is convincing if you then simply explain that the
robot does not get tired, that it can work 24 hours without a break, and that it does
not tremble, does not overlook anything, and much more.

11.2 Robots Help in Care

Science fiction movies often feature humanoid robots that help elderly people in
their daily lives. They move freely around the home, do the dishes, cook, clean,
and assist those in need of care with personal hygiene and can even carry on
conversations, much like human caregivers.
In these fictional scenarios, the impression is usually given that technological
progress will result in the loss of humanity and that at some point everything will
be done by robots. In fact, the exact opposite is true: digitization is bringing more
humanity back into nursing.
Apart from the repeatedly criticized poor pay for skilled workers, the biggest
problem in care today is the time factor. For older people in particular, interpersonal
relationships are enormously important. It’s one thing to be helped to take a shower
in the morning. But to be able to talk to the caregivers afterward about one’s
own worries and problems (or simply about the weather) is something completely
different. And it is precisely for this interpersonal communication that time is
usually lacking.
This problem can be significantly reduced by digitizing care. It starts with
simple things like electronic patient records, automated nursing documentation, or
networked route planning. It is precisely such administrative tasks that take up a lot
of time in practice. For people who need care but still live in their own homes, there
is also a lot of potential in telecare. After all, a 15-minute face time call is better
than no contact at all.
However, the idea is not for robots to replace human caregivers, but rather to
support them. This starts with simple robots that can clean the floor, for example.
The makers of the therapy seal PARO, which was specially developed for dementia
patients, are taking a completely different approach. The care robot Pepper, on
162 11 Robotics

the other hand, supports professionals in care facilities during day care. It can
tell jokes, read fairy tales, play music, demonstrate simple fitness exercises, etc.
Meanwhile, there are even solutions for private homes, such as Medisana’s Home
Care Robot, which enables video calls, regularly reminds people to check their
health, and can quickly summon help in an emergency.
“Movement disorders such as tremors, paralysis, or muscle tension disorders
affect many patients suffering from strokes or neurodegenerative diseases such
as Parkinson’s or multiple sclerosis. Recording the mobility restrictions precisely
and reproducibly is a major challenge in diagnostics and therapy monitoring,
as this requires experienced physicians who are not always available,” explains
Mathias Bähr, director of the Neurological Clinic at the University Medical Center
Göttingen.
This is where the “Deep Movement Diagnostics” project comes in. We will
combine our expertise in the fields of body and eye movements in humans and
monkeys, neurophysiology, and clinical neurology, as well as prosthetics and
robotics, says Alexander Gail.
Using the latest digital methods, walking and grasping movements are to be
measured and modeled with previously unfeasible precision in order to use them as
the basis for diagnostic tools for individualized therapy approaches, for example, in
Parkinson’s or stroke patients. In this context, motor function studies in monkeys
play an important role; they are the basis for later application in humans. “Our
goal is to develop a low-cost, easy-to-use system that can be used across the board
for diagnosis and therapy monitoring in movement disorders,” says project leader
Alexander Gail.
“Ease of use and low effort to perform the examinations are important factors
to increase patient acceptance,” said Melanie Wilke, director of the Institute of
Cognitive Neurology. “We expect the new video-based methods to provide a
qualitative leap over current clinical examination techniques.”
In addition to pure diagnostics, the research team plans to study complex
movement patterns in healthy subjects and in monkeys to better understand the
neurophysiological basis of movement disorders.

11.3 Other Interesting Applications

In the fall of 2018, scientists at the University of California in San Francisco

presented a pilot study with deep artificial neural networks in the journal Radiology,
which can detect Alzheimer’s disease on average 6 years before the final diagnosis
based on brain scans. Doctors often diagnose Alzheimer’s disease only when the
first symptoms appear. Even experienced doctors find it difficult to recognize and
correctly classify the small changes in the brain that occur in the early stages. That’s
why AI-assisted detection could make an important contribution to early detection
and thus treatment, he said. The network achieved a sensitivity of 100 percent, with
a correct-negative rate of 82 percent.
11.3 Other Interesting Applications 163

Fig. 11.4 A healthy lung and

a lung affected by cancer

In 2020, the Vienna-based AI lab Deep Insight released an artificial neural

network trained to classify whether a patient has COVID-19 based on CT scans
of the lungs, assuming the virus has already affected the lungs. The network
distinguishes between changes in the lung caused by COVID-19, other pathological
findings, and normal condition (Fig. 11.4).
In ophthalmology, AI-based approaches have been shown to outperform special-
ists for a variety of clinically relevant tasks. These include recognizing diagnoses
based on retinal imaging, calculating ideal lens power prior to cataract surgery,
and recognizing systemic risk factors (e.g., blood pressure) based on ophthalmic
imaging. In a head-to-head comparison between deep learning and 13 human
specialists, 25,326 fundus photographs (photos of the back of the eye) of diabetic
patients were subjected to a diagnostic evaluation of diabetic retinopathy. The
sensitivity of the neural network was 0.97 (0.74 for specialists) and the specificity
was 0.96 (0.98). The project was recognized as a demonstration of outstanding
human performance in the 2019 AI Index Report as a milestone. Interestingly, it
was shown that even gender can be identified from color fundus photos using AI.
This was previously considered impossible among specialists. Follow-up work was
able to show that AI applications can identify gender from subtle differences in the
course of vessels.
In cardiology, algorithms are in use that evaluate long-term ECGs for a physician
and find and list important arrhythmias within seconds. For the medical layperson,
there are low-threshold mobile applications with algorithms that can measure
pulse rate using photoplethysmography (similar to pulse oximetry). Arrhythmias,
164 11 Robotics

especially atrial fibrillation, are significant here because they increase the risk of
stroke.
Several projects show that AI is not always about whether the machine is better
than the human, but that the amount of work involved in diagnosing by doctors can
be reduced by almost 20 percent. For example, in August 2016 at the University of
Tokyo Medical Institute, the IBM Watson computer program was able to correct
a misdiagnosis made by doctors. The doctors diagnosed the patient with acute
myeloid leukemia. The therapy was unsuccessful, so they asked Watson for help. It
took the AI 10 minutes to match the woman’s DNA with 20 million cancer studies.
Watson identified a very rare form of leukemia that has only affected 41 patients and
is curable. However, IBM Watson’s treatment recommendations can also be flawed,
such as when too little training data is available. Corresponding reports of faulty
recommendations, the application of which put patients at risk, were published by a
medical portal in 2018. The malfunction is said to have been fixed in a later version,
according to IBM. In January 2018, scientists at Stanford University presented a
program that can calculate with a probability of 90 percent whether terminally
ill patients will die within the next 3 to 12 months from their disease data. This
could help terminally ill patients live out their final months with dignity, without
aggressive treatment methods, and possibly at home under palliative care, it said.
The wristwatch Apple Watch records a person’s heart rate, among other
things. Apple announced that programs have an 85 percent probability of detecting
diabetes mellitus in the wearer of the wristwatch from the analysis of heart
rate. The idea is based on the Framingham Heart Study, which recognized back in
2015 that diabetes can be diagnosed using heart rate alone. Apple had previously
succeeded in detecting abnormal heart rhythm from heart rate with 97 percent
probability, sleep apnea with 90 percent, and hypertension (high blood pressure)
with 82 percent.
Researchers at Mount Sinai School of Medicine demonstrated in
January 2018 how psychological interview transcripts with adolescents can tell
if they will develop psychosis in the next 2 years. Natural language processing
helped achieve up to 83 percent accuracy on standardized tests, such as those based
on disorganized thought processes, awkward phrasing, unclear associations, or
reduced language complexity. The subtle differences were apparent after training
with many such conversations, he said.
MIT researchers unveiled a program in September 2018 that can diagnose
depression in patients based on spoken text or written text. By themselves, doctors
and psychologists ask patients questions about lifestyle habits, behaviors, and sen-
sitivities to diagnose depression from answers. After training with such interviews,
the program also detected depression from everyday conversations with a hit rate of
83 percent—and in classifying the severity of depression on a scale of 0 to 27 with
a hit rate of 71 percent. The AI could assist doctors or permanently monitor them as
an app user to alert in an emergency. The researchers also want to detect dementia
from speech in the future.
According to the manufacturer, the health app Babylon Health is supposed
to be able to use a voice system (chatbot) based on AI to make a diagnosis in
11.3 Other Interesting Applications 165

conversation with patients that is about ten times more accurate than diagnoses made
by a GP. The development of the app was also co-funded by the British healthcare
system. The aim was to use it to reduce costs. Although the app is supposed to
significantly reduce doctor visits, patients quickly found out how to use the app to
get doctor appointments faster by describing symptoms incorrectly.
The app Ada from the Berlin-based company Ada Health uses a program
to assist with diagnoses based on symptom descriptions. According to the manu-
facturer, this should correspond to the quality of well-trained Western doctors. The
app Ada sent unauthorized marketing companies such as Amplitude and Adjust,
headquartered in San Francisco (USA), and in the course of app use regularly
Facebook.com personal data, even if you do not have a Facebook account.
The app has won awards from MIT and is funded by the Bill & Melinda
Gates Foundation. In 2019, Ada Health announced a collaboration with
Sutter Health. Especially in developing countries where there is a shortage
of medical personnel, the app can help build a health system.
Devices or software that use artificial intelligence must be CE marked as medical
devices in Europe and approved by the FDA in the USA. A comparison between
the USA and Europe from 2015 to 2020 yielded, a rapidly increasing number of
approvals, with CE marking slightly dominating in number (2019: US, 79; EU, 100).
Often, CE marking occurs before FDA approval, which may indicate a less rigorous
process. The emphasis is on radiology. Only 15 % of the products are aimed directly
at individuals (patients), the rest at professionals (physicians). During this period,
only 1 % of approvals are for the highest risk classes, e.g., those for the diagnosis of
breast cancer [46].
Even more so in Europe than in America, the authors of the study find a lack of
transparency in the device description and the process of evaluation. It reflects the
ethical responsibility of regulators as much as manufacturers. A publicly accessible
database of CE-marked devices and software is also urged.
Devices or software that use artificial intelligence must be CE marked as medical
devices in Europe and approved by the FDA in the US. A comparison between
the U.S. and Europe from 2015 to 2020 yielded, a rapidly increasing number of
approvals, with CE marking slightly dominating in number (2019 U.S. 79, EU 100).
Often, CE marking occurs before FDA approval, which may indicate a less rigorous
process. The emphasis is on radiology. Only 15 % of the products are aimed directly
at individuals (patients), the rest at professionals (physicians). During this period,
only 1 % of approvals are for the highest risk classes, e.g. those for the diagnosis of
breast cancer [46].
Even more so in Europe than in America, the authors of the study find a lack of
transparency in the device description and the process of evaluation. It reflects the
ethical responsibility of regulators as much as manufacturers. A publicly accessible
database of CE-marked devices and software is also urged.
The user wants to understand the reasons for an algorithmic decision. In artificial
intelligence and machine learning, algorithms are often completely opaque (“black
box”), typically in neural networks and the corresponding learning methods. To
counter this, the field of Explainable Artificial Intelligence is
166 11 Robotics

developed. In healthcare, trusted model developers, extensive external validation

via studies, and standardized evaluation procedures are discussed.
Boris Babic of the University of Toronto objects that Explainable Artificial Intel-
ligence would often only provide post hoc explanations for black-box algorithms.
These explanations, he said, are not reliable and could mislead the user. True white-
box algorithms that actually provide comprehensible explanations, on the other
hand, are clearly limited in their complexity and therefore hardly suitable for many
use cases. He therefore believes it is a mistake to prescribe the explainability of
algorithms, for example, as a requirement for approval, since this offers hardly any
advantages, but instead slows down innovations and leads to the use of algorithms
with lower accuracy. Instead, algorithms should be tested more in clinical trials to
ensure their effectiveness and safety.
In pharmaceutical research, automated high-throughput screening has estab-
lished itself as a method of finding so-called hits and thus candidates for lead
structures. British researchers at the University of Cambridge developed the automa-
tion further. The research robot Eve, which was presented in the Journal of the
Royal Society Interface in 2015, uses statistical models and machine learning to
produce and test assumptions, test observations, perform experiments, interpret
results, change hypotheses, and repeat this over and over again. This, they say,
allows the robot to predict promising compounds, making the process of finding
lead structures more efficient. Using this robot, the researchers found in 2018 that
triclosan, which is also used in toothpaste, could fight malaria infections at two
critical stages, infestation of the liver and blood. With the discovery by AI, a new
drug could now be developed.
The new Sars-CoV-2 coronavirus is now also occupying the attention of artificial
intelligence. It will help detect infections with the virus on X-ray images of
the lungs. At least that’s what radiologist Haibo Xu from Zhongnan Hospital in
Wuhan—the city where the virus probably first jumped to humans—is convinced
of. Now a team wants to test whether and how the software can help medical
staff diagnose new cases. The virus is new, but the idea is not: last year, German
scientists developed an algorithm that could detect black skin cancer better than
doctors in a direct comparison. There were similar results in the USA back in 2017.
Other studies successfully applied artificial intelligence to analyze tissue samples
for breast cancer and X-rays of pneumonia. Black skin cancer is also known as
malignant melanoma. Too much sun or ultraviolet (UV) light can increase the risk
of black skin cancer. People with light skin or a particularly large number of moles
have an increased risk of black skin cancer (Fig. 11.5).
References 167

Fig. 11.5 Black skin cancer

References

1. https://www.studycheck.de/studium/Medizinische-informatik
2. https://www.studieren-studium.com/studium/Medizinische-Informatik
Chapter 12
Artificial Intelligence in Finance

Unlike physics and many engineering sciences, there is no closed mathematical

theory for finance from which one can derive all rules and laws. Formulas and
equations exist only for small, singular tasks [1]. One has immense amounts of data
and tries to derive future favorable or unfavorable developments from them. The
digitization of all information available for the financial world and its immediate
availability via the Internet dominate as the basis for all decisions.
For stocks, for example, this includes the DAX, the Dow Jones, the TecDAX,
and the Nasdaq. For commodities, prices are given for gold, silver, oil, and natural
gas. The https://www1.oanda.com/currency/converter/ website shows the rates for
all currencies that exist and may be traded. Then there is a heading HotStuff;
this table lists the most searched stocks of the day with their gains and losses. Even
the title bar shows the breadth of information: Stocks, News, Indices, Funds, ETFs,
Certificates, Leverage Products, Bonds, Commodities, Cryptocurrencies, and Forex.
Around the world, artificial intelligence is seen as a promising innovation for the
financial sector. Here, one is certainly a bit more reserved with publications, because
of course an effective system based on neural networks is very advantageous in the
industry. In any case, neural networks and the systems that can be implemented with
them will play a major role, especially because of the amount of data that will be
available for many years. This will also help prevent fraud and automate resource-
intensive, repetitive processes and customer services without compromising quality:
banks and insurers from the DACH region have recognized the potential of artificial
intelligence in times of digitalization, but are not yet exploiting it across the board.
Although the majority of respondents (62%) consider artificial intelligence to be
an important innovation that will gain in importance in the financial sector over
the next 5 years, there is still a clear difference between the vision and the actual
situation. Currently, only nine percent of decision-makers see their company as
digitally very well prepared for the use of AI technologies. Banks and insurers are
only just beginning to look at specific areas of application for this rapidly advancing
technology. There is a wide gap between expectations and implementation. There

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 169
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_12
170 12 Artificial Intelligence in Finance

are many AI pilot projects at financial companies—but only very few of them
succeed in transferring these ideas into daily, operational business. Even companies
that already have AI expertise themselves often do not know how to approach the
topic in a meaningful way [2].
Currently, insurers and banks in the DACH region (Germany, Austria, Switzer-
land) are looking at the possible use of AI solutions primarily from a conventional
business perspective: 79% of those surveyed want to make business processes more
efficient digitally, almost three quarters generally want to save costs (73%), and
every second company expects to use AI to ensure compliance with all existing
regulations (50%). But also for new fields, such as chatbots, automation, and
predictive marketing, just over half of the respondents (55%) are already using the
new technology. However, many other opportunities remain untapped: for example,
the complexity of risk assessments and decision support in controlling can be greatly
reduced through automation if the underlying data is really analyzed intelligently.
In order to catch up with the European average, many financial companies are
currently examining which new projects are particularly suitable for the use of AI.
But even once these have been identified and defined, the road to implementation
is often still long: 69% of the companies surveyed identify the lack of available
data as an obstacle to adaptation. A good two-thirds of the companies surveyed are
also struggling with budget restrictions and insufficient funding for corresponding
projects (67%), and 64% of the companies simply lack employees with expertise to
answer questions about establishing AI:
• Which business unit provides an appropriate entry point for establishing AI
projects in operations?
• Which division provides funding for the integration process?
• Are AI-related projects to be regarded as a subdivision of IT or, as a strategically
relevant topic, dependent on independent management structures? [2]
Data is more than abundant in some subfields for almost 100 years. The problem
here is to get it all into one system, in a reasonable amount of time, and with a
reasonable amount of effort. Whether a body can be created to work across different
banks on these problems is not clear at this time. Normally, the various banks
are in a competitive situation. However, cooperation would clarify the problem of
resources. Independent management structures are recommended in any case, since
the IT departments are already busy as it is with all the problems of online banking.
However, if a large bank succeeds in overcoming these problems, it would have a
major advantage over its competitors.
The study also shows that using AI for day-to-day business and in established
processes, for example, with regard to personalization or new business models,
has so far been a comparatively low priority for financial service providers. In
addition, the often still low general comprehensibility in the financial sector is a
hurdle that should not be underestimated. Classic mathematical applications can
still be mapped with a comparatively simple algorithm, but closed models such as
deep neural networks are significantly more demanding and thus more difficult to
penetrate. This results in another problem: the financial services sector is highly
12 Artificial Intelligence in Finance 171

regulated. Companies are obliged to explain their processes and decisions in detail
to supervisory authorities and internal auditors. Here, AI is still considered a black-
box technology, so many companies act with restraint here, and not just in banking.
Another big area that could be revolutionized by AI is risk assessment [3]. This
is hugely important in many areas of finance. AI systems can draw on a wide range
of information to calculate the risk of default on loans. Algorithms can learn from
cases of credit decisions and customers’ repayment behaviors, identify patterns from
them, and create profiles. When a new credit request is made, the system compares
the customer’s data, checks for patterns, and ranks his or her creditworthiness
based on the analysis. The check can be used to make lending decisions. Machine
verification not only provides insights and greater certainty. It can also be used as
an argument to reject a loan. The final decision remains with the lender. However,
the lender is given the opportunity to make a quick and data-based decision.
Artificial intelligence can also be a powerful tool in market research. News from
all over the world, media activities, and studies can be evaluated to make forecasts
and identify investment trends at an early stage. With the help of such analysis
tools, geopolitical events can be taken into account and the stability of markets
can be estimated. Valuations for securities can be created from the analyses and
made available to employees for their decisions. Powerful systems can make this
assessment in real time, which holds great potential for high-frequency trading on
the stock market.
Know Your Customer programs can use artificial intelligence to screen
new customers. The algorithms detect suspicious patterns and activity and can alert
to money laundering and white-collar crime. Pre-screening customers is a critical
step for many companies and institutions in the financial sector. Therefore, a lot of
time and energy is invested in research and verification. Another example is claims
settlement in the insurance business. AI systems can be trained with data from a
large number of insurance cases, especially fraud cases. Conspicuous cases can be
flagged and presented to employees for further review.
On June 15, 2021, the German Federal Financial Supervisory Authority pub-
lished supervisory principles for the use of algorithms in decision-making processes
of financial companies. They are intended to lead to responsible use of Big Data
and Artificial Intelligence (BDAI) and to enable the associated risk to be controlled.
Background to the principles paper: technologies such as BDAI are increasingly
also being used by companies in the financial market. In 2018, BaFin had already
pointed out in its study “Big Data meets Artificial Intelligence” that this would
result in opportunities for companies, but also for consumers, but that it was also
important to manage the risks that BDAI applications would bring [4]. In this regard,
a fundamental problem underlies all of the regulatory issues surrounding BDAIs:
it is still difficult to distinguish BDAI procedures from procedures of classical
statistics. From a risk perspective, however, three characteristics can be identified
that are of particular importance in modern BDAI methods:
• First, the algorithms used are often particularly complex compared to classical
statistical methods. This complicates or prevents their traceability.
172 12 Artificial Intelligence in Finance

• Second, shorter and shorter recalibration cycles can be observed. This is due
to the combination of constantly learning algorithms and the fact that new data
is available almost daily. As a result, the boundaries between calibration and
validation are becoming increasingly blurred. Third, the use of BDAI methods
increases the degree of automation. This makes it increasingly easy to scale up
processes, and the impact of the individual algorithm increases.
In order to be able to formulate the principles as precisely as possible, the decision-
making process based on algorithms has been divided into two phases in a highly
simplified manner: the development phase and the application phase.
The development phase is about how the algorithm is selected, calibrated, and
validated. For this, there are principles for data strategy and for documenting
internal and external traceability. In the application phase, the results of the
algorithm must be interpreted and integrated into decision-making processes. This
can be done automatically, but also by involving experts on an ongoing basis. In
any case, functioning rules of procedure must be established, including sufficient
control mechanisms and appropriate feedback to the development phase. These two
phases are flanked by overarching principles, for example, on the need for a clear
responsibility structure and adequate risk and outsourcing management (Fig. 12.1).
Here you can see the transition to the theory of fractal systems. The DAX curves
are self-similar. The curves always look the same, no matter whether the values are
plotted daily, weekly, monthly, or annually. This transfer of the theory of fractals
to finance goes back to Benoît Mandelbrot (Fig. 12.2). In a founding series for
the renowned journal Quantitative Finance, he summarized his view of markets.
Here he described a fractal recursive construction process for price time series that
recursively decomposes trends into smaller trends (Fig. 12.1).

Fig. 12.1 This information regarding the DAX is offered in real time
12 Artificial Intelligence in Finance 173

Fig. 12.2 Benoît Mandelbrot

(1924–2010)

Answering the question of what a trend is does not seem difficult. Mathemati-
cally, however, it is not so simple. If you want to define trends, you have to specify
two things: the exact measurement methodology and the scale used to measure.
The moving average (also moving average) is a method of smoothing time or
data series. The smoothing is done by removing higher-frequency components. As
a result, a new data point set is created consisting of the averages of equally sized
subsets of the original data point set:

1
n−1
m(t) =
. x(t − i).
n
i=0

For many values from everyday stock market life, the values .n = 38 are chosen
for 1 month and .n = 200 for half a year. Strong swings of the values up and down
can compensate each other, if they occur equally often and have approximately the
same size. The structured monthly investment in stocks that have historically risen
in price hereby appears clear. Obviously, this effect exploits overlong trends in real
data [5].
Analogously, stock prices may also be assumed to be fractal. In 1968, Mandel-
brot, together with van Ness, generalizes the model of a random walk, which is
fundamental for financial mathematics, in which he introduces a further parameter,
the so-called Hurst exponent, in addition to the growth of returns (drift) and their
fluctuation (volatility). This parameter can be used to describe the scaling behavior
of prices. Here, values <0.5 can be interpreted as mean reversion, while values
>0.5 can be understood as trending. Only in the limiting case, where the Hurst
exponent is equal to 0.5, one may speak of efficient markets. In this case,
the time series would be statistically trend-free, i.e., it would have no relation to the
past.
174 12 Artificial Intelligence in Finance

Fig. 12.3 Average trend lengths

Toward the end of his career, Mandelbrot outlines a construction principle for
trends that follows the construction principle of fractals. Starting from a main trend,
he applies a subdivision rule that divides this main trend into three smaller trends. In
a next step, these smaller segments are again further decomposed into even smaller
trends. The whole leads to a fractal in which trends can be described by smaller
trends (Fig. 12.3).
In an article [6], it is shown that well-known factor strategies, such as momentum
(strongly rising stocks), low volatility (stocks that fluctuate weakly), and
value (stocks that are considered favorable due to economic key figures), can
be described and replaced by systematic trend-based strategies. For this purpose,
only the last visible trend is evaluated for all stocks and then sorted by slope. The
differences in the strategies can be described by different wavelet scales, which are
used to calculate the last visible trends.
This view of investment styles (factor investments) is also almost complete,
i.e., one is able to describe market returns with high statistical quality by these
investment styles. These styles are additionally dominant in different market
regimes, i.e., they generate excess returns in different phases. For example, investing
in defensive stocks is attractive when the market is in a down phase, while value
often generates very high returns when markets are considered cheap and market
participants expect prices to recover. With momentum, on the other hand, corporate
earnings revisions play a dominant (but not exclusive) role.
• The rise of artificial intelligence requires new thinking and new content for
finance education. AI experts are already a shortage; firms in finance are com-
12 Artificial Intelligence in Finance 175

peting with technology-based firms, startups, and other fields. The application of
efficient algorithms leads to profits that are higher than the general market rate:
• Algorithmic credit scoring can support or fully automate credit decisions.
• Algorithms can detect unusual patterns, helping to identify fraud or even prevent
it.
• Algorithms can be trained to optimally execute derivative trading.
• Algorithms can be used to design and manage funds intended for retirement.
The algorithms work better and more effectively than humans due to the
following factors:
• It is possible to cope with large amounts of data.
• The instability of financial markets does not follow constant laws.
• Many relations that occur are nonlinear. Tiny changes in the initial data can lead
to significant changes in the market.
• The high dimensionality of many problems can be handled by programs.
• Traditional econometrics dealt primarily with regression problems. Classification
problems, however, can be much better handled by learning procedures.
• Artificial intelligence learning techniques can handle structured and unstructured
data simultaneously.
A major advancement in hardware has been the development of graphics
processing units (GPU). GPUs handle the calculations for 2D and 3D graphics. This
relieves the CPU (central processing unit). This takes care of all computing tasks that
have nothing to do with graphics. Compared to the GPU and CPU, a GPU is usually
built on a lot of computing cores to be able to handle many tasks at the same time.
The calculation of single pixels is not very complex in comparison, but millions of
pixels have to be calculated at the same time. The GPU is either installed directly
on the CPU, on the motherboard, or as a plug-in card for the computer. Several
GPUs can also be installed on a plug-in card—for example, the PowerColor Radeon
R9. Today, three major manufacturers share the market for the production of GPUs:
Intel, AMD, and Nvidia.
For the future, three possibilities are mentioned.
• The monopoly: a financial institution achieves a dominant position because it
is the first to be able to apply high-quality AI methods first.
• The oligopoly: a market form characterized by few market participants.
People are currently working hard on the following problems:
• Privacy: the large-scale application of AI methods requires the use of private
data. This brings the risk of data being stolen or used improperly.
• Bias: the algorithms may take on these characteristics if it is already present in
the data.
• Explainability of the results is a major problem. In some cases, it is required
by investors; in others it is required by law. Although this area is being studied
intensively, no progress is currently seen.
176 12 Artificial Intelligence in Finance

• Habituation effect: this effect can occur when many do the same thing. In this
regard, one can study the materials on the 2008 financial crisis. A similar situation
can occur when many companies use the same or similar programs.
• Jurisprudence and financial policy must keep up with these developments and not
lag behind them.

References

1. Yves Hilpisch. Artificial Intelligence in Finance – A Python-Based Guide, O’Reilly, Beijing –

Boston – Farnham – Sebastopol – Tokyo 2021, 2021, 978-1-492-05543-3
2. https://www.pwc.de/de/finanzdienstleistungen/kuenstliche-intelligenz-im-finanzsektor.html
3. https://blog.fintechcube.com/ki-im-finanzwesen
4. Big Data und künstliche Intelligenz: Neues Prinzipienpapier der BaFin, https://www.bafin.
de/SharedDocs/Veroeffentlichungen/DE/Meldung/2021/meldung 210615 Prinzipienpapier BD
KI.html Ba Fin, Bundesanstalt für Finanzdienstleistungsaufsicht, Juni 2021
5. https://mandelbrot.de/fraktale maerkte i ineffiziente maerkte mandelbrots alternativ sicht, Teil I,
Mandelbrot Asset Management GmbH, 2019
6. https://mandelbrot.de/fraktale maerkte ii ineffiziente maerkte mandelbrots alternativ sicht, Teil
II, Mandelbrot Asset Management GmbH, 2019
Chapter 13
Artificial Intelligence and Education

13.1 Education of AI Professionals

The main problem in creating intelligent systems is the interdisciplinary collabora-

tion of institutions to achieve a critical mass of resources and knowledge. Putting
hardware aside, there remains the problem of having enough experts in a particular
area where one wants to introduce intelligent systems to model and program the
problems of the area in a way that is appropriate for computer science. In addition,
the introduction of such systems must be carried out very prudently. Very quickly,
one gets into the situation where the old system is no longer there, but the new one
is not yet working. This can be a loss-making situation.
The DFKI takes a good position here [2]. For example, they offer the following
certificate:
Artificial intelligence (AI), machine learning and autonomous systems will have a decisive
influence on the development of companies in the coming years. What does this mean for
your own company? Can AI be used to support the future value creation of a company?
Which AI concepts, which fundamentals, methods and concrete applications are hidden
behind the often strained buzzwords? Which fundamentals do you need to master, which
research institutes and companies can you work with in the field of artificial intelligence, and
where are there already experiential examples of good practice? Answers to these questions
are provided by the so far unique certificate course “Training as AI Manager” in Berlin -
the first training in the field of artificial intelligence with a certificate.

The practice-oriented certificate course is conducted by leading experts from the

German Research Center for Artificial Intelligence DFKI, SAP, IBM Watson, and
selected partner institutions.
The course is conducted using a digital learning environment. Participants can set
thematic priorities and discuss and work on these with the lecturers in the respective
modules. Knowledge transfer is rounded off and ensured through versatile practical
exercises and group work. Participants are thus enabled to take the knowledge

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 177
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_13
178 13 Artificial Intelligence and Education

they have acquired and apply it in their own day-to-day business. Upon successful
completion of the examination, participants receive a Bitkom personal certificate.
The part-time certificate course “Training as AI Manager” consists of five mod-
ules in which participants are taught the fundamentals of artificial intelligence and
hybrid value creation through heuristics, knowledge-based systems, and machine
learning. Application examples on artificial intelligence and practical experiences
from research and development as well as from companies and startups are an
integral part of the training. The participants test AI services on the Internet, reflect
on them against the background of their own business fields, and are enabled to
design and critically reflect on use cases for generating AI-based hybrid value
creation for their own company.
The five modules of the AI Manager training are the following:
• What is Artificial Intelligence: history, examples, society, ethics (2 days).
• Basics and prerequisites of AI: methods, tools, hands-on, machine learning, and
data mining (2 days)
• Hybrid value creation through AI: fundamentals and methods. Applications
Industry 4.0 and Smart Services (1 day)
• Artificial Intelligence in Research and (Business) Application. Cognitive AR,
Retail Solutions, Watson, Leonardo (2 days)
• Development of business areas, change of corporate culture, and design of
transformation processes (1 day)
The Chamber of Industry and Commerce (IHK) has a range of offerings for
technical training and continuing education available through the IHK Academy
in Munich and Upper Bavaria, as well as through the various regional locations—
including AI and topics related to data handling. The Stuttgart Region IHK has
already invited participants to a number of work camps and other events on AI in
2019.
The Rhine-Neckar Chamber of Commerce and Industry has addressed AI and
analytical platforms as well as “From Big Data to Smart Data.” The Artificial
Intelligence Working Group (AKKI) of the Lübeck Chamber of Commerce and
Industry has held several lecture series on the topic together with the local university
and MEVIS. A look at the respective further education and training measures of the
local IHK can therefore also be worthwhile for smaller companies.
If you have a higher-level qualification for the relevant area of the company—
a degree in business administration, for example—a full-fledged computer science
degree with a specialization in AI is hardly worthwhile, but it might be for one
or two junior staff members in the IT department. WirtschaftsWoche has compiled
a list of German universities offering such courses. Berlin’s Beuth University of
Applied Sciences even advertises its new Humanoid Robotics course on YouTube.
The Technical University of Munich (TUM) recently introduced the interdisci-
plinary master’s program “Robotics, Cognition, Intelligence.” Bergische Universität
Wuppertal has launched an “Interdisciplinary Center Machine Learning and Data
Analytics,” or IZMD for short. Bachelor’s degree programs with an AI focus are
available in Stuttgart, Marburg, Göttingen, Karlsruhe, and Deggendorf and at the
13.2 The Application of Artificial Intelligence in Education 179

Beuth University of Applied Sciences in Berlin. Master’s degree programs are avail-
able at LMU Munich, Darmstadt University of Applied Sciences; the universities of
Marburg, Potsdam, and Bielefeld; Chemnitz University of Technology; and Stuttgart
Media University.
Seminars and support in the area of AI and data science are also offered by
consulting firms such as All for One Group, headquartered in Filderstadt
near Stuttgart. The leading IT and consulting group in the DACH region includes
a number of subsidiaries and branches, including B4B Solutions and Allfoye Man-
agementberatung GmbH, which support corporate customers in terms of change
management in order to master the challenges of digital transformation.

13.2 The Application of Artificial Intelligence in Education

This area is hotly debated and extensive. Only a few examples will be given that
relate directly to education in a particular subject.
3D printing of body parts is ushering in a new era in anatomy education.
Especially in countries where exhumation of cadavers is a problem, the 3D printer
can fundamentally change medical education. The first product for this application is
likely to be Monash University’s 3D Printed Anatomy Series’ [1]. The series of body
parts created with a 3D printer works without body parts from deceased people, but
still provides all the important parts of a body needed for scientific teachings in
anatomy (tendons, ligaments, muscles, bones, etc.).
According to Professor Paul McMenamin, director of the Centre for Human
Anatomy Education at Monash University, the use of the low-cost anatomy kit
will improve the knowledge of future doctors and other medical professionals. He
says the technology could even lead to the development of new surgical treatment
procedures.
The anatomy kit is likely to generate increased interest, particularly in developing
countries. There, the handling of cadavers is often problematic for religious or
cultural reasons. The 3D-printed body models in the anatomy kit are quick and
relatively inexpensive to produce. In addition, they are not subject to constant
change. This means that with the acquisition of a 3D printer and the digital CAD
files, new models can be produced at any time worldwide (Fig. 13.1).
In a first step, real body parts are scanned by computer tomograph or a surface
laser scanner. Then the body parts are printed out of a plaster-like powder or plastic
with lifelike colors in higher resolution.
The combination of artificial intelligence and 3D printing can also help expand
the range of compatible materials to meet the needs of industries such as aerospace,
which mostly require high-temperature materials. The “futureAM” project launched
by the Fraunhofer Institute IWS in 2017 is a perfect example of this. Professor
Dr. Ing. Frank Brückner, head of the business unit “Generate and Print” at the
Fraunhofer IWS and AMCD (Additive Manufacturing Center Dresden) explains:
180 13 Artificial Intelligence and Education

Fig. 13.1 The 3D replica of

a hand

Fig. 13.2 Complicated shapes do not cause difficulties

“Aircraft engines could operate much more efficiently at higher temperatures if most
materials did not fail at temperatures above 1,200 degrees.”
So where does AI come into play here? Prof. Dr. Ing. Frank Brückner explains
this to us as follows: “The processing of new high-performance materials is very
complex and requires fine-tuning of all process parameters. That is why we monitor
the 3D printing process with a large number of different sensors. Using AI, we then
evaluate this flood of data and identify hidden correlations that are not recognizable
to humans.” This is precisely the advantage of artificial intelligence: it can process
many times more data than humans—and all of it much faster. Thanks to this
work, researchers can process complex alloys and obtain the exact properties of
the materials.
AI can also help improve the 3D printing process. For example, the printability
of an object can be analyzed before it is printed to ensure that it is suitable for 3D
printing. But also quality prediction as well as quality control of the printing process
can avoid errors in the final product and thus lead to a better result (Fig. 13.2).
The startup Printsyst is trying to implement this with the help of their
AI engine—in this case specifically in the aerospace industry. The goal of the
company’s patented AI algorithm is to identify part functionality and improve
References 181

printing success rates. In doing so, Printsyst’s integrated solution basically learns
from experience gained from previous projects in which objects were additively
manufactured. To meet the very stringent industry specifications that apply in
aerospace, it suggests specific printing parameters that have been proven in the
past on previously printed parts and are highly likely to deliver a successful
first print in the future. Also based on accumulated experience data, Printsyst’s
solution can accurately estimate the cost of required components, increasing not
only productivity but also delivery readiness, at the lowest possible cost.
Eitan Yona, co-founder of Printsyst explains with an example: “We reduce the
average 3D print preparation process from 30 minutes to 5 seconds for each job.
By reducing this time, we increase printer utilization, and by eliminating errors, we
reduce iterations.” So we see that due to the high complexity that a manual process
can bring, the use of AI makes perfect sense. For example, all parameters can be set
and optimized by the algorithm in a matter of seconds—significantly better results
can thus be achieved with the help of AI compared to manual processing.
Another field of application in the area of quality optimization is the printability
analysis (part evaluation) already mentioned above, which takes place before
printing. The company AMFG, mentioned above, has implemented a feature in their
comprehensive software package that ensures that a 3D-printed part is ultimately
really suitable for 3D printing. This is especially important for companies that
work efficiently with additive manufacturing. Here, too, previously obtained data
is processed and thus the printability of a part can be predicted. For example, if the
part has many weak points in terms of stability, shape, or even resilience, preventing
printing can save enormous costs.
There are many smaller systems for training in various subjects (chemistry,
physics, biology, foreign languages). These are essentially used in self-study by
individuals. There is often a lack of general application in schools from grade 1 to
grade 12 or 13. There is an urgent need to work out appropriate didactics and bring
the systems into schools.

References

1. https://www.3d-grenzenlos.de/magazin/zukunft-visionen/3d-druck-anatomie-ausbildung-
chance-dritte-welt-2755963
2. https://www.bitkom-akademie.de/lehrgang/ausbildung-zum-ki-manager
Chapter 14
Artificial Intelligence in the Arts

The application of intelligent software to painting and drawing has already been dis-
cussed in detail in the chapter on image processing. Therefore, only the applications
of artificial intelligence methods in music and architecture will be discussed in this
chapter.

14.1 Artificial Intelligence and Music

Music is the most vivid and sensual of all human arts. So it’s no wonder that
computer science tried very early on to use machines to create sounds or teach
artificial intelligences to compose melodies. The use of computers and algorithms
to produce music has a long history.
As artificial intelligence and machine learning continue to evolve, so do the
capabilities of machine-generated music. Computers can derive rules and guidelines
from real, human music examples.
The Illiac Suite, later renamed String Quartet No. 4, is a 1957 composition that is
considered the first score composed by an electronic computer. It was programmed
by researcher Lejaren Hiller in collaboration with Leonard Isaacson,
both professors at the University of Illinois. They used the ILLIAC I (Illinois
Automatic Computer) computer for this purpose.
The Illiac Suite consists of four movements corresponding to four experiments:
the first is about the generation of Cantus Firmi; the second generates four-part
segments with different rules; the third deals with rhythm, dynamics, and playing
instructions; and the fourth with different models and probabilities for generative
grammars or Markov chains. On paper, the Illiac Suite was a real masterpiece, but
in reality the piece sounds very tortured and not quite fully developed.
• In 1965, inventor Ray Kurzweil premiered a piano piece created by a
computer that was able to recognize and analyze patterns in various compositions

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 183
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_14
184 14 Artificial Intelligence in the Arts

and create new melodies. This computer was first used on Steve Allen’s CBS
game show “I’ve Got a Secret.”
• In 1980, professor and composer David Cope at the University of California,
developed a system called EMI (Experiments in Musical Intelligence). It was
based on the idea of Markov chains, that is, on giving probabilities for the
occurrence of future events. iCope used a program to analyze existing passages
of music and create new pieces from them, which was considered a real
breakthrough. By analyzing various works, EMI was able to generate unique
structured compositions within the framework of different genres. In total, the
system created over a thousand works based on the works of 39 composers with
different musical styles.
• Aiva Technologies is one of the leading companies in the field of music
composition. It was founded in 2016 in Luxembourg and London by Pierre
Barreau, Denis Shtefan, Arnaud Decker, and Vincent Barreau.
Their AI is named “Aiva” (Artificial Intelligence Virtual Artist), which has been
taught to compose classical music. Having already released her first album called
“Genesis” as well as numerous individual tracks, Aiva has officially received
worldwide composer status. She was registered with the French and Luxembourg
Society of Copyright (SACEM), where all her works are registered with a
copyright in her own name.
• Shimon is a marimbaplaying robot from Georgia Tech professor Gil
Weinberg that sings, dances, writes lyrics, and can even compose some
melodies. Weinberg and his students trained Shimon with datasets of 50,000
lyrics from jazz, prog rock, and hip-hop.
• Noah 9000 is a music project by Michael Katzlberger, CEO of the
Viennese creative agency TUNNEL23. 9000 is a reference to the neurotic arti-
ficial intelligence “HAL 9000” from Stanley Kubrick’s 1968 masterpiece,
“2001: A Space Odyssey.”
For this project, the program was trained with piano sonatas by Ludwig van
Beethoven, among others. In order to create the perfect piece, it can sometimes
take many hundreds or thousands of iterations before correct, melodious sounds
are produced for the human ear. After all, probably the most interesting challenge
in producing modern music lies not only in the composition itself but also in the
instrumentation and sound design. Katzlberger, as curator, selected ten program-
generated generated tracks for the album Spring 9000 from around 2000 AI
compositions. It is an attempt to use artificial intelligence to create music that
feels human.
Anyone can play the steelpan via Virtual Piano. You don’t need to be able to read
music, and you don’t need to have played steelpan before. Virtual Piano’s sheet
music contains letters that correspond to the keys on a QWERTY keyboard. So it is
quick and easy to learn. You can start playing steelpan [1] immediately.
The platform virtualpiano.net allows you to play the piano on your
computer keyboard, cell phone, or tablet [2].
14.2 Artificial Intelligence and Architecture 185

14.2 Artificial Intelligence and Architecture

There are many ways that artificial intelligence is changing the field of architecture
[3].
• Programs can now make building calculations and environmental analysis a
simple task. Today, so much data is available to an architect that retrieving
information such as temperature and weather data, material ratings, etc., which
would otherwise take a lot of time to compile, is much easier.
• The application of intelligent systems can significantly reduce the time
required to plan and design a structure through building information
modeling (BIM).
• computer-aided design (CAD) has been instrumental in creating 2D
and 3D models of buildings. BIM goes a step further, integrating product
information, time, and cost to give an architect the full scope of a project. BIM
works with other design software programs to provide a complete picture of a
structure, including conceptual design, detailed design, analysis, documentation,
fabrication, construction, operation and maintenance after the building is built,
and renovation or demolition plans.
• Through appropriate use, programs can leverage this data and identify trends
or inefficiencies. This allows architects to design better, longer-lasting projects
without spending a lot of time analyzing data.
• Smart technology also gives architects the ability to incorporate smart lighting
or smart stormwater management systems into plans. None of this was possible
a decade ago. The proliferation of technology is allowing smaller practices to
take on larger projects, as they can use the Internet and all the resources at their
disposal to develop proposals.
• About seven percent of the world’s workforce is employed in the construction
industry, but it has traditionally been one of the least technologically advanced
sectors. However, there is great potential for integrating AI into construction, and
it could reduce construction costs by up to 20 percent.
• Video security has long been used in businesses, offices, and campuses to monitor
who comes and goes. However, due to the high volume of footage collected,
things can get overlooked if no one is watching the video. An intelligent
security system integrated into a building, on the other hand, can quickly scan,
automatically detect suspicious activity, and send alerts to the building owner.
• AI-based energy management platforms can set usage patterns to create ideal
conditions for tenants, saving them both energy and money. The Nest Thermostat
is an example of such a system that is popular with consumers—it adjusts to
keep a building at a safe and optimal temperature and can alert the user if the
temperature drops or rises to a dangerous level. AI devices can also take data
from sensors and analyze it to monitor for leaks or malfunctions. They make it
easier than ever to keep an eye on a building’s performance and efficiency.
AI can also be used in the form of intelligent locking systems that can restrict
access to certain areas unless the user has a key card or code. In addition to using
186 14 Artificial Intelligence in the Arts

Fig. 14.1 The architecture of the future

autonomous or semiautonomous construction equipment to support excavation

and preparation work, computers can analyze construction sites and identify
potential risk factors, reducing safety risks and associated delays.
Video feeds are being used to collect data on human behavior and usage
patterns around the world. For example, AI is already being used to optimize
operations in museums and airports. Companies could design buildings that
integrate to create entire smart cities based on how people interact with their
surroundings and how they feel about public spaces.
From the Great Wall of China to the Egyptian pyramids, architectural
innovation has been about making the most of available technology. As human
needs and technologies evolve, AI is poised to take humanity’s architectural
achievements to the next level (Fig. 14.1).

References

1. https://virtualpiano.net/virtual-steelpan
2. https://virtualpiano.net
3. https://blog.dormakaba.com/de/sieben-wege-wie-kuenstliche-intelligenz-die-architektur-
revolutioniert
Chapter 15
Artificial Intelligence in Law

The legal system is also beginning to include artificial intelligence in its con-
siderations, relatively hesitantly and cautiously in Germany, although research
also already exists in this field, which began more than 30 years ago and is
associated with the person of Professor J. van den Herik from the Netherlands. At
present, Leiden University can be considered as a center of connection of artificial
intelligence and law.
There are the following institutes in Leiden:
• Law
• Institute for the Interdisciplinary Study of the Law
• Van Vollenhoven Institute
• Constitutional and Administrative Law
• Institute of Private Law
• Child Law
• eLaw—Center for Law and Digital Technologies
• Europa Institute (Fig. 15.1)
• What opportunities and threats arise from institutional and normative diversity
in the field of fundamental rights for effective protection of these rights in a
pluralistic world? [2]
The research program “Effective Protection of Fundamental Rights in a
Pluralistic World” (EPFR) was launched in its current form in 2015. It builds on a
strong tradition of stimulating and high-quality human rights research at Leiden
University and seeks to further expand and strengthen this line of research. The
program is unique in the Netherlands, as it is the only stand-alone research
program on fundamental rights at a Dutch law school.
The EPFR program explores the dynamics of institutional and normative
diversity in the field of fundamental rights protection against the backdrop of
sociocultural, political, and economic pluralism, which is a prominent feature of

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 187
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_15
188 15 Artificial Intelligence in Law

Fig. 15.1 Computers and

law

today’s world, both globally and locally. The program will focus on the following
topical and intersecting issues:
• Many handwritten and illustrated archives contain a wealth of information, but
are largely unexplored because they are complex and difficult for computers to
decipher. The goal of this project is to develop a digital environment that solves
this challenge and connects heterogeneous archival content with other digital
sources [4].
The project will focus on one of the Naturalis Biodiversity Center’s most
important collections: the Natuurkundige Commissie archive and collection,
which contains a rich verbal and pictorial record of the nature, cultures, and
economies of the Indonesian archipelago (1820–1850). Researchers will use an
advanced handwriting and image recognition system (Monk) supplemented by
contextual information on species, locations, and habitats. Naturalis’ taxonomic
expertise will be used in combination with methods from the history of science to
further refine the system. The outcome of the project will enable Brill to offer the
system as an online heritage service. This will serve both curators of illustrated
manuscript archives and researchers seeking to further the understanding of these
collections.
As digitization continues, the legal work environment continues to change [1],
from the introduction of electronic records in government to sophisticated software
solutions to the use of specialized algorithms [3].
According to a survey conducted by Freie Fachinformation Verlag in 2020, the
greatest motivation among lawyers to deal with legal tech is to increase efficiency
while reducing costs and maintaining competitiveness. At the same time, around
30 percent of respondents specifically stated that they were planning to introduce
individual legal tech solutions. This shows that there is no way around legal tech.
During the corona crisis in particular, some of the legal tech startups were able to
record considerable growth. Canceled flights, train rides, or package tours led to
more and more consumers wanting to enforce their legal rights—with as little cost
and risk as possible.
15 Artificial Intelligence in Law 189

Legal tech generally stands for IT-supported applications that support legal
work processes. The aim is to make workflows more efficient and thus more cost-
effective. The use of various software programs and databases is now commonplace
in many law firms. The software works largely with if-then rules. Artificial
intelligence goes one step further: artificial intelligence deals with the automation of
intelligent behavior and machine learning. Here, algorithms also initially work with
rules, but in contrast to software, they repeatedly adapt them based on problems
and data that have already been solved and try to recognize patterns. In this way,
even complete legal services can be mapped and human or legal decision-making
processes can be emulated.
A large number of young companies have discovered this topic and know how to
exploit it commercially. Platforms such as flightright.de, which specializes
in airline passenger compensation, or wenigermiete.de, whose core business
is unlawful rent increases, are flooding onto the legal market. The lockdown with its
consequences has done its bit to make legal portals highly popular with consumers.
The company RightNow, for example, is advertising claims for refunds of gym
fees during the pandemic-related shutdown or compensation claims for data mishaps
at the vaccination center.
The advantages for consumers are obvious: they receive a quick assessment of
their chances of success in enforcing their claims at low cost or even completely
free of charge. The critical point here is that these programs cannot replace an
initial consultation with a lawyer, but customers are not necessarily aware of
this. At the same time, they are significantly cheaper and therefore also tempting.
Consumer protectionists also criticize the lack of cost transparency in some cases.
The commissions due in the event of success are not always transparent on the
websites and may even be higher than with a law firm.
The IT skills of lawyers will continue to develop in line with technological
developments. IT skills will go far beyond operating a computer and will be an
essential part of the legal profession. Lawyers are just as technology-savvy as other
professions. The young people currently studying at universities belong to the group
of digital natives and have grown up with the new technologies. Education
should be more interdisciplinary. The law faculty in particular must integrate the
topic of digitization directly into the curriculum.
Ultimately, not all regulatory issues have been clarified for the use of deep
learning systems. Only in April of this year did the EU Commission publish a
draft for a regulation establishing harmonized rules for AI. The German Federal
Bar Association sees this as a first important step, but also expresses a need for
further clarification.
Meanwhile, on the legal market, the dispute between legal tech companies and
bar associations continues to smolder, as consulting services in particular may only
be provided by lawyers to protect against unqualified legal advice. The fact that
this is not always clear is shown by the recent ruling of the Federal Court of
Justice on online contract offers by Smartlaw. The BGH ruled in favor of the
online contract provider, while the plaintiff bar association still sees a risk of false
advice for consumers. And while the ruling may be a small victory for providers
190 15 Artificial Intelligence in Law

of automated contract offers, the BGH made it clear that this was no substitute for
legal advice from a lawyer. Even if some legal tech companies are also consolidated
or partially disappear in the future, it brings movement into the market.
In the long term, AI and algorithms will certainly be able to compete with law
firms in certain legal matters. However, they will not replace them. It is a matter of
actively shaping the development of AI in use in the legal profession—technically,
ethically, regulatory, and culturally. During the corona crisis in particular, many
processes in the legal sector were digitized, which would otherwise have taken much
longer. This effect must be further leveraged, because ultimately, it will come down
to a combination of humans and machines that complement each other.
To achieve this goal, experts and specialists in this field are needed above all
to support law firms on this path. So well-trained personnel and know-how will
remain the drivers of innovation in the future—only the way in which this is used
will change.
LawBot, the chatbot for legal issues launched by a group of Cambridge
University law students, is expanding its capabilities by launching in seven countries
and adding a case outcome prediction feature that can be used to estimate the odds
of winning a legal case analyzed by the bot.
Bister, LawBot-X focused on English criminal law, but the new version now
supports the USA, Canada, Hong Kong, Singapore, Australia, and New Zealand, as
well as a Chinese language option. It can also analyze information provided by users
before using data science techniques to calculate a lawsuit’s chances of winning.
According to CEO Ludwig Bull, the system currently predicts the outcome of the
lawsuit with 71% accuracy.
“We are proud to make this service available to the public. LawBot-X is
important both for basic research in the application of AI to legal services and for
advancing the rule of law.” This means that once the AI learns the user’s situation,
it will be able to select actual lawyers in the right country or jurisdiction who can
provide assistance.
LawBot’s head of marketing, Rebecca Agliolo, said, “Following the success
of LawBot’s first pilot, we are excited to bring to market a platform to analyze
the quality of users’ claims. We have evolved from decision trees to data-driven
intelligence.” LawBot was founded to explain the complex law surrounding sexual
offenses. It was then expanded to include a total of 26 serious crimes.
In the USA, there is also an extensive effort to make AI reside in the legal sector.
Earlier this year, Ginni Rometty, chair, president, and CEO of IBM, told CNBC’s
@Work Talent + HR Conference: “I expect AI to change 100 percent of jobs within
the next five to 10 years. This is going to require skills development on a massive
scale, and companies are going to have to do it in a way that is inclusive - including
people who may not have a four-year college degree. The last time we saw such a
big paradigm shift was when the Internet was introduced. The evolution is not really
a question of whether, but rather when and how.”
Many don’t like change. Lawyers, in particular, are notorious for it. In the two decades that
I have been working with lawyers from all over the United States, I have found that among
all professional service providers, lawyers are among those most resistant to change. I find it
15 Artificial Intelligence in Law 191

fascinating that a group of trained logicians do not always recognize the difference between
optional and mandatory changes. Whether they like it or not, law firms are businesses, and
the most successful law firms pay attention to business fundamentals such as return on
investment (ROI), marketing, and so on [5].

Recognizing and adapting to change has always been—and will continue to be—
the hallmark of successful companies. Take e-discovery, for example. There
was a time when the use of e-discovery tools was rare in organizations. It was simply
unthinkable that a firm that had a veritable warehouse of employees sifting through
documents could ever be overtaken by a firm that used e-discovery technology.
Today, the exact opposite is true: a firm that reviewed large volumes of documents
and did not use technology would be an underdog. The e-discovery pattern has
changed relatively quickly.
In June 2007, Apple released its first iPhone. Until then, Apple was known
primarily for computers and iPods. Steve Jobs, however, recognized early on
the value of combining computer technology with smartphones. In fiscal 2007,
Apple generated $24.6 billion in revenue, with the fledgling iPhone accounting
for a relatively small share. In contrast, in the fourth quarter of 2019 alone, Apple
generated nearly $63 billion in revenue. Today, just 12 years after its launch, the
iPhone accounts for nearly 70 percent of Apple’s revenue (estimated at more than
$400 billion). The ability to recognize and proactively respond to the coming change
has made Apple a powerhouse company. Apple could have buried its head in the
sand and continued to focus on iPods and Macs. Had it done so, Apple might not
even be in business today. The iPhone was first an iPod in a phone. Now it gives
users access to Siri, a close application of AI.
From services like time tracking to contract review to brief analysis and research,
new technology companies are emerging all the time to meet the needs of lawyers.
Some of these tech companies are quickly going under, and that’s despite the fact
that their ideas are often ahead of very promising ones. Add to this the widespread
fear of change among lawyers. The result is that the advances the legal field could
make to increase its overall efficiency are almost nonexistent.
LawGeex is an automated contract review platform that uses AI to answer
the simple question “Can I sign this?”. The idea is simple: a contract is sent to
a company and uploaded to LawGeex’s AI, which then reviews the contract. If
problems are found, the contract is forwarded to a legal team that highlights the
questionable wording. The time saved is immense. More importantly, the accuracy
is unparalleled. In a study of nondisclosure agreements (NDAs), five contracts were
analyzed by 20 experienced lawyers.
There are over two and a half thousand startups in Hong Kong, employing nearly
10,000 people. Companies like Zegal are tapping into this large pool of talent and
changing the game for companies that have previously shied away from complex
legal documents [7].
“Historically, law has been an intimidating area for individual businesses, espe-
cially small- and medium-sized businesses and law firms,” says Tai Hung-Chou,
CTO of Zegal. “We provide a place where clients and lawyers can collaborate to
draft and execute contracts - all online. In the future, we will use advanced natural
192 15 Artificial Intelligence in Law

language processing and artificial intelligence to provide access to a variety of

precedents that can be inserted during the drafting process.”
Over the following decades, Case Reporters were joined by a wealth of secondary
sources, including treatises, restatements, and various legal journals. The Case
Reporters themselves were supplemented by Pocket Parts, i.e., booklets inserted
into special pockets in the back of the Reporters that updated the cases contained in
the Reporter. Lawyers of a certain age will remember the Pocket Parts, perhaps with
nostalgia or perhaps with horror [6].
But by and large, as Don MacLeod of Debevoise explains, things didn’t change
all that much for the longest time. Law was found in books that sat on dusty shelves
in law school, the bar, or the county law library. It was there and in the minds of
experienced lawyers that knowledge was kept. Junior Associates had to go
to the library, sit in the car, go through the books, and report back to the partners.
Then, as MacLeod puts it, the asteroid struck, namely, the advent of computerized
legal databases such as Westlaw and LexisNexis, in which cases were
continuously published, updated, and annotated. These services came on the market
in the 1970s and enjoyed great popularity during the PC revolution in the 1980s.
Today, online legal research services provided via the Internet (as opposed to
dedicated terminals) dominate the landscape. Law students and lawyers do most
of their legal research online. Camping out at the library is no longer necessary.
“Today we can access a wide range of information and analyze it quickly and
comprehensively, which wasn’t possible before,” MacLeod says. “And now we have
24/7, mobile access to information no matter where we are.” Online legal research
services have transformed the work of attorneys to a degree that John B. West’s
clients could never have imagined. It’s fair to say that overall legal research today is
faster, easier, and more accurate than ever before. But powerful new tools bring new
challenges and opportunities, which we now turn to.
Legal searches will always be more complex and sophisticated than consumer
searches. But those higher expectations from the consumer space are migrating into
the legal space, leading lawyers to look for search tools that are faster and more
intuitive. “There’s so much new machine learning that’s now going into what you
type into the search bar,” Tonya Custis says. “With Westlaw Edge, we analyze your
search linguistically. Are you looking for a group of cases on a particular topic? Are
you looking for an ALR article with a question in the title? Are you looking for the
answer to a specific, narrow question? We’ve used intelligent programs to figure out
user intent so you get exactly what you’re looking for”.
In other words, according to Khalid Al-Kofahi, Westlaw Edge is better
able to distinguish between different types of requests and respond accordingly. But
that’s not all, Al-Kofahi continues. Westlaw Edge helps researchers formulate the
right questions, helping attorneys find that elusive issue or decision that’s on the
tip of their tongue and they can’t think of at the moment—which is just one way
Westlaw Edge saves attorneys time and increases their confidence in their research.
Of course, Westlaw Edge gets much of its power from the improvements made
to cases by a team of attorney editors. Other legal research tools also use artificial
intelligence, but they do not apply artificial intelligence to the information that has
15 Artificial Intelligence in Law 193

been improved through organization and annotation by hundreds of highly skilled

and experienced lawyer editors working for Thomson Reuters.
Take, for example, Westlaw Edge’s predictive typeahead feature, where a user
enters a natural language query and receives suggestions for research questions
that then lead the user to the exact part of the document that answers the
questions. Sometimes the answers include West headnotes, which is just one
way Westlaw Edge combines artificial and human intelligence. “The technology
wouldn’t work the same way without Westlaw’s editorial enhancements,” Leann
Blanchfield explains. “If you have the technology but compare it to content without
enhancements, you don’t get the same results.” The cycle works both ways: the data
improvement helps the technology, and the technology helps the data improvement.
Artificial intelligence and other tools enable Blanchfield and her content team to sift
through the vast ocean of legal content more quickly and to flag cases and other
authorities more accurately and consistently.
Litigation analytics uses artificial intelligence to sift through a vast
amount of historical data, giving lawyers a comprehensive picture of how a judge
might rule on a particular claim, how long such a ruling might take, and a whole
host of other factors that can help lawyers develop a litigation strategy and advise
clients on how to proceed. The tool uses graphics and data visualization to present
this information to attorneys in an aesthetically pleasing, easy-to-understand format.
“Attorneys often formulate their litigation strategy based on their own experience
and insights,” explains Khalid Al-Kofahi. “That’s fine, but we believe we can
do better.” Legal analytics is about putting data at the fingertips of lawyers
alongside personal insights. It’s about providing lawyers with statistics related to
the likelihood of a court or a judge granting a motion, denying a motion, and ruling
on a motion in a certain time period—whatever.”
Estimating the likelihood of a particular decision has historically been a chal-
lenge because of the sheer volume of data that needs to be analyzed, said Barb
McGivern, vice president of strategy and marketing at Thomson Reuters: “Imagine
many rooms filled from floor to ceiling with pieces of paper: court transcripts. You
want to know how the judge will rule on your motion, and somewhere in those huge
rooms is your answer. That’s been the problem with Big Data in law. But now, using
the latest technologies, legal analytics can surface those insights and present them
clearly using visualization and other tools.”
“I’m really looking forward to talking to clients about litigation analytics,” Jon
Meyer says. “These days, lawyers are trying to figure out how successful a particular
argument might be in a particular court or judge based on anecdotal evidence—but
with litigation analytics, we now have the data to make informed decisions.”
With litigation analytics, lawyers can quickly understand how judges might
rule with a much higher level of confidence—based on comprehensive, empirical
information, not hunches or gut feelings.
Another area of concern for attorneys: whether they are citing good law. For
many years, KeyCite, Westlaw’s citation service, has pointed out potential problems.
A red flag in KeyCite indicates that the case in question is no longer established law
(on at least one legal issue), and a yellow flag in KeyCite means that the case has
194 15 Artificial Intelligence in Law

a negative history, such as criticism from another court. KeyCite Overruling

Risk, another feature added in Westlaw Edge, introduces an orange warning icon:
an indicator that a case may be implicitly overruled because it relies on an overruled
or otherwise invalid prior decision. This feature allows attorneys to check the status
of a case in a way that was not possible before.
In the past, said Mike Dahn, senior vice president of Westlaw Product
Management, a status flag required an explicit reference to the flagged case
in a subsequent instance. This was good for capturing cases that were directly
overturned or explicitly discarded. But it did not capture the many cases where
later developments in case law turned that case into bad law. Thanks to KeyCite
Overruling Risk, attorneys are now alerted to situations where a case may no
longer be valid due to new developments in case law. Attorneys are directed to
the authorities challenging their case so they can determine if those authorities are
accurate—and the risk of citing bad law is dramatically reduced.

References

1. https://www.perconex.de/de/perconex-personaldienstleister.html
2. https://www.universiteitleiden.nl/en/research/research-projects/law/effective-protection-of-
fundamental-rights-in-a-pluralist-world
3. https://www.universiteitleiden.nl/en/research/research-projects/law/the-legitimacy-and-
effectiveness-of-law--governance-in-a-world-of-multilevel
4. https://www.universiteitleiden.nl/en/research/research-projects/science/making-sense-of-
illustrated-handwritten-archives
5. https://growpath.com/future-of-law-artificial-intelligence-for-lawyers
6. https://abovethelaw.com/law2020/how-artificial-intelligence-is-transforming-legal-research
7. https://zegal.com/en-hk
Chapter 16
The Development in South and Central
America and the Caribbean

Based on my work at the University of the West Indies, I would like to make a few
more remarks about artificial intelligence in this region. This is another feature of AI
that has not yet been mentioned, but which is of paramount importance, especially
in countries as small as the Caribbean islands. You develop a high-quality system in
one place and can immediately pass it on to others with ease. With a good division
of labor, you can bring several countries up to a high level.
“Some of the push in AI adoption in these countries has come from academics
and researchers, like the ones at the University of Sao Paulo who are developing AI
to determine the susceptibility of patients to disease outbreaks, or Peru’s National
Engineering University where robots are being used for mine exploration to detect
gases, or Argentina’s National Scientific and Technical Research Council where AI
software is predicting early onset pluripotent stem cell differentiation.
These and other truths were revealed to me at a Latin America and Caribbean
(LAC) Workshop on AI organized by Facebook and the Inter-American Devel-
opment Bank in Montevideo, Uruguay, in November this year. I was the lone
Caribbean participant in attendance, presenting my paper entitled: AI & The
Caribbean: A Discussion on Potential Applications & Ethical Considerations, on
behalf of the Shridath Ramphal Centre (UWI, Cave Hill)” [3].
So what is happening here in the Caribbean?
AI remains, at best nascent, with limited R&D. Reports of its limited application
in The Bahamas, Belize, and Guyana and the absence of policy discussions belie the
significant potential for AI here.
We could dream of a day when Fedo, a risk stratification AI system for
predicting patient susceptibility to noncommunicable diseases (NCDs), is used in
the Caribbean’s health sector, where NCD mortality is highest in the Americas, or
when Dragon Medical Virtual Assistants assist the region’s critical nurse shortage,
which in 2010 measured 1.25 nurses for every 1000 people. How about See &
Spray, a weed control AI robot that could reduce herbicide expenditure by 90%?
Or AI harvest robots replacing 30 workers in the Caribbean’s agricultural sector,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 195
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_16
196 16 The Development in South and Central America and the Caribbean

where the food import bill is expected to reach USD 8–10 billion by 2020? Could
we ever see AI systems developed by Google, Harvard, and NASA which predict
earthquake aftershocks, flooding, and hurricanes as part of the Caribbean’s disaster
management and mitigation efforts, to save lives and mitigate potential loss?
Instead of dreaming, I propose the following three steps that the Caribbean can
take to better position itself to harness AI’s potential.
• First, we must develop an appetite for such technologies.
South America’s engagement in this field is a testament of the region’s innovative
capabilities and appetite for such technologies. This cannot be done without
firms and governments that are willing to adopt and utilize these systems
in their provision of goods and services. In addition, we need research and
studies that demonstrate how can AI be leveraged to solve some of the region’s
developmental challenges. It falls to the region’s academia and private sector to
find innovative AI solutions and spur demand for their subsequent development
and adoption.
• Second, we must form strategic partnerships.
Google is developing an AI system to predict and send flood warnings in India;
Unilever is testing various AI solutions in South America; and MIT and Harvard
are hosting AI R&D conferences in Uruguay, but, who are we partnering with in
the Caribbean?
Recognizing the importance of strategic partnerships and taking steps to reach
out to organizations like the IDB to fund such initiatives, or companies like
Facebook and Google to develop and test AI solutions in the region, or AI R&D
centers and universities to partner with are all potential avenues for overcoming
existing financial and resource constraints that hinder our progress in this field.
• Third, we must initiate AI-related policy discussions.
Realizing the wider ethical and legal considerations arising from the application
of AI, we must ask probing questions like the following: are the existing
frameworks capable of addressing our concerns? And how can we mitigate risks
and instill public confidence in such technologies?
Beyond technologists like engineers and developers, the discussion must involve
policy-makers who must be on the front lines in developing adaptive and
anticipatory frameworks. Similar to Mexico’s move toward an AI strategy, which
aims to transform the country from an observer to a key actor, we must look
toward the development of holistic approaches.
While not exhaustive, this list of recommendations is a start to riding the AI wave.
It is now up to us to either learn how to ride it, like our South American neighbors
are doing, or we get washed ashore.
16.1 Guyana 197

16.1 Guyana

v75 Inc. a local technology company has teamed up with Professor Jason Mars, creator
of Jaseci – a leading artificial intelligence programme – to create Guyana’s own Silicon
Valley. Once established, Guyana’s talented young minds will be able to build their own AI
products to be marketed and sold to top companies around the world. Already a team of five
Guyanese engineers who recently concluded an apprenticeship programme are employed
with Jaseci and are creating their own AI products. Part of the wider intention is to upskill
Guyanese in AI and create tangible opportunities to retain those persons here. [1]

• Crop and soil monitoring

Micro- and macronutrients in the soil are critical factors for crop health and both
the quantity and quality of yield. Then, once crops are in the soil, monitoring the
stages of growth is also essential to optimizing production efficiency. It is vital
to understand interactions between crop growth and the environment in order to
make adjustments for improved crop health.
Now, traditionally soil quality and crop health were determined by human
observation and judgment. But this method is neither accurate nor timely. Instead,
we can now use drones (UAVs) to capture aerial image data and train computer
vision models to use this for intelligent monitoring of crop and soil conditions.
• Crop maturity
Manual observation of wheat head growth stages is just the kind of labor-
intensive process that AI can help with in precision agriculture. Researchers
achieved this by collecting images of wheat at different “heading” stages across
3 years and in different lightings, which enabled them to create a “two-step
coarse-to-fine wheat ear detection mechanism.”
This computer vision model was then able to outperform human observation in
accurately identifying wheat growth stages, meaning that the farmers no longer
had to make daily treks into the fields to examine their crop.
Another study examined how well computer vision can detect maturity in toma-
toes. Researchers created an algorithm that analyzed color from five different
parts of the tomato and then made maturity estimates based on this data. The
algorithm achieved a successful detection and classification rate of 99.31%.
Observing and estimating crop growth and maturity is a hard, labor-intensive
work for farmers. But AI is proving capable of handling much of that work with
both ease and impressive accuracy.
• Track crop health
Getting back to the importance of soil, another study set out to see how well
computer vision can characterize soil texture and soil organic matter (SOM).
Ordinarily, evaluating soil requires farmers to dig up samples and bring them
to a lab for time- and energy-intensive analysis. Instead, researchers decided to
see if image data from an inexpensive hand-held microscope could be used to
train an algorithm to do the same thing. Sure enough, the computer vision model
managed to make sand content and SOM estimates with accuracy comparable to
costly lab processing.
198 16 The Development in South and Central America and the Caribbean

Fig. 16.1 Potato beetles

So, not only can computer vision eliminate a large amount of the difficult,
manual labor involved in crop and soil monitoring; in many cases, it does it more
effectively than humans can.
• Finding bugs with code
And say you’d like to know not only if your crops have pests, but how many
there are, computer vision systems for insect detection has that covered as well
(Fig. 16.1).
This works for flying insects too. They are certainly not the most fun to capture
and count by hand.
Researchers first set up a sticky trap to capture six different species of flying
insect and collect real-time images. They then based the detection and coarse
counting method on YOLO object detection and the classification and fine
counting on support vector machines (SVM) using global features.
When all was said and done, their computer vision model was able to identify
bees, flies, mosquitoes, moths, chafers, and fruit flies with an accuracy of 90.18%
and count them with 92.5% accuracy.
These studies show that the future of AI computer vision for monitoring
the health of our food systems is promising. Not only can it reduce labor
inefficiencies, but it can do so without sacrificing reliability of the observations.
• Livestock health monitoring
So far we have focused mainly on plants, but there is more to agriculture than
wheat, tomatoes, and apples.
Animals are another major component of our agriculture systems, and they tend
to need a bit more tracking than plants. Can computer vision keep up with cows,
chickens, and pigs on the move?
16.1 Guyana 199

Well, if it can track a fly, it can certainly track a cow.

CattleEye’s training data allows for tracking and annotating of cattle using
bounding boxes and key points.
The algorithms are trained to look at video data and determine what the chickens
are up to—whether they are drinking, eating, sleeping, or doing something odd
that may be indicative of disease or behavioral problems.
• Intelligent spraying
We have seen that computer vision is good at spotting disorders in agriculture,
but it can also help with preventing them.
UAVs equipped with computer vision AI make it possible to automate spraying
of pesticides or fertilizer uniformly across a field.
With real-time recognition of target spraying areas, UAV sprayers are able to
operate with high precision both in terms of the area and amount to be sprayed.
This significantly reduces the risk of contaminating crops, humans, animals, and
water resources.
While the potential here is great, currently some challenges still exist. For
example, spraying a large field is much more efficient with multiple UAVs, but
assigning specific task sequences and flight trajectories for individual crafts can
be tricky.
But that does not mean that the game is over for intelligent spraying.
Researchers from Virginia Tech have devised a smart spray system based on
servo motor-controlled sprayers that use computer vision to detect weeds. A
camera mounted on the sprayer records the geo-location of weeds and analyzes
the size, shape, and color of each pesky plant in order to deliver precise amounts
of herbicide with precision targeting.
In other words, it is a kind of weed terminator. But unlike the terminator, the
accuracy of the computer vision system allows it to spray with such accuracy
that it manages to avoid collateral damage to crops or the environment.
• Automatic weeding
Intelligent sprayers are not the only AI getting into weeding. There are other
computer vision robots taking an even more direct approach to eliminating
unwanted plants.
Now, spotting a weed in the same way that computer vision can spot an insect
or oddly behaving chicken does not actually eliminate very much work for the
farmer. To be of even greater help, the AI needs to both find and remove the
weed.
Being able to physically remove weeds not only saves the farmer quite a bit of
work but also reduces the need for herbicides and thus makes the whole farming
operation much more environmentally friendly and sustainable.
• Robots in the weeds
Luckily, object detection can do a great job of identifying weeds and distinguish-
ing them from the crops. However, the real power comes when computer vision
algorithms are combined with machine learning to build robots that perform
automatic weeding.
200 16 The Development in South and Central America and the Caribbean

All this pretty well introduces BoniRob, an agricultural robot that uses camera
and image recognition technology to find weeds and remove them by driving a
bolt into the earth.
It learns to distinguish between weeds and crops through image training on leaf
size, shape, and color. That way, BoniRob can roll through a field eliminating
undesirable plants without the risk of destroying anything of value.
And while our AI friends are out in the fields anyway, perhaps there are other
jobs they could do.
A group of scientists is working on making this a reality with designs for
agricultural robots that detect weeds as well as soil moisture content. This way,
it can move through a field, removing weed and delivering appropriate amounts
of water to the soil as it goes.
Experimental results for this system show that its plant classification and weeding
rates are both at or above 90%, all the while keeping deep soil moisture content
at 80 ± 10
AI-driven agriculture bots are developing quite the resume!
• Aerial survey and imaging
At this point, it is probably unsurprising that computer vision also has some
terrific applications for surveying land and keeping an eye on crops and livestock.
But that does not make it any less significant for smart farming.
AI can analyze imagery from drones and satellites to help farmers monitor crops
and herds. That way they can be notified immediately if something looks amiss
without having to constantly observe the fields themselves.
Aerial imaging is also useful for boosting the precision and efficiency of pesticide
spraying. As mentioned previously, ensuring that pesticides only go where
they’re intended saves money as well as the surrounding environment.
• Produce grading and sorting
Finally, AI computer vision can continue to help farmers even once the crops
have been harvested.
Just as they are able to spot defects, disease, and pests as the plants are growing,
imaging algorithms can also be used to sort “good” produce from the defective
or just plain ugly.
By inspecting fruits and vegetables for size, shape, color, and volume, computer
vision can automate the sorting and grading process with accuracy rates and
speed much higher than even a trained professional.
Take carrot sorting, for example. It is laborious and usually done by hand.
However, researchers have developed an automated sorting system that uses
computer vision to pick out carrots that have surface defects or are not the correct
shape and length. A good carrot, then, is one that is the right shape (a convex
polygon) and does not contain any fibrous roots or surface cracks.
On these three criteria, the computer vision model was able to sort and grade
carrots with accuracy rates of 95.5%, 98%, and 88.3%, respectively.
Further, bringing us back to the classic tomato, another study found that AI with
machine learning was able to use image data with seven input features to grade
tomato quality with 95.5% accuracy.
16.2 The Caribbean 201

In both cases, the amount of painstaking manual labor saved is enormous. And it
is all thanks to a bit of AI training on what a good carrot or tomato looks like.
• The future of AI in agriculture: farmers as AI engineers?
Throughout human history, technology has long been used in agriculture to
improve efficiency and reduce the amount of intensive human labor involved
in farming. From improved plows to irrigation, tractors to modern AI, it’s an
evolution that humans and agriculture have undergone since the invention of
farming.
The growing and increasingly affordable availability of computer vision stands
to become another significant step forward here.
With considerable changes occurring in our climate, environment, and global
food needs, AI has the ability to transform twenty-first century agriculture by:
Increasing efficiency of time, labor, and resources
Improving environmental sustainability
Making resource allocation smarter
Providing real-time monitoring to promote greater health and quality of produce
Of course, this will require some shifts in the agricultural industry. Farmers’
knowledge of their field will need to be translated into AI training, and this will
depend on greater technical and educational investments within the agricultural
sector.
But then again, innovation and adaptation are nothing new in agriculture.
Computer vision and agricultural robotics are just the latest way farmers can
adopt new technology to meet growing global food demands and increase food
security.
If you are interested to learn more about AI applications across other industries,
check out: v7labs

16.2 The Caribbean

Many Caribbean Small Island Developing States (SIDS) have not yet developed
strategies and national policies to manage the impact of AI on their societies. The
Caribbean Artificial Intelligence Initiative is led by the UNESCO Cluster Office for
the Caribbean and aims to develop a subregional strategy on the ethical, inclusive,
and humane use of AI in the Caribbean SIDS. The Caribbean AI Initiative raises
awareness, informs, engages in open discourse, strengthens capacities, produces and
shares knowledge, and gathers expertise from various stakeholders from different
sectors in the English- and Dutch-speaking Caribbean SIDS to identify the way
forward in developing regional and national AI strategies. Against this background,
the Caribbean AI Initiative was launched on the occasion of the Sub-Regional
Consultation on the Draft Recommendation on the Ethics of AI, which took place
on August 5, 2020 [2].
202 16 The Development in South and Central America and the Caribbean

16.3 Belize

The development of AI systems always starts with the necessary education. An

excellent example is Belize [4]. The web page of this event has the following points:
• What is the need for AI Expert Certification?
Artificial intelligence has been gaining significant importance across various
nations. In recent years, there have been exceptional instances where AI is
being applied across various industries. This has necessitated businesses to
incorporate artificial intelligence into their framework. Implementing AI requires
professionals who possess a deeper understanding of artificial intelligence and
its technical context. As a result, there has been an increase in the demand for
professionals who can demonstrate their skills in applying AI. To stand out
from the regular crowd of AI professionals, individuals can take up AI Expert
Certification.
AI Expert Certification in Belize City assists professionals in grasping the diffi-
cult concepts of artificial intelligence and its terminologies. The certification of
Artificial Intelligence Expert is based on the five major AI domains. Professionals
who gain a vast knowledge of these domains can ace the exam and end the AI
Expert Certification in Belize City easily.
• AI Expert Certification Training in Belize City
AI Expert Training is conducted by our in-house training in Belize City.
Individuals get to interact with their trainers to understand the terms better.
Enrolling in AI Expert Training helps individuals to acquire practical insights
into AI and be ready for the industry. The Artificial Intelligence Certification is
globally recognized enabling professionals to pursue their dream jobs on a global
scale. Owning AI Expert Certification in Belize City further entitles individuals
to earn higher incomes and take up managerial positions. In addition, AI Experts
attain reliability and are credible among their peers in the field.
• Key Features of AI Expert Certification Training in Belize City
AI Expert Certification Training in Belize City is conducted through various
training methods. Individuals enrolling for AI Expert Training can engage in
practical exercises such as case studies and discussions. This makes it easier for
aspiring AI Experts to showcase their competencies while looking forward to
achieving their career goals. At the end of the 3-day training program, candidates
are to take up the AI Expert Exam. Participants who score 65% or more can
obtain their AI Expert Certification from Unichrone.
The schedule is excellent. It is an excellent examples for many places all over the
Caribbean and South America!
• Day One
Introductions and Arrangements
Human and Artificial Intelligence—Pt I
Human and Artificial Intelligence—Pt II
Ethics and Sustainability—Trustworthy AI: Pt I
16.4 Argentina and Uruguay 203

Ethics and Sustainability—Trustworthy AI: Pt II

Sustainability, Universal Design, Fourth Industrial Revolution, and Machine
Learning
Artificial Intelligent Agents and Robotics
Being Human, Conscious, Competent, and Adaptable
• Day Two
What Is a Robot?
Applying the Benefits of AI—Challenges and Risks
Applying the Benefits of AI—Opportunities and Funding
Building a Machine Learning Toolbox—How Do We Learn from Data?
Building a Machine Learning Toolbox—Types of Machine Learning
Building a Machine Learning Toolbox—Two Case Studies
Building a Machine Learning Toolbox—Introductory Probability and Statistics
Building a Machine Learning Toolbox—Introductory Linear Algebra and Vector
Calculus
• Day Three
Building a Machine Learning Toolbox—Visualizing Data
A Simple Neural Network Schematic
Open Source ML and Robotic Systems
Machine Learning and Consciousness
The Future of Artificial Intelligence—The Human + Machine
Learning from Experience—Agile Projects
Conclusion
An excellent approach that can be used in a similar way by many other
institutions. It is important to find teachers with the necessary background.

16.4 Argentina and Uruguay

In this latest installment of Oxford Insights’ AI strategy series, we turn our attention
to the Latin American region to take a closer look at two neighbors’ approaches to
artificial intelligence: Argentina and Uruguay [5].
In this year’s Government AI Readiness Index, Uruguay topped the Latin
American region, with an overall score of 55.57. Meanwhile, Argentina came in
a regional fourth, behind Chile and Colombia, with an overall score of 50.75. As
two regional leaders in this sphere, how these neighboring governments tackle AI
is likely to have an impact on the future digital and economic landscape in Latin
America more widely.
There are a number of parallels between Argentina and Uruguay’s AI strategies.
They were published only months apart in 2019, and both were designed by previous
administrations that have since been defeated in national elections. Yet, as we
explore in this blog, each is underpinned by seemingly very different priorities.
204 16 The Development in South and Central America and the Caribbean

For each area considered, we have determined which country’s strategy has the
comparative advantage according to our analysis.
Argentina has a broader AI strategy, which was published in 2019 under the
government of the then-president Mauricio Macri. The document seeks to promote
AI in the private sector, minimize ethical risks, and develop talent, among other
objectives. Uruguay’s AI’s strategy, also published in 2019 under the administration
of a former president, Tabaré Vázquez, is much briefer, dealing specifically with
promoting the use of AI within public administration. Nonetheless, Uruguay’s
Agenda Digital 2020, another output of the previous government, takes a much
broader view of digital transformation in the country and is in many ways more
comparable to Argentina’s strategy in this sense. As such, this blog also considers
the documents, reports, and agendas that accompany the AI strategies. In doing this,
we hope to achieve a fairer comparison, which accounts for discrepancies between
the ways in which each country structures its strategic documents.
• Background
Before diving into the analysis of each strategy, it’s worth acknowledging their
political backdrops.
As of May 2020, a report published by the Banco Interamericano de Desarrollo
(BID) claimed that Argentina’s AI strategy can be classed as to be continued, i.e.,
not currently implemented. In October, Mauricio Macri’s incumbent conservative
government was defeated by Alberto Fernández’s left-wing coalition Frente
de Todos. Although the Fernández administration did initially show some
willingness to commit to digital transformation by vowing to re-establish the
Argentine Ministry of Science and Technologies—which had been demoted to
a secretariat under the previous administration—Fernández’s mandate has since
been dominated by the coronavirus response.
During its presidential elections in 2019, Uruguay experienced something of a
reverse of the Argentine situation, when the incumbent socialist government was
replaced by a center-right President, Luis Lacalle Pou. Pou’s presidency seems
to have run more smoothly than his Argentine counterpart’s to date; as opposed
to Argentina, which was hit relatively hard by the virus, registering more than
44,000 deaths from COVID-19, Uruguay has only recorded 231 deaths at the time
of writing. Digitalization remains a government priority, and unlike Argentina,
the BID considers Uruguay’s strategy to be a completed document which is
currently in implementation. Uruguay immediately looked to technology in its
virus response, creating online resources through which people could seek advice
and report potential cases. This caught the attention of Apple and Google, and
subsequently, Uruguay was the first country in Latin America to implement their
track and trace technology.
While it’s possible that the Argentine government may disregard or modify the
2019 strategy, it’s nonetheless the most up-to-date vision of artificial intelligence
in Argentina and therefore our main point of reference for this blog.
• Vision: draw Both the Argentinian and the Uruguayan AI strategies set out
ambitious goals for a digital future and received full marks in the vision
16.4 Argentina and Uruguay 205

dimension of our index. Nonetheless, the visions projected by each strategy

have somewhat different focuses. Each strategy’s key objectives are summarized
below.
Argentina’s AI strategy imagines a much more commercially focused approach,
whereas its Uruguayan counterpart places much more emphasis on improving
the use of AI in government. These divergent priorities are reflected in our
index scores. As explored further below, Argentina scored higher public sector
readiness indicators.
• Creating a market for AI growth: Argentina (for now) According to our index,
last year, Argentina outperformed Uruguay when it comes to AI readiness in
the tech sector. This is based on data that indicates that Argentina has over
6000 tech startups, compared to only 185 in Uruguay. Similarly, an analysis
conducted by the Banco Interamericano de Desarrollo also concludes that
Argentina has a stronger entrepreneurial ecosystem, since it is home to several
technology unicorns, such as Mercado Libre (a Latin American equivalent to
eBay), Despegar (a travel booking website), or Globant (an IT and software
development company).
Argentina’s fluctuating economy is no stranger to high levels of inflation and
economic crashes, meaning the country is sometimes regarded as a difficult
business environment in which to operate. Nonetheless, the 2019 AI strategy
places significant emphasis on easing the path for tech giants and startups alike,
potentially contributing to Argentina’s success in this dimension. The strategy
references several initiatives in this regard, including;
• The creation of the National Fund of Entrepreneurial Capital (FONDCE) to
support startups.
Changes in legislation which make it possible to start a business in 24 hours.
Tax benefits for individuals who invest in startups and SMEs.
Meanwhile, the Uruguayan strategy makes little mention of the economy, given
its focus on public administration. The Agenda Digital 2020 is slightly more
commercially orientated, but focuses on “sustainable economic development”
and reducing the economic risks associated with AI, adopting a much safer tone
than its growth-led Argentine equivalent.
Despite this, many have noted that the Uruguayan tech sector is thriving,
with new unicorns like dLocal poaching executives from companies based in
Argentina, particularly since the Fernández administration introduced more inter-
ventionist government controls on the economy. Therefore, whether Argentina
will be able to hold onto this lead in the coming year remains to be seen.
• Public sector readiness: Uruguay
Whereas the Argentine strategy performs well under an economically focused
lens, the Uruguayan government outstrips Argentina when it comes to the
government indicators in our index.
Argentina’s AI strategy does propose some government-centric reforms, stress-
ing the need for the digitalization of government services, and highlighting
existing initiatives, such as Boti, a virtual assistant which allows citizens to
interact with local government in Buenos Aires.
206 16 The Development in South and Central America and the Caribbean

However, public sector reform hardly shines through as a central focus of the
strategy, as in the case with Uruguayan AI strategy, which is primarily focused
on AI reforms in the public sector. From the outset, the Uruguayan strategy
explicitly sets out to “consolidate a closer relationship between the people and
the state” and “provide more efficient and innovative services” to citizens, by
increasing digital capacity within government. To achieve this, the Uruguayan
strategy proposes to:
• Create a training program around the use of AI in public services.
Train all government departments using the aforementioned program.
Define standards, guidelines, and recommendations for auditing of the decision-
making algorithms used in government, among other initiatives.
• Feasibility and implementation: Uruguay Having a high-reaching vision for
change is one thing, but as our initial ‘What makes a good AI strategy?’ blog
sets out, AI strategies need to set out measurable goals if they are to translate
ambitious ideas into practical change.
This is something the Uruguayan strategy and its accompanying initiatives
do particularly well. Uruguay Digital’s website allows visitors to search for
particular digital initiatives proposed in the Agenda Digital 2020 and see how
close they are to full implementation, measured in percentage points. Elsewhere,
AGESIC, the ministry for digital government and society responsible for creating
Uruguay’s AI strategy, places significant emphasis on “actually making things
happen” and “improving the improvable”—suggesting that the Uruguayan gov-
ernment’s digital departments have considered the importance of a culture of
tangible change.
Example progress markers on Uruguay Digital’s website.
Source: https://uruguaydigital.gub.uy.
The Argentine government lacks a similar tool, making it hard to track progress
on some of the claims made in the AI strategy. To give credit where it’s due,
many of the goals proposed in the strategy do have clear markers. For instance,
the previous Argentine government sought to measure their investment in human
capital and digital education by monitoring tracking the number of degrees
related to AI available, scholarship programs abroad, and papers published,
among other indicators. Yet the government does not seem to have published any
data surrounding whether these goals are being met, likely due to institutional
discontinuity.
• Final takeaways
Many question marks hang over both Argentina and Uruguay’s AI strategies.
Are they simply vestiges of old administrations, as the apparent abandonment of
Argentina’s strategy seems to suggest? How will each country’s new government
address the challenges and opportunities associated with AI?
To date, it seems like Luis Lacalle Pou’s relatively new Uruguayan administration
has been quicker to take up the baton on AI than its Argentine equivalent,
integrating AI into a coronavirus response which has been praised internationally.
Meanwhile, following a change in administration, very little information is
available at all surrounding the Fernández administration’s approach to AI;
References 207

Argentina’s AI strategy, released by the ArgenIA group in 2019, has not yet been
approved by the new government in a resolution.
As the Uruguayan case demonstrates, AI can be a powerful tool for governments
coordinating a public health response to the virus. Yet, looking further forward,
the digital sector will continue to be crucial for administrations as they seek to
revitalize economies in the wake of the crisis. As inflation in Argentina nears
40% and the Argentina Central Bank runs low on dollars, how the Argentine
government addresses AI in the coming years has the potential to define the
nation’s economic trajectory. At Oxford Insights, we’re eager to see how both
administrations will make their mark on national AI strategies in Argentina and
Uruguay as they approach 1 year in office [5].

References

1. https://www.jaseci.org/the-future-of-artificial-intelligence-in-guyana/
2. https://en.unesco.org/caribbean-artificial-intelligence-initiative
3. Chelceé Brathwaite: Artificial Intelligence (AI) in South America—Setting the Bar for the
Caribbean, Caricom Today, JULY 27, 2023
4. https://unichrone.com/bz/artificial-intelligence-expert-training/belize-city
5. Jasmine Kendall: Oxford Insights’ AI Strategy Series: Argentina and Uruguay, 13/January/2021,
https://www.oxfordinsights.com/insights/2021/1/8/oxford-insights-ai-strategy-series-
argentina-and-uruguay
Chapter 17
Future Developments

Already, we are seeing major changes across the board. These developments will
continue, leading to sweeping changes in all areas of life.
In light of these developments, it is not surprising that currently 62 percent of HR managers
expect AI to fundamentally change working life in the next five years. Our working world
and job descriptions will change dramatically. For example, research by Erik Brynjolfsson,
director of the Massachusetts Institute of Technology (MIT) Initiative on the Digital
Economy, found that machine learning rarely replaces entire professions. Instead, it takes
over or transforms specific work processes within a job description. Individual jobs will
certainly disappear. Predominantly, however, many tasks will be performed jointly by
humans and AI, entirely in the sense of AI as a personal assistant. Machines can indeed take
over tasks with a high workload and routine tasks. However, they work less well on abstract
processes where skills such as empathy, judgment, creativity, inspiration and leadership are
required. These include areas such as innovation and personnel management [3].

Given these developments, the importance of people being flexible and open to
changing for new tasks, as well as to continuous learning, becomes clear. As AI
continues to evolve, the competence of employees must also follow these advances.
It is therefore necessary to develop further training strategies in order to also be
able to use the potential of AI. This is not just a matter for IT staff, but also for the
specialist departments. Salesforce, for example, offers numerous free training
opportunities via a learning platform Trailhead. The more these technologies
can best support our efforts to create better jobs and living conditions, the more
successful companies and progressive societies will exist.
The effects of the further development of artificial intelligence can be very
diverse and will be assessed very differently depending on the economic and social
context in the coming decades. In most industrialized economies, an (unconditional)
basic income is very likely—not least because it is even being pushed by the current
economic elite. Who pays the taxes? Well, the robots and algorithms! [4]
“The picture of the future in the technological realm is often referred to as
‘transmodern liberty’.” The term is composed of the elements of a transmodern

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 209
C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1_17
210 17 Future Developments

society as well as a new quality of the concept of freedom and is an optimistic

version of the future. My scenario includes the following aspects, among others:
• Technology makes people human.
What sounds contradictory, I am completely serious about. The right use of
technology, above all artificial intelligence, could in the best case lead to all
people being allowed to be human for the first time in the history of mankind.
What does human mean? To me, it means not having to worry about providing
for existential needs on a daily basis. This privilege has been reserved (if at all)
for a small elite since the beginning of humanity. But if basic needs can be met by
automated processes, at least the possibility arises that gainful employment will
become obsolete—instead, people will go to work because it suits their individual
strengths and interests and they thereby contribute to society.
• Sapiens 2.0 is born.
• Technology makes people human.
“Often in my lectures I talk about the Homo Prospectus as a forward-thinking
human being. This is a component of a new form of humanity; this does not
mean genetically optimized people, but those with a cosmopolitan mindset and
an understanding of Transmodern Liberty. They are based on an updated version
of the Enlightenment and humanism, recognize all life as equal, and are free to
unfold their own biographies without external constraints.”
• Automation of diplomacy and global governance.
In addition to the production of everyday goods and infrastructure, governance
must of course also be automated. Who benefits if within a few closed nation-
states the economy is automated, but the neighboring state regularly causes
diplomatic crises? Here we increasingly see the trend towards a true Tianxia, i.e.,
a globally inclusive political system: on the one hand, it recognizes the cultural
and historical differences of people and territories, but on the other hand, it levels
out human abysses such as greed, hatred, or revenge through clever diplomacy.
If a computer wins against a human in Go, it will soon be able to moderate a
diplomatic crisis. Just ask the people in crisis and war zones, they already have
no desire for terror.
• Back to the present.
Artificial intelligence will change the world at least as much as electricity or
the taming of fire. I don’t say that, Google CEO Sundar Pichai 2018
does, but I agree with him. At the same time, I am committed to providing
more education and demystifying the topic. Expectations for the current decade
range from fear of a Terminator scenario or complete automation to a lack of
understanding for that annoying computer science chatter and the devastating
digitization emergency in Germany. This must finally come to an end. We
need more enlightenment, better education, more information, but also spirited
discussions in prime time [5].
17 Future Developments 211

The “Seven Deadly Sins” of Future Development [5]:

• sin 1: Overestimating and underestimating
• sin 2: Magic ideas
• sin 3: Concrete performance versus general competence
• sin 4: One word—a suitcase full of meanings
• sin 5: Exponentialism
• sin 6: Hollywood scenarios
• sin 7: The pace of implementation
Hollywood scenarios have taken up a lot of space in film and television in
recent decades. In [1] talk about alien races (Fig. 17.1). To a first group of
extraterrestrial races belong the grays from the star system Zeta Reticulum and
the Orion constellation, which are ubiquitous in the representations of ETs. They are
mentioned in most researches related to abductions and UFO crashes.
The research team, which had already developed the first living robots (so-
called xenobots, which are made of frog cells and can perform simple tasks), now
discovered that the organisms, which were designed on the computer and assembled
by hand, can swim around in a petri dish and find individual cells. Further, they can
round up hundreds of these cells and use their mouths, reminiscent of the computer
game character Pac-Man, to assemble them into mini-xenobots. Within a few days,
these develop into new xenobots that look and move the same. In addition, these

Fig. 17.1 There are no limits

to the imagination
212 17 Future Developments

new xenobots can also swim around, find cells, and build copies of themselves.
This process can be repeated as often as desired, as the scientists describe in their
work, which was published in the journal Proceedings of the National Academy of
Sciences.
Joshua Bongard, a robotics expert and one of the study leaders, explained that
with the right design, the xenobots can spontaneously self-replicate. Normally, these
embryonic stem cells would develop into the skin layer of a clawed frog (Xenopus
laevis). These cells would then normally sit on the outside of the tadpole, fending
off pathogens and secreting mucus. But in the experimental design of their trials, the
researchers expose the cells to a new environment and context. Michael Levin, one
of the researchers, explained that this opens up the possibility for the stem cells to
remember genetic code, allowing them to develop into other cells. Sam Kriegman,
another scientist involved in the study, explained:
These are frog cells replicating in a way that is very different from the way frogs do it. No
animal or plant known to science replicates in this way.

For a long time, it was believed that all possibilities of reproduction of life-
forms had been researched and were known. But according to the scientists, this is
something that has never been observed before. The cells have the genome of a frog,
but since they do not develop into tadpoles, they can use their collective intelligence
to accomplish something so extraordinary. In previous experiments, researchers
were already amazed at the possibility of being able to engineer xenobots to perform
simple tasks. Now the researchers were all the more amazed to find that this
computer-designed collection of cells began to multiply spontaneously. According
to Levin, there was no indication from the complete frog genome that the cells could
work together and condense functional copies of themselves from the collected
cells.
A milestone toward organ cultivation was achieved in China. Researchers grew
human-monkey embryos for the first time [2]. The parent, consisting of about 3,000
cells, is shaped like a sphere and forms offspring on its own. Typically, this system
dies off quickly, so it is usually difficult to get it to continue reproducing. But
the researchers use an artificial intelligence program on a supercomputer cluster
and used an evolutionary algorithm to test billions of different body shapes in a
simulation, from spheres to pyramids to the shape of starfish. In doing so, they found
a shape that allowed cells to be as effective as possible in kinematic replication based
on their movement. As Kriegman, one of the study authors, explains, this allowed
them to figure out how to match the shape of the original parents:
“The AI, after months of work, came up with some strange designs, including
one reminiscent of Pac-Man. It’s very unintuitive. It looks very simple, but it’s not
something a human engineer would come up with.” But in practice, he said, that very
system has been shown to work: the parent xenobots were able to build children,
who then assembled grandchildren, who in turn produced great-grandchildren, and
who then produced great-great-grandchildren. With the right design, the number of
generations could be greatly expanded. Until now, kinematic replication was only
known at the level of single molecules, but was never observed at the level of whole
17 Future Developments 213

cells or organisms. With the researchers’ work, this has now changed. They have
discovered, they say, that there is a vast, previously unknown space within living
systems. According to the researchers, life harbors surprising behaviors just beneath
the surface.
Stem cell researchers grew embryos composed of parts of humans and monkeys
that survived in the culture dish for an astonishingly long time. The breeding of such
mixed creatures sparked a debate about ethical issues in research on such chimeras.
Almost 2 years ago, stem cell researcher Juan Carlos Izpis.úa Belmonte had already
made an announcement that caused some controversy: In July 2019, Ispizúa told
the Spanish newspaper El Pa.ís that he had worked with Chinese scientists to breed
embryos that were part human and part ape. Now, the associated publication has
appeared in the journal Cells, which is likely to reignite the debate about ethical
issues in this type of research.
With his team, the stem cell researcher had grown embryos from Javanese
monkeys. After 6 days, they implanted these with highly transformable human stem
cells. The cells then actually fused, and 132 embryos were created from parts of
humans and monkeys. After 10 days, 103 of the human-monkey chimeras were still
alive, and after 19 days, three were still alive.
The experiment is astonishing not only because of the ethical issues involved
but also that it worked at all. Until now, chimera embryos have never survived this
long. Earlier attempts at mixed human-mouse and human-pig embryos had failed.
The reason for this was probably the high species barrier. The chimera of monkey
and human, on the other hand, survived for quite a long time. But of course this
also raises the fear that this discovery could be misused for ethically more than
questionable purposes: if the chimera embryo were implanted in a human or animal
surrogate mother, a new, independent life-form could possibly develop on it.
Izpis.úa assured that he did not intend to do so; moreover, ethics committees had
previously reviewed his experiments. With his experiments, he pursues the goal of
better understanding embryonic development. He has also long been working on
growing human organs and tissue in pigs for sick people, he said:
It is our responsibility as scientists to think through our research well and follow all ethical,
legal and social guidelines.

According to Stefan Schlatte, director of the Center for Reproductive Medicine

at Münster University Hospital, the work in chimera research represents a break-
through. Thanks to basic research in this area, he said, we will have a much
better understanding of the formation of organs and tissues in the human embryo.
However, such studies must take place under the strict proviso that under no
circumstances is the birth of a hybrid being the goal. Rüdiger Behr, head of the
Department of Degenerative Diseases at the Leibniz Institute for Primate Research
in Göttingen, told the Süddeutsche Zeitung that human stem cells introduced
into a monkey embryo could potentially give rise to organs with completely new
properties:
In such chimeras, for example, pig-human hybrids, there would be a good chance
to produce organs for transplantation. Whether the rescue of such terminally ill
214 17 Future Developments

people is sufficient justification for the creation of such chimeras is still an open
question. This would have to be decided by each individual, by society as a whole,
and then finally by the legislature in a legally binding manner, Behr said.
But Michael Coors, head of the Institute for Social Ethics at the University of
Zurich, demands that this decision-making process must finally begin now, because
there is an urgent need for legal regulation of this type of research. In view of
the fact that Izpis.úa already announced his attempt 2 years ago, the facts now
do not exactly come as a surprise. It is all the more astonishing that the process
has hardly progressed. It is not only about mixed creatures but also about the
instrumentalization of the animals used and the potential suffering that might be
inflicted on such chimeras.
Some people may look with horror at the idea of self-replicating nanorobots, but
the scientists explained that they are primarily concerned with understanding the
processes involved. The millimeter-sized living machines, which are exclusively in
the lab, are easy to eliminate once the experiments are over and have also been
reviewed by ethics experts at various levels. Bongard said this is not something
that keeps him up at night. On the other hand, he said, one has a moral obligation
to understand the conditions of self-replicating systems under which they can be
controlled, directed, or contained. In this context, he also refers to the corona crisis:
The speed with which we can develop solutions is of great importance. If we can develop
technologies that learn from xenobots, and quickly tell the AI: We need a biological tool
that does X and Y and suppresses Z, then that could be very useful. Today, that takes a long
time.

The new technology could be used to accelerate the development of solutions

to problems in the future—for example, by using living machines to remove
microplastics from waters or to develop new drugs. The team also sees their research
as promising in the field of regenerative medicine. Levin explained: “If we knew
how to make collections of cells do what we want them to do, that would ultimately
be regenerative medicine—the solution to traumatic injury, birth defects, cancer, and
aging.”
So far, he said, we can’t predict or control which groups of cells will form or what
various problems in medicine will result from. With xenobots, he said, we now have
a new platform to learn and explore this.
We have already experienced the use of robots on the moon and on Mars, and
it will continue. Eventually, you will have factories outside of Earth that will again
build robots and rockets that will go further into space.
It is almost impossible to predict the future development a little more precisely;
it is already extraordinarily difficult to understand the present. If one subscribes to
some newsletters of suitable magazines and follows the new editions of books, then
one would like to add some new sections every day. Thus, I have omitted some
problems and must ask the reader to continue to follow the development himself.
For example, I pretty much left out programming. It was only mentioned that
many AI programs were programmed in Python; however, any other programming
languages could be used. Surely, however, it goes without saying that DeepMind
makes programming itself the subject of learning [6].
17 Future Developments 215

The introduction to this paper states: “Programming is a powerful and ubiq-

uitous problem-solving tool. Developing systems that can assist programmers or
even create programs independently could make programming more productive
and accessible, but so far incorporating innovations in AI has proven difficult.
Recent large-scale language models have impressively demonstrated their ability
to generate code and solve simple programming tasks. However, these models
still perform poorly on more complex unknown problems that require problem-
solving capabilities beyond simply translating instructions into code. For example,
competitive programming tasks that require an understanding of algorithms and
complex natural language remain challenging. To address this gap, we introduce
AlphaCode, a code generation system that can create novel solutions to these
problems that require deeper reasoning. Evaluating recent programming competi-
tions on the Codeforces platform, AlphaCode achieved an average ranking of 54.3
% in programming competitions with more than 5,000 participants. Participants. We
have found that three key components are critical to good and reliable performance:
• (1) a large and clean dataset for programming contests for training and scoring,
• (2) large-scale and efficient sampling-based architectures,
• (3) large-scale model sampling to explore the search space, followed by filtering
based on program behavior on a small set of inputs.”
The most comprehensive work on the future of AI is titled “I 2041 - Ten Visions
of the Future” [7]. It will certainly be interesting to pick up this book again in
20 years.
This study exploited the representation of unmanned systems to present military
personnel with challenges that the use of autonomous weapons may bring, partic-
ularly concerning the ethical challenges that may arise. Interviews with military
experts revealed that the ethics of AI, particularly in warfare, is expected to keep
evolving as an appealing and perplexing topic. Peculiarly, the acuteness of the topic
was not observed or spoken about as a moral aspect directly by agents of military
ethics. Ethical issues will likely emerge and become more evident if participants
have accepted AI weaponization as a tangible concept.
Military professionals see the development of AI-driven weapons systems as
an inevitable and notable process that will change future armed conflict inde-
terminately. The findings of this study indicate that approval of or opposition
against autonomous systems by most participants is contextual and is packed
with uncertainty in applicability. Mainly, it is believed that the decision-making
process by machines will be very problematic or impossible since even commanders
struggle with decisions with limited time and information. On the one hand,
autonomous systems could offer a significant advantage in increasing a unit’s
combat potential; conversely, it involves higher risk due to the military’s lack of trust
in AI-embedded systems. Much emphasis has been put by practitioners on trust and
building confidence in the unmanned system, while senior officers acknowledge that
disruptive technology requires a shift in military thinking and training. However,
the general understanding was that future utilization of AWS will not necessarily
change the art of war. The argument is that warfare is in the end, a struggle of will,
216 17 Future Developments

and machines do not have free will; it always remains between human leaders who
can eventually impose their will.
Regarding autonomous supporting functions, several participants referred in their
response to human-machine teaming issues, which is believed relevant because
unmanned systems are assumed unable to share combat experiences and to com-
municate military issues. Therefore, many respondents think it is still too early to
discuss acquisition of AWS, especially if they cannot replace soldiers or demand
more human resources to keep unmanned machines operating. Moreover, soldiers’
inner will and patriotic mentality to fight for freedom was underlined, which was
explained as inconceivable for an autonomous system, so unmanned platforms can
hardly ever be equipped with patriotical algorithms.
Unmanned warfare is a relatively new topic, for which reason I recognize that
there were limitations to this study. Participants had to imagine AI implications
for warfare, leading to very diverse representations even in the same field of
expertise, which renders results challenging to interpret. Although expert interviews
allow for in-depth examination, it is not enough to comprehensively capture the
essence of a novelty subject with a questionnaire alone. However, concerns about
key categories were asked explicitly; participants should be presented with case
studies that include these categories, which facilitate setting a focus on a particular
issue. Moreover, such designed vignettes might encompass cultural differences
and capture distinctive perspectives for this field of research. As a result, study
outcomes intend to reveal relevant aspects of AI to understand how it can be applied
responsibly in a military domain and the implications for the following research. In
that respect, it may help researchers and system engineers understand AWS’s main
benefits and principal vulnerabilities from an end-user point of view.
The datasets generated and analyzed during the current study are not publicly
available due the fact that they constitute an excerpt of research in progress but
are available from the corresponding author on reasonable request. The study was
accepted by the ethics committee of the University of Tartu. Informed consent was
obtained from all individual participants included in the study.

References

1. https://transinformation.net/ein-bericht-ueber-ausserirdische-rassen-die-mit-der-menschheit-
in-interaktion-stehen
2. Meilenstein in Richtung Organzucht: Forscher züchten in China erstmals Mensch-Affen-Embryo
https://de.rt.com/asien/90867de.rt, August 2019
3. Frank Engelhard. Die Zukunft der KI—und ihr Einfluss auf die Geschäftswelt, Silicon.de, 2019
4. Kai Gondlach. https://www.kaigondlach.de/artikel/wie-wird-kuenstliche-intelligenz-die-
zukunft-veraendern, 2021,
5. https://algorithmenethik.de/
6. https://news.mit.edu/2022/new-programming-language-high-performance-computers-0207
7. Kai-Fu Lee, Qiufan Chen, KI 2041—Zehn Zukunftsvisionen, Campus Verlag Frankfurt/New
York, 2021, 978-3-593-51549-6
Index

A Coloring problems, 59
Algorithm, 20, 27 Competence, 9
Alphazero, 76 Complexity, 41, 44
Alzheimer, 127 butterfly effect, 41
Analytic geometry, 136 exponential complexity, 41, 44
Axioms, 49 linear search, 44
logarithmic complexity, 44
polynomial complexity, 41, 44
B Computability, 30
Backpack problem, 62 Computer algebra, 37
Big Data Mathematica, 39
Alipay, 97 Microsoft Mathematics, 37
Apache Software Projects, 95 Conjunction, 16
Association Analysis, 94 CorelDRAW, 143
Clustering, 94 Correlation coefficient, 92
Corvid-19, 96 Covariance, 92
Hadoop, 95 Covid-19, 128
MapReduce, 95 Cybernetics, 63, 64
Predictive analysis, 94 control, 64
Spark, 95 feedback, 64
Terark, 97
WeChat, 97
Bog data D
regression analysis, 94 Dartmouth Conference, 19
Boole, G., 15 Database, 83
Boolean Algebra, 16 Brockhaus, 83
Brain research, 6 database language, 83
Browser, 26 dictionary, 83
management software, 83
management system, 83
C relational database, 84
Church-Turing thesis, 30 SQL, 83
Circuits, 17 Digital camera, 113
Cloud computing, 85 Digital Revolution, 113

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 217
Springer Nature Switzerland AG 2024
A. C. Posthoff, Artificial Intelligence for Everyone,
https://doi.org/10.1007/978-3-031-57208-1
218 Index

Digital Transformation, 113 inherited intelligence, 6

Digitization social intelligence, 7
agriculture, 118 swarm intelligence, 7
digital transformation, 120 Intelligence Quotient (IQ), 10
education, 120 Intelligence test, 10
GPS, 119 psychological test, 10
RGB color space, 116 Internet, 26
Unicode, 114 Javascript, 26
Disjunction, 16
Dual system, 14
L
Lambda-calculus, 21
E Learn, 74
Earthquake, 123 overlearn, 74
E-commerce, 86 prediction accuracy, 74
Amazon, 86 test dataset, 74
ebay, 86, 89 training data set, 74
Marketplace, 86 validation dataset, 74
market transparency, 86 Legal, 187
YouTube, 88 Legal Tech, 188
Electronic file, 188 Leibniz, 13
Equivalence, 16 LINUX, 25
Exclusive-or, 16 Logics, 15

F M
Fisheye projection, 136 Magnetic Resonance Imaging (MRI), 125
Mersenne’s prime, 2
Moore’s Law, 1
G
Game theory, 63
minimax theorem, 63 N
n-person games, 63 Negation, 16
Gravitational wave, 117 Netflix, 89
Network
feed-forward network, 73
H feedback, 73
HTML, 26 fully-connected, 73
HTTP, 26 recurrent network, 73
short-cuts, 73
Neural network, 69, 74
I folding network, 74
Image processing Neuron, 69
brain tumor, 125 action potential, 72
pixel, 133 membrane potential, 72
Information theory, 63 threshold, 69
complete information, 63 weight, 72
Intelligence, 5
Artificial Intelligence, 5
collective intelligence, 7 P
definition, 5 Pearson’s rank correlation, 92
emotional intelligence, 6 Perceptron, 72
environmental intelligence, 6 single perceptron, 72
human intelligence, 6 Polar coordinates, 136
Index 219

Problem of the traveling salesman, 60 R

Programming, 20 Robotics, 147
logical programming, 25
object-oriented programming, 24
Programming Language S
ADA, 25 Sampling, 116
ALGOL, 22 bit depth, 116
Backus-Naur Form (BNF), 23 Satisfiability problem, 47
BASIC, 23 Shannon, C., 17
C, 23 Standard deviation, 92
C++, 25 Supercomputer, 18, 32
COBOL, 22
FORTRAN, 22
LISP, 22 T
MODULA, 24 Translation, 78
PASCAL, 23 Braille, 81
PL/1, 23 DeepL, 78
PROLOG, 25 Google, 80
SCHEME, 23 Tripadvisor, 89
SIMULA, 24 Turing award, 27
SMALLTALK, 24 Turing-machine, 27
Program testing, 20 final-state, 29
Psychology, 6 single-band machine, 29
cognitive psychology, 6 start-state, 29
developmental psychology, 6 Turing-test, 27, 28
differential psychology, 6
general psychology, 6
neuropsychology, 6
U
URL, 26
Q
Quantumbit, 34
Quantum computer, 34
W
quantum entanglement, 32
Web server, 26
superposition principle, 32