115,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Wissenschaft und neue Technologien
- Sprache: Englisch
CYBERNETICAL INTELLIGENCE Highly comprehensive, detailed, and up-to-date overview of artificial intelligence and cybernetics, with practical examples and supplementary learning resources Cybernetical Intelligence: Engineering Cybernetics with Machine Intelligence is a comprehensive guide to the field of cybernetics and neural networks, as well as the mathematical foundations of these technologies. The book provides a detailed explanation of various types of neural networks, including feedforward networks, recurrent neural networks, and convolutional neural networks as well as their applications to different real-world problems. This groundbreaking book presents a pioneering exploration of machine learning within the framework of cybernetics. It marks a significant milestone in the field's history, as it is the first book to describe the development of machine learning from a cybernetics perspective. The introduction of the concept of "Cybernetical Intelligence" and the generation of new terminology within this context propel new lines of thought in the historical development of artificial intelligence. With its profound implications and contributions, this book holds immense importance and is poised to become a definitive resource for scholars and researchers in this field of study. Each chapter is specifically designed to introduce the theory with several examples. This comprehensive book includes exercise questions at the end of each chapter, providing readers with valuable opportunities to apply and strengthen their understanding of cybernetical intelligence. To further support the learning journey, solutions to these questions are readily accessible on the book's companion site. Additionally, the companion site offers programming practice exercises and assignments, enabling readers to delve deeper into the practical aspects of the subject matter. Cybernetical Intelligence includes information on: * The history and development of cybernetics and its influence on the development of neural networks * Developments and innovations in artificial intelligence and machine learning, such as deep reinforcement learning, generative adversarial networks, and transfer learning * Mathematical foundations of artificial intelligence and cybernetics, including linear algebra, calculus, and probability theory * Ethical implications of artificial intelligence and cybernetics as well as responsible and transparent development and deployment of AI systems Presenting a highly detailed and comprehensive overview of the field, with modern developments thoroughly discussed, Cybernetical Intelligence is an essential textbook that helps students make connections with real-life engineering problems by providing both theory and practice, along with a myriad of helpful learning aids.
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 705
Veröffentlichungsjahr: 2023
Ähnliche
Table of Contents
Cover
Table of Contents
Series Page
Title Page
Copyright Page
Preface
About the Author
About the Companion Website
1 Artificial Intelligence and Cybernetical Learning
1.1 Artificial Intelligence Initiative
1.2 Intelligent Automation Initiative
1.3 Artificial Intelligence Versus Intelligent Automation
1.4 The Fourth Industrial Revolution and Artificial Intelligence
1.5 Pattern Analysis and Cognitive Learning
1.6 Cybernetical Artificial Intelligence
1.7 Cybernetical Intelligence Definition
1.8 The Future of Cybernetical Intelligence
Summary
Exercise Questions
Further Reading
2 Cybernetical Intelligent Control
2.1 Control Theory and Feedback Control Systems
2.2 Maxwell’s Analysis of Governors
2.3 Harold Black
2.4 Nyquist and Bode
2.5 Stafford Beer
2.6 James Lovelock
2.7 Macy Conference
2.8 McCulloch–Pitts
2.9 John von Neumann
Summary
Exercise Questions
Further Reading
3 The Basics of Perceptron
3.1 The Analogy of Biological and Artificial Neurons
3.2 Perception and Multilayer Perceptron
3.3 Activation Function
Summary
Exercise Questions
Further Reading
4 The Structure of Neural Network
4.1 Layers in Neural Network
4.2 Perceptron and Multilayer Perceptron
4.3 Recurrent Neural Network
4.4 Markov Neural Networks
4.5 Generative Adversarial Network
Summary
Exercise Questions
Further Reading
5 Backpropagation Neural Network
5.1 Backpropagation Neural Network
5.2 Gradient Descent
5.3 Stopping Criteria
5.4 Resampling Methods
5.5 Optimizers in Neural Network
Summary
Exercise Questions
Further Reading
6 Application of Neural Network in Learning and Recognition
6.1 Applying Backpropagation to Shape Recognition
6.2 Softmax Regression
6.3
K
‐Binary Classifier
6.4 Relational Learning via Neural Network
6.5 Cybernetics Using Neural Network
6.6 Structure of Neural Network for Image Processing
6.7 Transformer Networks
6.8 Attention Mechanisms
6.9 Graph Neural Networks
6.10 Transfer Learning
6.11 Generalization of Neural Networks
6.12 Performance Measures
Summary
Exercise Questions
Further Reading
7 Competitive Learning and Self‐Organizing Map
7.1 Principal of Competitive Learning
7.2 Basic Structure of Self‐Organizing Map
7.3 Self‐Organizing Mapping Neural Network Algorithm
7.4 Growing Self‐Organizing Map
7.5 Time Adaptive Self‐Organizing Map
7.6 Oriented and Scalable Map
7.7 Generative Topographic Map
Summary
Exercise Questions
Further Reading
8 Support Vector Machine
8.1 The Definition of Data Clustering
8.2 Support Vector and Margin
8.3 Kernel Function
8.4 Linear and Nonlinear Support Vector Machine
8.5 Hard Margin and Soft Margin in Support Vector Machine
8.6 I/O of Support Vector Machine
8.7 Hyperparameters of Support Vector Machine
8.8 Application of Support Vector Machine
Summary
Exercise Questions
Further Reading
9 Bio‐Inspired Cybernetical Intelligence
9.1 Genetic Algorithm
9.2 Ant Colony Optimization
9.3 Bees Algorithm
9.4 Artificial Bee Colony Algorithm
9.5 Cuckoo Search
9.6 Particle Swarm Optimization
9.7 Bacterial Foraging Optimization
9.8 Gray Wolf Optimizer
9.9 Firefly Algorithm
Summary
Exercise Questions
Further Reading
10 Life‐Inspired Machine Intelligence and Cybernetics
10.1 Multi‐Agent AI Systems
10.2 Cellular Automata
10.3 Discrete Element Method
10.4 Smoothed Particle Hydrodynamics
Summary
Exercise Questions
Further Reading
11 Revisiting Cybernetics and Relation to Cybernetical Intelligence
11.1 The Concept and Development of Cybernetics
11.2 The Fundamental Ideas of Cybernetics
11.3 Cybernetic Expansion into Other Fields of Research
11.4 Practical Application of Cybernetics
Summary
Exercise Questions
Further Reading
12 Turing Machine
12.1 Behavior of a Turing Machine
12.2 Basic Operations of a Turing Machine
12.3 Interchangeability of Program and Behavior
12.4 Computability Theory
12.5 Automata Theory
12.6 Philosophical Issues Related to Turing Machines
12.7 Human and Machine Computations
12.8 Historical Models of Computability
12.9 Recursive Functions
12.10 Turing Machine and Intelligent Control
Summary
Exercise Questions
Further Reading
13 Entropy Concepts in Machine Intelligence
13.1 Relative Entropy of Distributions
13.2 Relative Entropy and Mutual Information
13.3 Entropy in Performance Evaluation
13.4 Cross‐Entropy Softmax
13.5 Calculating Cross‐Entropy
13.6 Cross‐Entropy as a Loss Function
13.7 Cross‐Entropy and Log Loss
13.8 Application of Entropy in Intelligent Control
Summary
Exercise Questions
Further Reading
14 Sampling Methods in Cybernetical Intelligence
14.1 Introduction to Sampling Methods
14.2 Basic Sampling Algorithms
14.3 Machine Learning Sampling Methods
14.4 Advantages and Disadvantages of Machine Learning Sampling Methods
14.5 Advanced Sampling Methods in Cybernetical Intelligence
14.6 Applications of Sampling Methods in Cybernetical Intelligence
14.7 Challenges and Future Directions
14.8 Challenges and Limitations of Sampling Methods
14.9 Emerging Trends and Innovations in Sampling Methods
Summary
Exercise Questions
Further Reading
15 Dynamic System Control
15.1 Linear Systems
15.2 Nonlinear System
15.3 Stability Theory
15.4 Observability and Identification
15.5 Controllability and Stabilizability
15.6 Optimal Control
15.7 Linear Quadratic Regulator Theory
15.8 Time‐Optimal Control
15.9 Stochastic Systems with Applications
Summary
Exercise Questions
Further Reading
16 Deep Learning
16.1 Neural Network Models in Deep Learning
16.2 Methods of Deep Learning
16.3 Deep Learning Frameworks
16.4 Applications of Deep Learning
Summary
Exercise Questions
References
Further Reading
17 Neural Architecture Search
17.1 Neural Architecture Search and Neural Network
17.2 Reinforcement Learning‐Based Neural Architecture Search
17.3 Evolutionary Algorithms‐Based Neural Architecture Search
17.4 Bayesian Optimization‐Based Neural Architecture Search
17.5 Gradient‐Based Neural Architecture Search
17.6 One‐shot Neural Architecture Search
17.7 Meta‐Learning‐Based Neural Architecture Search
17.8 Neural Architecture Search for Specific Domains
17.9 Comparison of Different Neural Architecture Search Approaches
Summary
Exercise Questions
Further Reading
Final Notes on
Cybernetical Intelligence
Index
End User License Agreement
List of Tables
Chapter 1
Table 1.1 Feature‐wise difference between parametric and nonparametric algo...
Chapter 2
Table 2.1 Key aspects of control theory and feedback control systems.
Chapter 3
Table 3.1 Summarizing the key differences between artificial and biological...
Table 3.2 Summarizing the differences between perception and multilayer per...
Table 3.3 Comparative of various activation functions in neural network.
Chapter 4
Table 4.1 Summarizing the different layers that can be present in a neural ...
Chapter 5
Table 5.1 Various parameters involved in neural network training.
Table 5.2 Summarizing some common optimizers used in neural network trainin...
Chapter 8
Table 8.1 Various kernels used in SVM.
Table 8.2 Summarizing the hyperparameters of support vector machine.
Chapter 10
Table 10.1 Representation of the payoff matrix.
Chapter 11
Table 11.1 The yarns of game theory.
Chapter 12
Table 12.1 Table summarizing the parameters of a Turing machine.
Table 12.2 Summarizing some key concepts in computability theory.
Chapter 14
Table 14.1 Comparing different sampling methods used in machine learning.
Chapter 17
Table 17.1 Comparing neural architecture search and neural networks.
List of Illustrations
Chapter 1
Figure 1.1 Workflow of intelligent automation.
Figure 1.2 Examples of parametric algorithms along with their characteristic...
Figure 1.3 Examples of nonparametric algorithms along with their characteris...
Figure 1.4 Historical overview and development of cybernetics.
Chapter 2
Figure 2.1 Understanding the concept of control system.
Figure 2.2 McCulloch and Pitts neuron model.
Chapter 3
Figure 3.1 Analogy of real neurons and artificial neurons.
Figure 3.2 The structure of the neural network.
Figure 3.3 Schematic diagram of neural network.
Figure 3.4 Sigmoid activation function.
Figure 3.5 Tangent activation function.
Figure 3.6 Rectified linear unit.
Figure 3.7 Linear activation function.
Chapter 4
Figure 4.1 Architecture of recurrent neural network.
Chapter 5
Figure 5.1 Principle diagram of a cybernetic system with a feedback loop.
Figure 5.2 Process of Backpropagation Neural Network.
Figure 5.3 Procedure of gradient descent.
Figure 5.4 Graphical representation for loss function.
Figure 5.5 Local minimum and global minimum in peaks.
Figure 5.6 Data splitting scheme from the initial data split to resampling....
Figure 5.7 Optimizers in neural network.
Chapter 6
Figure 6.1 Illustration of simple 4‐layer neural network.
Figure 6.2 Schematic diagram of RNs.
Figure 6.3 A single layer of a simple GNN. A graph is the input, and each co...
Figure 6.4 The CNN on the left side and GCN on the right side.
Figure 6.5 Cybernetical layout of the neuron based on perception theory.
Figure 6.6 Schematic diagram of neural network in overfitting, optimal, and ...
Figure 6.7 Structure of typical confusion matrix.
Figure 6.8 Visual representation of schematic diagram for ROC‐ACU.
Chapter 7
Figure 7.1 SOM network structure diagram.
Figure 7.2 Schematic diagram of the two‐dimensional rectangular plane SOM.
Figure 7.3 Node growth options in GSOM: (a) one new node, (b) two new nodes,...
Figure 7.4 Generative topographic map.
Chapter 8
Figure 8.1 Schematic diagram of Euclidean distance.
Figure 8.2 Schematic diagram of Manhattan distance.
Figure 8.3 Procedure of support vector machine.
Figure 8.4 Sample dataset for Linear SVM.
Figure 8.5 Hard margin in support vector machine.
Chapter 9
Figure 9.1 Initialization of genetic algorithm.
Figure 9.2 Example of genetic algorithm. (a) represents the crossover point,...
Figure 9.3 Example of the procedural steps from (a) to (b), and then to (c) ...
Figure 9.4 Flowchart of the bees algorithm.
Figure 9.5 The flowchart for the artificial bee colony.
Figure 9.6 The flowchart for the cuckoo search algorithm.
Figure 9.7 Flowchart for the particle swarm optimization.
Figure 9.8 Flowchart for the bacterial foraging optimization.
Figure 9.9 Flowchart for the gray wolf optimizer.
Figure 9.10 Flowchart for the firefly algorithm.
Chapter 10
Figure 10.1 Workflow of multi‐agent system.
Figure 10.2 Workflow of discrete element method.
Figure 10.3 Workflow of SPH.
Chapter 11
Figure 11.1 The structure and development of cybernetics. SISO stands for Si...
Figure 11.2 Cybernetic dichotomies of behavior.
Figure 11.3 The flow chart of Re
8
analysis for cybernetical intelligence neu...
Figure 11.4 Software testing as a control problem in cybernetics.
Figure 11.5 PCBTA modeling process.
Figure 11.6 The relationship between perceptual control system and environme...
Chapter 14
Figure 14.1 Categories of sampling methods.
Figure 14.2 Geometrical interpretation of the transformation method for gene...
Figure 14.3 Difference between original and sample data.
Figure 14.4 The generic overview of the sampling technique applied to the da...
Figure 14.5 Overview and difference between undersampling and oversampling....
Figure 14.6 Basic principle of synthetic minority oversampling technique.
Figure 14.7 Generic working of ADASYN algorithm.
Figure 14.8 Workflow of ensemble sampling technique.
Figure 14.9 Ensemble learning bagging vs boosting.
Figure 14.10 Process of active learning in machine learning and computer vis...
Figure 14.11 Workflow of Bayesian optimization.
Figure 14.12 Schematic of Bayesian optimization framework.
Chapter 16
Figure 16.1 Basic convolutional neural networks structure.
Figure 16.2 Network structure of ConvLSTM.
Figure 16.3 The basic recurrent neural networks.
Figure 16.4 The general workflow of GAN.
Figure 16.5 Structure diagram of p‐GANs generator
G
network.
Figure 16.6 Network structure diagram of p‐GANs discriminator
D
.
Figure 16.7 Architecture of the p‐GANs.
Figure 16.8 Structural principle of U‐Net.
Figure 16.9 Schematic diagram of the network structure principle of 3D resid...
Figure 16.10 The general architecture of variational autoencoders.
Figure 16.11 The overall architecture of transformer encoder.
Figure 16.12 The basic architecture of attention based models.
Figure 16.13 Generic architecture of capsule network.
Chapter 17
Figure 17.1 The generic working of NAS.
Figure 17.2 NAS implementation using reinforcement learning.
Figure 17.3 An example of the two types of mutations in evolutionary NAS.
Figure 17.4 Flow chart of Bayesian optimization.
Figure 17.5 One‐shot neural network model.
Figure 17.6 Meta learning for NAS.
Figure 17.7 Generic overview of intelligent control system.
Guide
Cover Page
Series Page
Title Page
Copyright Page
Preface
About the Author
About the Companion Website
Table of Contents
Begin Reading
Final Notes on Cybernetical Intelligence
Index
WILEY END USER LICENSE AGREEMENT
Pages
ii
iii
xv
xvi
xvii
xix
xxi
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
69
70
71
72
73
74
75
76
77
78
79
80
81
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
399
400
401
402
403
404
405
406
407
408
409
IEEE Press445 Hoes LanePiscataway, NJ 08854
IEEE Press Editorial BoardSarah Spurgeon, Editor in Chief
Jón Atli Benediktsson
Behzad Razavi
Jeffrey Reed
Anjan Bose
Jim Lyke
Diomidis Spinellis
James Duncan
Hai Li
Adam Drobot
Amin Moeness
Brian Johnson
Tom Robertazzi
Desineni Subbaram Naidu
Ahmet Murat Tekalp
Cybernetical Intelligence
Engineering Cybernetics with Machine Intelligence
Kelvin K. L. Wong
Copyright © 2024 by The Institute of Electrical and Electronics Engineers, Inc. All rights reserved.
Published by John Wiley & Sons, Inc., Hoboken, New Jersey.Published simultaneously in Canada.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per‐copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750‐8400, fax (978) 750‐4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748‐6011, fax (201) 748‐6008, or online at http://www.wiley.com/go/permission.
Trademarks: Wiley and the Wiley logo are trademarks or registered trademarks of John Wiley & Sons, Inc. and/or its affiliates in the United States and other countries and may not be used without written permission. All other trademarks are the property of their respective owners. John Wiley & Sons, Inc. is not associated with any product or vendor mentioned in this book.
Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.
For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762‐2974, outside the United States at (317) 572‐3993 or fax (317) 572‐4002.
Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com.
Library of Congress Cataloging‐in‐Publication Data applied for:
Hardback: 9781394217489
Cover Design: WileyCover Image: Courtesy of author
Preface
Life evolves into existence from the edge of chaos and builds up a new mechanism based on a set of rules that governs the law of survival, reproduction, and evolution. This complex set of rules that allows a living thing to interact with its environment comes into being from the beginning of life itself, which is something one can understand as intelligence. All the wonders of art, design, sciences, etc., in the world make us ponder upon the question of the ages, on the origin of creation and the existence of life itself and the evolution of intelligence that comes into being.
“The important thing is not to stop questioning,” is a famous quotation by Albert Einstein from 1955. A robot gains intelligence by questioning and seeking answers, which form examples for the labels for the unseen examples. However, will the robot have a creative mind similar to that of Einstein? This leads us to the question of whether creativity can be programmed. Can an analogy be bridged between the robot's experience in developing multiple search way paths for the optimal solution and an intelligent being's intuition to design creative solutions?
Artificial intelligence (AI) is built on the pillars of a few major branches of science and engineering, namely, systematology, information theory, and cybernetics, which is typically based on control theory that was derived from the studies of Norbert Wiener, the world‐renowned father of cybernetics. In 1954, Hsue‐sen Tsien founded engineering cybernetics by publishing the famous engineering cybernetics in America. On the basis of cybernetics, a predictive system may be regarded as a multiple feedback system. The framework of a multilayer perceptron as well as that of a backpropagation neural network can be based on the theoretic of system control in modern cybernetics. With this type of thinking, perceptron theory offers a cohesive approach to the statistical mechanics and principles of cybernetics as a basis for the successful neural network modeling.
A feedback controller's operation is to change the behavior of a system fundamentally. Feedback control systems sample a system's outputs, compare them to a set of desired outputs, and then utilize the resulting error signals to compute the system's control inputs in such a way that the errors are minimized. Artificially built feedback control systems, which are utilized to govern industrial, automotive, and aeronautical systems, are responsible for today's aerospace achievements. Biological systems are full of naturally occurring feedback controls. The cell, one of the most basic of all life forms, regulates the potential difference across the cell membrane to preserve homeostasis. Although neural network controllers are adaptive learning systems, they do not need the conventional assumptions of adaptive control theory, such as parameter linearity and the presence a regression matrix. It is demonstrated in detail the process to create neural network based controllers for cybernetical systems, a general category of nonlinear systems, complicated industrial systems with vibrations and flexibility effects, force control, motor dynamics control, and other applications. These are given for both continuous‐time and discrete‐time weight tuning.
Integration of AI and cybernetics can produce applications in predictive control, pattern recognition, and classification, which essentially are based on the same fundamentals. This book proposes for the first time the novel perspective of machine intelligence, which is termed as Cybernetical Intelligence. Such a new field will have extensive and practical applications in not just the combinatorial optimization problems but also in pattern recognition, data mining, and other related machine intelligence based cybernetics problems.
The key concept of Cybernetical Intelligence grew from a desire to understand and build systems that can achieve goals, whether complex human goals or just goals. It is even deeper underlying conceptual term. Cybernetics holds the world sufficiently to gain feedback in order to correct the actions to achieve goals. It is mutual combination of automated communication and control system between artificially intelligent machines and the environment with subsequent strong support from machine learning; the concepts of systems thinking and systems theory became integral parts of the established scientific language of Cybernetical Intelligence and can lead to numerous new methodologies and applications.
The basic ideas of Cybernetical Intelligence can be treated without reference to electronics, but they are fundamentally challenging; so although advanced techniques may be necessary for advanced applications, a great deal can be done, especially in biological sciences, by the use of mathematical derivations, provided they are used with a clear and deep understanding of the principals involved.
This book is intended to provide a concise conceptualization of Cybernetical Intelligence. It starts from common place and well‐understood concepts and proceeds, step by step, to show how these concepts can be made exact and how they can be developed until they lead into subjects such as feedback, stability, regulation, ultrastability, information, coding, noise, and other cybernetic topics. Closed‐loop applications and features of neural network are examined and developed in great detail in this book, employing mathematical stability proof approaches that illustrate how to construct neuro‐controllers while also ensuring their stability and performance. Control engineering based concepts, a family of multi‐loop neuro‐controllers for various applications have been created methodically.
There are strategies for both continuous‐time and discrete‐time weight tuning given. The book is intended for students taking a second semester in control theory, as well as engineers in academia and industry who construct feedback controllers for complex systems found in commercial, industrial, and military applications. The many types of neuro‐controllers are organized in tables for simple reference when it comes to design procedures.
This material is a comprehensive exploration of the advanced terminologies in AI and cybernetics. In Chapter 1, the concept of AI and its relation to cybernetics are introduced. Chapter 2 delves into the theory of cybernetical intelligence and control. Chapter 3 covers the basics of perceptron, including its activation function. The structure of the multilayer perceptron neural network is discussed in Chapter 4, while Chapter 5 covers the backpropagation algorithm and its derivatives, as well as the resampling rate. Chapter 6 focuses on neural network applications in learning and recognition. Chapter 7 explores self‐organizing and its applications in AI, and Chapter 8 covers support vector machines and their applications. Chapters 9 and 10 delve into bio‐ and life‐inspired Cybernetical Intelligence. Chapters 11 and 12 revisit cybernetics and its relation to Cybernetical Intelligence and Turing machines. Entropy concepts and sampling methods in Cybernetical Intelligence are covered in Chapters 13 and 14. Chapters 15 and 16 describe linear systems and deep learning, including their methods and applications. Finally, Chapter 17 focuses on neural architecture search, including its methods and applications. Every chapter presents its own characteristic concept, and the concatenation of these concepts generates a mind map and general framework for the formulation of machine learning from the cybernetics perspective and encompassing the Cybernetical Intelligence philosophy. The philosophical insights and mathematical theories in this book will give us the adequate knowledge necessary for building AI.
It is the author’s belief that the subject founded is well understood and is then built up carefully, step by step, with advanced mathematical, computing, and engineering knowledge. Having spent years consolidating and developing the conceptual roadmap of machine learning from the cybernetics perspective, the author is proud to present the novel work on Cybernetical Intelligence to the academic community with the ultimate aim of training the next generation of AI cybernetists.
About the Author
Prof. Dr. Kelvin K. L. Wong is a distinguished expert in medical image processing and computational science, who earned his Ph.D. from The University of Adelaide. With a strong academic background from Nanyang Technological University and The University of Sydney, he has been at the forefront of merging the fields of cybernetics and artificial intelligence (AI). He is widely recognized for introducing the term “Cybernetical Intelligence” and is the inventor and founder of the Deep Red AI system. Dr. Wong's impactful research in AI has yielded significant achievements with the potential to positively impact humanity. He is the author of influential books such as Methods in Research and Development of Biomedical Devices and Computational Hemodynamics–Theory, Modelling, and Applications. With extensive experience as an associate editor and guest editor for esteemed biomedical engineering and computational intelligence journals, he has contributed extensively to the field. As an internationally recognized biomedical engineering scientist and AI cybernetist, Dr. Wong was named among Stanford University's top 1.3% biomedical engineering researchers in 2020. He has actively participated in researching the management and control of COVID‐19 and is a dedicated supporter and donor to UNICEF, advocating for kindness and human rights. Throughout his professorship, he has mentored numerous students, providing invaluable guidance and shaping their careers. Leading a team of experts in AI, healthcare, disease management, and diagnosis, Dr. Wong's expertise has been instrumental in supporting government projects and initiatives in developing countries.
About the Companion Website
This book is accompanied by a companion website:
www.wiley.com/go/cyberintel
This website includes:
Assignments and Solutions
1Artificial Intelligence and Cybernetical Learning
Artificial intelligence (AI) is a field of engineering cybernetics that focuses on the development of intelligent machines that can simulate human‐like behaviors such as learning, problem‐solving, reasoning, and decision‐making. AI technology involves the use of algorithms and computational models to analyze vast amounts of data, recognize patterns and make predictions, and interact with humans through natural language processing (NLP) and other forms of communication.
This chapter will comprehensively explore various aspects of AI, including its relation to cybernetics and the fundamental principles governing it. Additionally, it will delve into the nuances of parametric and nonparametric algorithms and core concepts of cybernetical intelligence (CI). Through a systematic and rigorous exposition, readers will acquire a robust understanding of the key principles and algorithms that underlie AI. Consequently, they will be well equipped with the requisite knowledge to develop their own AI applications, leveraging the insights gained from this chapter.
1.1 Artificial Intelligence Initiative
Intelligence includes the capacity for abstraction, logic, learning, reasoning, communication, and inference. It can learn from the environment both actively and passively and use the knowledge to obtain adaptive ability. AI can be defined as a human‐made machine with human‐like intelligence. The use of AI in education has produced effective pedagogical effects in addition to technical advancements and theoretical developments. Automated target identification, automatic medical diagnosis, and audio recording are a few interesting uses. AI may be utilized to provide customized assistance and increase knowledge‐gap awareness, allowing educators to deliver individualized and adaptable education with efficiency and effectiveness. Enabling computers to simulate intelligent behavior using prestored world models is the main goal of AI. The AI simulates human cognitive processes such as reasoning, learning, pattern recognition, knowledge reasoning, and machine learning (ML). ML refers to the creation of automated systems capable of processing massive volumes of data for data mining and is one of the more traditional fields of computing intelligence.
ML is a part of AI that allows machines to obtain intelligence from data without being explicitly programmed. Therefore, it often associates ML with data mining. ML enables a cyber system to possess intelligence by using massive data. Based on that data, ML models or algorithms can mine the knowledge, rules, and laws behind the data. ML identifies underlying functional links in systems between sets of variables and individual variables. The goal of combining the fields of ML and cybernetics is to identify different ways that systems interact with one another through various methods for learning from data. Equation (1.1) illustrates how ML may be summed up as learning a function (f) that maps input variables (x) to output variables (y).
The configuration of the function is unknown, but the ML algorithms learn to map the target function from the training data. It is necessary to assess many algorithms to determine which one is best at modeling the underlying function because they all reach different conclusions or exhibit biases on the function's structure. In theory, there are two types of ML algorithms: parametric algorithms and nonparametric algorithms. Additionally, three well‐known techniques are used to train ML algorithms. Supervised learning, unsupervised learning, and reinforcement learning are the three categories of ML.
The most significant methodology in ML is supervised learning, which is especially crucial in the processing of multimedia data. This kind of learning is comparable to how humans learn from their past experiences to obtain new information and improve their capacity to carry out activities in the actual world. Models for supervised learning are designed to predict the appropriate label for newly presented data. Unsupervised learning is typically used to identify patterns in the input data that propose candidate features prior to the application of supervised learning, and feature engineering changes these candidate features to make them more appropriate for supervised learning. It is quite time‐consuming to identify the correct category or response for every observation in the training set in addition to the characteristics. With the help of semi‐supervised learning, one may train models with very little labeled data, which will reduce the labeling work.
Unsupervised learning can be motivated from information‐theoretic and Bayesian principles. It empowers the model to work independently to identify previously unnoticed patterns and information. Take into account a device (or living thing) that gets a series of inputs, such as x1, x2, …, xt, where xt is the sensory input at time t. This input, which is known as sensory data, can be a representation of a retinal image, a camera's pixels, or a sound waveform. The most famous technique is clustering in which each observation belongs to at least one of the k clusters, while i and j belong to centroid of each cluster. Furthermore, variation within each cluster is achieved by minimizing the sum of the squared Euclidean distance between each observation within a cluster, as shown in Equation (1.2).
where μk represents the centroid of the kth cluster, Xi is the ith data point in the kth cluster, and Ck represents the set of indices of data points assigned to the kth cluster. Reinforcement learning, on the other hand, is heavily influenced by the theory of Markov decision processes and deals with the ability to learn the associations between stimuli, actions, and the occurrence of positive events. The agents are taught a reward and punishment scheme in reinforcement learning. For wise actions, the agent is rewarded, and for poor ones, they are penalized. While doing this, the agent tries to minimize the undesirable motions while maximizing the desirable ones. It is hardly unexpected that reinforcement learning has been noticed in the really distant past given its clear adaptive benefit. A few cybernetics experiments have made use of reinforcement learning. Robots can learn skills that a human instructor is unable to teach, adapt a learned ability to a new task, and accomplish optimization even in the absence of an analytical formulation with the help of this sort of ML. The predicted total of the immediate reward and the long‐term reward under the best feasible policy (Max Policies), as given in Equation (1.3), is utility u (over a limited agent lifespan):
where st is the state at time step t, R (st, a) is the immediate reward of executing an action in state st, N is the number of steps in the lifetime of the agent, and R is the reward time step t. The operator stands for taking an expectation over all sources of randomness in the system. Here, st denotes the state at time step t, R(st, a) is the instantaneous benefit of carrying out an action in state st, and N denotes the total number of steps the agent will take throughout its lifespan. Taking an expectation across all system randomness sources is what the operator. The configuration of the function is unknown, but the ML algorithms learn to map the target function from the training data. It is necessary to compare multiple algorithms to determine which one is the best successful at modeling the underlying function since different algorithms reach different conclusions or have different biases on the structure of the function. As a result, ML algorithms may be divided into parametric and nonparametric varieties, which will be covered in the following subsections.
1.2 Intelligent Automation Initiative
Intelligent automation initiative (IAI) is an emerging technology‐driven approach to optimize business processes and decision‐making through a combination of AI, robotic process automation (RPA), and other advanced technologies. It aims to streamline repetitive and mundane tasks, improve productivity, reduce errors, and enable employees to focus on higher‐value‐added activities. The IAI strategy involves the integration of different technologies to automate various aspects of the business, including customer service, supply chain management, finance, human resources, and more. The main components of IAI include:
Artificial intelligence (AI): A subset of computer science that focuses on developing algorithms that can mimic human intelligence, such as speech recognition, NLP, ML, and computer vision. AI helps organizations to make sense of vast amounts of data, predict trends, and make informed decisions.
Robotic process automation (RPA): A software tool that uses bots to automate repetitive and rule‐based tasks, such as data entry, invoice processing, and report generation. RPA can reduce operational costs, improve accuracy, and increase efficiency.
Advanced analytics: It involves the use of statistical models, data mining, and predictive analytics to analyze data and extract insights. This can help organizations to make informed decisions and improve business outcomes.
Chatbots: AI‐powered virtual assistants that can interact with customers, answer queries, and resolve issues in real time. Chatbots can improve customer satisfaction, reduce response times, and free up resources for other tasks.
Machine learning: A subset of AI that focuses on developing algorithms that can learn from data without being explicitly programmed. ML can be used to make predictions, identify patterns, and automate decision‐making.
Cognitive automation: Involves the use of AI and other advanced technologies to automate complex tasks that require human‐like reasoning and decision‐making. This can include tasks such as fraud detection, risk analysis, and supply chain optimization.
1.2.1 Benefits of IAI
IAI is a strategic approach to integrating advanced technologies, such as AI, RPA, and ML, to automate business processes and workflows. Here are some of the benefits of implementing IAI:
Increased productivity: Automation can perform repetitive and time‐consuming tasks faster and with fewer errors than humans, leading to increased productivity and efficiency. By freeing up employees from these mundane tasks, they can focus on higher‐value tasks that require creativity and critical thinking.
Cost savings: Automation can help reduce labor costs, as companies no longer need to hire additional staff to perform repetitive tasks. Additionally, automation can help reduce operational costs by streamlining processes and reducing the potential for errors and delays.
Improved accuracy and quality: Automation can perform tasks with a high degree of accuracy, consistency, and quality, reducing the potential for errors and improving the quality of work produced.
Faster processing times: Automation can help speed up processing times for tasks such as data entry, data analysis, and report generation. This can lead to faster decision‐making and improved business agility.
Enhanced customer experience: Automation can help improve the customer experience by enabling faster response times to inquiries, reducing errors, and providing more accurate and personalized services.
Increased scalability: Automation can help businesses scale their operations more easily by enabling them to handle higher volumes of work without the need for additional staff.
Better data insights: Automation can help businesses gather and analyze data more quickly and accurately, enabling them to make better‐informed decisions.
Overall, the benefits of IAI can help businesses streamline their operations, reduce costs, and improve their ability to compete in an increasingly fast‐paced and competitive market.
1.3 Artificial Intelligence Versus Intelligent Automation
AI and intelligent automation (IA) are two related technologies that are transforming the way businesses operate. AI is the simulation of human intelligence processes by machines, while IA refers to the automation of processes using AI and other advanced technologies. IA combines RPA, ML, and other AI technologies to automate repetitive and time‐consuming tasks. It allows businesses to automate processes that were previously done manually, which saves time, reduces costs, and improves accuracy.
AI, on the other hand, is a broader field that encompasses a range of technologies, including ML, NLP, and computer vision. These technologies enable machines to perform tasks that would typically require human intelligence, such as understanding language, recognizing images, and making decisions based on data. When AI and IA are combined, businesses can achieve even greater benefits. For example, IA can be used to automate processes such as data entry and document processing, while AI can be used to analyze that data and provide insights for decision‐making. This can help businesses make more informed decisions faster, which can lead to improved efficiency, productivity, and profitability. Moreover, AI can help automate decision‐making processes by analyzing vast amounts of data and providing recommendations based on that data. IA can then be used to execute those decisions automatically, further streamlining business processes. The complete workflow of how IA works is shown in Figure 1.1.
1.3.1 Process Discovery
Process discovery involves using mathematical equations and algorithms to analyze business processes and identify areas where automation can be applied. One example of a mathematical equation used in process discovery is the process cycle efficiency (PCE), as shown in Equation (1.4).
where value‐added time (VT) is the time spent on activities that directly add value to the customer, and cycle time (CT) is the total time taken to complete the process, including both value‐added and non‐value‐added activities. The PCE formula helps businesses identify areas where there is wastage or inefficiency in the process. A high PCE indicates that a process is highly efficient and that there is minimal wastage, while a low PCE suggests that there is a lot of wastage that can be eliminated through automation.
The first step in process discovery is to collect data on the current business processes. This can be done by conducting interviews with key stakeholders, analyzing documentation such as process maps, or observing the processes in action. Once the data has been collected, the next step is to map out the processes using visual diagrams such as flowcharts or swim lane diagrams. This helps to identify the different steps involved in the process, as well as the inputs and outputs at each stage. After the process has been mapped out, the next step is to analyze it in detail. This involves looking for inefficiencies or bottlenecks in the process that could be improved through automation or optimization. Based on the analysis, the process can be optimized by identifying areas where automation can be applied to reduce manual effort, speed up processing times, or reduce errors. This may involve using RPA to automate repetitive tasks or using ML or AI to analyze data and make predictions about future outcomes.
Figure 1.1 Workflow of intelligent automation.
Once the process has been optimized, it is important to test it thoroughly to ensure that it works as intended. This may involve conducting user acceptance testing (UAT) or running pilot programs to ensure that the process is reliable and effective. Once the testing is complete, the optimized process can be implemented into production. Overall, process discovery is an essential part of IA, as it helps organizations identify and optimize their existing business processes, reducing costs, increasing efficiency, and improving the overall customer experience. By leveraging technologies, such as RPA, ML, and AI, organizations can automate repetitive tasks, make better decisions based on data, and streamline their operations to stay competitive in a rapidly evolving business landscape.
1.3.2 Optimization
Optimization is an important aspect of IA. It refers to the process of improving the efficiency and effectiveness of automated processes over time by continuously analyzing and refining them. Optimization involves using advanced technologies such as ML and AI to analyze data generated by automated processes. By analyzing this data, businesses can identify areas where the process can be improved and make adjustments to improve efficiency and effectiveness.
One example of how optimization can be achieved is with predictive analytics. Predictive analytics uses statistical algorithms and ML techniques to analyze historical data and make predictions about future outcomes. By using predictive analytics, businesses can identify potential problems in their automated processes before they occur, allowing them to take corrective action to prevent issues from arising. Optimization is an ongoing process in IA, and it requires businesses to continuously monitor and analyze their automated processes to identify areas for improvement. By doing so, businesses can improve efficiency, reduce costs, and improve the quality of products or services produced, leading to improved customer satisfaction and profitability. Here are the key steps involved in optimizing processes using IA:
Identify the processes to be optimized: The first step is to identify the processes that are causing bottlenecks, delays, or inefficiencies. This can be done by analyzing data, conducting surveys, or observing the workflow. Once the processes have been identified, it is important to prioritize them based on their impact on the business.
Define the objectives: The next step is to define the objectives of the optimization process. This could be reducing costs, improving quality, increasing productivity, or enhancing customer satisfaction. The objectives should be specific, measurable, and achievable.
Collect data: Data is essential for IA to work effectively. It is important to collect relevant data related to the processes being optimized, such as CT, throughput, error rates, and customer feedback. The data can be collected from various sources, such as sensors, databases, or manual inputs.
Analyze the data: The data collected needs to be analyzed using advanced analytics techniques, such as ML or statistical analysis. This will help identify patterns, trends, and correlations that can provide insights into the root causes of the problems. Based on these insights, it is possible to develop strategies to optimize the processes.
Implement intelligent automation: The next step is to implement IA to automate the processes. This can be done using a combination of technologies, such as RPA, NLP, and ML. The automation can be used to eliminate manual tasks, reduce errors, and speed up the process.
Monitor and refine: Once the optimization process has been implemented, it is important to monitor the performance of the processes and refine the strategies if needed. This can be done using key performance indicators (KPIs) such as CT, cost per unit, and customer satisfaction scores. The data collected can be used to fine‐tune the automation algorithms and make continuous improvements.
Overall, the combination of optimization and IA can provide significant benefits to organizations, such as cost savings, increased productivity, and improved customer satisfaction. By following these steps, businesses can create a data‐driven approach to process optimization that leverages the power of advanced technologies.
1.3.3 Analytics and Insight
Analytics and insights are critical components of IA. They refer to the process of collecting and analyzing data generated by automated processes to gain insights into how the processes are performing and identify areas for improvement. Analytics involves the use of advanced technologies such as ML and AI to analyze large volumes of data and identify patterns and trends. For example, businesses can use analytics to identify bottlenecks in their processes, areas where there is wastage or areas where automation can be applied to improve efficiency. Insights involve using the data generated by analytics to inform business decisions. For example, businesses can use insights gained from analytics to identify opportunities for process improvement, inform strategic decision‐making, or identify areas where additional automation can be applied. Analytics and insights are essential in IA because they help businesses identify areas for improvement, optimize processes, and make data‐driven decisions. By continuously analyzing data generated by automated processes and using insights to inform decision‐making, businesses can improve efficiency, reduce costs, and improve the quality of products or services produced. IA is the use of AI and ML technologies to automate processes, tasks, and workflows. When applied to analytics, IA can significantly enhance the speed, accuracy, and efficiency of data processing and analysis.
By leveraging IA, businesses can quickly and easily extract insights from large amounts of data, identify trends and patterns that might be difficult to spot manually, and make data‐driven decisions based on accurate and reliable information. This can help companies improve their performance, optimize their processes, reduce costs, and stay ahead of the competition.
In conclusion, AI and IA are two powerful technologies that are transforming the way businesses operate. By combining these technologies, businesses can automate processes, analyze data, and make more informed decisions faster, which can lead to improved efficiency and profitability.
1.4 The Fourth Industrial Revolution and Artificial Intelligence
AI has played a crucial role in every industrial revolution from the first to the fourth. In the First Industrial Revolution, machines were primarily powered by steam, water, and coal. However, the Second Industrial Revolution brought about the rise of electricity and the development of the telegraph and telephone. The use of computers and automation in manufacturing was first introduced during the Third Industrial Revolution, which led to the development of AI. Then, AI technology was further improved in the Fourth Industrial Revolution (4IR), which brought about the Internet of Things (IoT), Big Data, and cloud computing. The use of AI in these revolutions has brought about increased efficiency, productivity, and automation of various industries. In the 4th Industrial Revolution, AI is being used to optimize manufacturing processes, improve supply chain management, and revolutionize healthcare by developing more accurate diagnoses and personalized treatments. AI has become a crucial tool in various industries, and it is expected to continue playing an essential role in future revolutions.
The 4IR is a term used to describe the current era of rapid technological advancement that is transforming the way of living, working, and communicating. It builds upon the previous industrial revolutions, which were characterized by the mechanization of production (1IR), the introduction of mass production and assembly lines (2IR), and the automation of production through the use of computers and robotics (3IR). The 4IR is characterized by the integration of physical, digital, and biological systems and the use of technologies such as AI, the IoT, and robotics to drive innovation and productivity.
AI is one of the key technologies driving the 4IR. It refers to the ability of machines to perform tasks that typically require human intelligence, such as learning, reasoning, problem‐solving, and perception. AI systems are designed to simulate human cognitive abilities and can be used to automate a wide range of tasks across different industries. One of the most significant impacts of AI is its ability to analyze vast amounts of data quickly and accurately. This makes it particularly useful for applications, such as predictive analytics, where it can be used to identify patterns and make predictions based on historical data. AI is also increasingly being used for NLP, which enables computers to understand and process human language. Another key application of AI in the 4IR is in the development of autonomous systems, such as self‐driving cars and drones. These systems use a combination of sensors, algorithms, and ML to navigate their environment and make decisions in real time.
AI is also being used to improve healthcare, with applications such as personalized medicine, medical imaging, and drug discovery. In the financial sector, AI is being used for fraud detection and risk assessment, while in manufacturing, it is being used to optimize production processes and improve product quality. However, the increasing use of AI also raises concerns about issues such as job displacement, bias in decision‐making, and data privacy. As such, there is a growing need for ethical frameworks and regulations to ensure that AI is used in a responsible and transparent manner. The Fourth Industrial Revolution will be characterized by the widespread use of AI and Big Data. AI can be categorized into three stages: narrow artificial intelligence (ANI), artificial general intelligence (AGI), and super artificial intelligence (ASI). The ultimate goal is to achieve intelligence or even wisdom.
1.4.1 Artificial Narrow Intelligence
Artificial Narrow Intelligence (ANI) is a type of AI that is designed to perform specific tasks within a limited range of functions. ANI systems are built using ML algorithms and statistical models, and they are trained on large amounts of data to perform specific tasks. ANI systems use mathematical equations and algorithms to process data and make decisions based on that data. These mathematical models are often complex and can involve several different types of algorithms and techniques. One example of an ANI system is a computer vision system that is designed to recognize objects in images. This type of system uses deep learning algorithms, such as convolutional neural networks (CNNs), to analyze images and identify patterns that are associated with specific objects.
The mathematical equations used in ANI systems vary depending on the specific application, but they generally involve techniques such as linear algebra, probability theory, and optimization. Here are some examples of mathematical equations that are commonly used in ANI:
Linear regression is a statistical technique that is used to model the relationship between two variables. In ANI, linear regression can be used to predict the value of an output variable based on the values of one or more input variables. The equation for linear regression is:
where y is the output variable, x1, x2, …, xn are the input variables, and b0, b1, b2, …, bn are the coefficients of the regression equation. Bayes' Theorem is a mathematical equation that is used to calculate the probability of an event based on prior knowledge of related events. In ANI, Bayes' Theorem can be used to make predictions based on data that has been collected previously. The equation for Bayes' Theorem is:
where P(A∣B) is the probability of event A given that event B has occurred, P(B∣A) is the probability of event B given that event A has occurred, P(A) is the prior probability of event A, and P(B) is the prior probability of event B.
Gradient descent is an optimization algorithm that is used to find the minimum of a function. In ANI, gradient descent can be used to adjust the parameters of an ML model to minimize the error between predicted and actual values. The equation for the gradient descent algorithm is:
where b is the next position of the migrating point, while a represents its current position. The minus sign refers to the minimization part of the gradient descent algorithm. The γ is a waiting factor and the gradient term ∇f(a) is simply the direction of the steepest descent.
In summary, ANI is a type of AI that uses mathematical equations and algorithms to process data and perform specific tasks. The specific mathematical models and equations used in ANI systems depend on the specific application, but they generally involve techniques such as linear algebra, probability theory, and optimization.
1.4.2 Artificial General Intelligence
Artificial General Intelligence (AGI) is a theoretical type of AI that can perform tasks that typically require human‐level intelligence.
The idea behind AGI is that an AI system would be able to learn and adapt to new situations, just as a human would. This would require the AI to be able to reason, make decisions, and solve problems in a variety of contexts. One of the key challenges in creating AGI is developing algorithms that can handle the complexity of human‐like thinking.
One approach to developing AGI is through deep learning, which uses neural networks to simulate the function of the human brain. Neural networks consist of interconnected nodes that perform computations based on input data. These computations are typically represented as mathematical equations. The connections between nodes are weighted based on the strength of the correlation between the input and output data. The weights are adjusted during the learning process, allowing the neural network to improve its predictions over time.
Another approach to developing AGI is reinforcement learning, which involves training an AI system to make decisions based on feedback from its environment. Reinforcement learning uses a reward‐based system to encourage the AI to take actions that lead to positive outcomes. The goal is to develop an AI system that can learn from its mistakes and make better decisions over time. There are many mathematical equations used in the development of AGI, including:
Gradient descent: This equation is used in deep learning to adjust the weights of the connections between nodes in a neural network. It involves calculating the gradient of the error function with respect to the weights and then adjusting the weights in the direction of the gradient.
Bellman equation: This equation is used in reinforcement learning to calculate the expected value of a decision based on the potential future rewards. It takes into account the immediate reward as well as the expected future reward based on the decision.
Bayes' Theorem: This equation is used in probabilistic reasoning to update the probability of a hypothesis based on new evidence. It is often used in ML algorithms that involve uncertainty.
Overall, AGI is a complex and challenging field of study that involves many different mathematical approaches. While still there has not been a fully functioning AGI system, ongoing research is pushing the boundaries of what is possible with AI and bringing us closer to creating machines that can reason, learn, and adapt like humans.
1.4.3 Artificial Super Intelligence
Artificial Super Intelligence (ASI) refers to the hypothetical future state of AI where machines will surpass human intelligence and become capable of performing tasks that are currently considered impossible for machines. While there is no universally accepted definition of super AI, one way to conceptualize it is through the concept of an AGI. An AGI is an AI system that is capable of understanding or learning any intellectual task that a human being can, including those that are currently beyond the capabilities of any machine. ASI could be seen as an even more advanced version of AGI, capable of not just performing any intellectual task but surpassing human intelligence in all areas. The development of ASI would likely involve significant advances in fields such as ML, artificial neural networks (ANNs), and reinforcement learning, as well as the development of entirely new approaches to AI. Some possible mathematical equations and concepts that could be involved in the development of super AI include:
Neural networks: ANNs are a mathematical model that is inspired by the structure and function of biological neural networks in the brain. ANNs consist of layers of interconnected nodes (also known as neurons), which are capable of processing information and making predictions. ANNs can be trained using algorithms such as backpropagation to adjust the weights and biases of the nodes to improve their performance. Super AI could potentially involve the development of more advanced and complex neural networks, with a greater number of layers and nodes.
Reinforcement learning: A type of ML that involves an agent learning through trial and error in an environment where it receives feedback in the form of rewards or punishments. The agent's goal is to learn a policy (i.e. a set of actions) that maximizes its long‐term reward. Super AI could potentially involve the development of more advanced reinforcement learning algorithms, such as deep reinforcement learning, which uses deep neural networks to represent the agent's policy.
Bayesian networks: A probabilistic graphical model that represents a set of random variables and their conditional dependencies using a directed acyclic graph. Bayesian networks can be used for reasoning, prediction, and decision‐making under uncertainty. Super AI could potentially involve the development of more advanced Bayesian networks, capable of handling larger and more complex data sets.
Information theory: Information theory is a mathematical framework for quantifying and analyzing the amount of information in a message or data set. Information theory can be used for tasks such as data compression, error correction, and signal processing. ASI could potentially involve the development of more advanced information‐theoretic approaches to AI, such as the use of entropy‐based measures to optimize learning algorithms.
