Swarm Intelligence and Deep Evolution

Evolutionary Approach to Artificial Intelligence

Hitoshi Iba

• Publisher: CRC Press
• Year: 2022
• ISBN:
  9781000579932
• Edition: 1
• Topic: Computer Science
• Subtopic: Software Development
• Index: Computer Science

The book provides theoretical and practical knowledge about swarm intelligence and evolutionary computation. It describes the emerging trends in deep learning that involve the integration of swarm intelligence and evolutionary computation with deep learning, i.e., deep neuroevolution and deep swarms. The study reviews the research on network structures and hyperparameters in deep learning, and attracting attention as a new trend in AI. A part of the coverage of the book is based on the results of practical examples as well as various real-world applications. The future of AI, based on the ideas of swarm intelligence and evolution is also covered.

The book is an introductory work for researchers. Approaches to the realization of AI and the emergence of intelligence are explained, with emphasis on evolution and learning. It is designed for beginners who do not have any knowledge of algorithms or biology, and explains the basics of neural networks and deep learning in an easy-to-understand manner. As a practical exercise in neuroevolution, the book shows how to learn to drive a racing car and a helicopter using MindRender. MindRender is an AI educational software that allows the readers to create and play with VR programs, and provides a variety of examples so that the readers will be able to create and understand AI.

Chapter 1 AI: Past and Present

I have always had some disquiet about the term “artificial intelligence” and only rarely identified myself as working primarily in that area. However, I remember the first time I met Edsger Dijkstra. He was noted not only for his pioneering contributions to computer science, but also for having strong opinions and a stinging wit. He asked me what I was working on. Perhaps just to provoke a memorable exchange I said, “Al.” To that he immediately responded, “Why don’t you work on I?” [143].

1.1. AI and its History

The word “artificial intelligence” (hereafter, AI) originated from the 1956 Dartmouth workshop. Many leading future researchers of AI research, including Marvin Minsky¹, John McCarthy², Herbert Simon³, and⁴, participated in this conference. Early AI was actively studied in relation to game and theorem proving. This was followed by studies connected to the real world, such as those on machine inference, problem solving, image understanding, natural language learning, and expert systems. Moreover, academic fields connected to AI, such as cognitive science, which is the field where the mechanism of the human mind centered on “intelligence” is studied, and knowledge engineering, which is the field that aims to endow computers with human knowledge for engineering purposes, have also been established.

1 Marvin Minsky (1927–2016): An American computer scientist and cognitive scientist. He is known as the “father of artificial intelligence.” He is famous for his work on frame theory and the society of mind. His book [100] showed that a simple perceptron cannot identify linearly inseparable patterns, ending the first neural network boom in the 1960s and causing the “AI winter” of the 1970s. 2 John McCarthy (1927–2011): An American computer scientist and cognitive scientist. The term “Artificial Intelligence” was invented by him for the Dartmouth workshop. He is the developer of the LISP language. 3 Herbert Alexander Simon (1916–2001): American cognitive psychologist and information scientist. He was awarded the Nobel Prize in Economics in 1978 for his lifelong research on managerial behavior and decision-making in large organizations. 4 Allen Newell (1927–1992): American researcher in computer science and cognitive psychology. He is well known for Logic Theory Machine (1956) and General Problem Solver (1957), early AI programs he developed with Herbert Simon.

Broadly speaking, early AI research tended toward three main trends: the first trend was based on the theory of logical thinking from the late 1940s, which led to research on predicate logical languages, such as the Prolog; the second trend was the neural network approach (connectionism) from the 1960s; while the third trend was heuristic programing from the 1950s.

As research on AI advanced, early optimism started to wane, and many criticisms and challenges emerged. For example, the above mentioned three approaches received criticisms, such as the inability of logical methods to process the flexible thinking of humans, the difficulty of connectionism to express high-level knowledge, and the ad hoc nature of heuristics.

AI is an interdisciplinary field related to many other fields, such as psychology, engineering, information science, mathematics, philosophy, and brain science, through the keyword “intelligence,” and this relationship itself is AI. The ancestors of AI in psychology are Jean Piaget⁵ and Sigmund Freud⁶. Piaget conducted the pioneering study on the knowledge acquisition of children through an empirical development study. His book “The Origins of Intelligence in Children” emphasizes the role of structuring in cognition, and its influence on AI is significant. Piaget advanced the theory of staged mental development, which proposed that the development of thought is through a gradual development of structure. It was claimed that children try out logical combinations through various plays and acquire knowledge through the same to develop. This theory is called “constructivism.” Meanwhile, Freud was the founder of psychoanalysis and advanced a unique theory on ego and the unconscious.

5 Jean Piaget (1896–1980): Swiss psychologist. He proposed “genetic epistemology.” His work in developmental psychology has influenced the theory and practice of the new education. 6 Sigmund Freud (1856–1939): Austrian psychiatrist. After working as a neuropathologist, he became a psychiatrist and conducted research on neurosis, free association, and the unconscious. He is known as the founder of psychoanalysis. He proposed the theory of psychosexual development, libido theory, and infantile sexual desire.

However, there are claims that the theories of Freud and Piaget have already been disproved [111]. For example, paleontologist S.J. Gould⁷ views that the theory of Freud (i.e., interpretation of sexual developmental stage and neurosis patients) reflects the recapitulation theory⁸ and Lamarckism⁹ in biology. He wrote that the disproving of these two theories consequently ended the validity of the theory of Freud [51]. From the viewpoint of an evolutionary psychologist, Steven Pinker¹⁰ also claimed that we must understand the purpose of the design of our mind in the environment we evolved in. He explained the harm of the theory of Piaget in education as follows [111]:

7

Stephen Jay Gould (1941–2002): American paleontologist and evolutionary biologist. He published an essay in the American scientific journal “Natural History” every month. His many books collecting these essays became bestsellers. While both being researchers of evolutionary theory, he was an opponent of Richard Dawkins (see 41 page). 8 Theory by German biologist Ernst Haeckel (1834–1919), which postulates that phylogeny repeats ontogeny. An animal repeats the path of evolution when growing from the embryo. For instance, the theory regards the gill slit seen in the human embryo to be the residue of the time when we were fish and the tail of the embryo that is later absorbed to be the residue of our amphibian ancestors. This theory has been disproven, but it persists as an urban legend. 9 The view that acquired characters are genetically inherited. This is also disproved by the knowledge obtained through genetics. See Section 2.2.6 and 62 page. 10

Steven Arthur Pinker (1954–): American cognitive psychologist. He is a Harvard University Professor in psychology and the author of a large number of scientific publications including “How the Mind Works,” “The Blank State: The Modern Denial of Human Nature,” and “The Language Instinct: How the Mind Creates Language.”

Ever since the Swiss psychologist Jean Piagct likened children to little scientists, psychologists have compared the person in the street, young and old, to the person in the lab. (…) Natural selection, however, did not shape us to earn good grades in science class or to publish in refereed journals. It shaped us to master the local environment, and that led to discrepancies between how we naturally think and what is demanded in the academy.

Particularly, with the exception of the instinctual counting of small numbers and simple calculations, children will not be able to learn how to handle mathematical concepts spontaneously. Similarly, when teaching children how to read, the constructivist method postulating that “reading is a human instinct that develops naturally” (which has not been proven by evolutionary psychology) and contending that it is enough to place them in an interpersonal environment rich with texts is not correct. Moreover, in the field of evolutionally cognitive science, it is regarded that cognitive science not based on the human genome or biological function is impossible. The mechanism of human cognition is ...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Preface
• Table of Contents
• 1. AI: Past and Present
• 2. Evolutionary Theories for AI
• 3. Evolutionary Computation
• 4. Swarm Intelligence
• 5. Deep Learning and Evolution
• 6. Deep Swarms and Evolution
• 7. Emergence of Intelligence
• A. Software Packages
• References
• Indices