AI and SWARM

Evolutionary Approach to Emergent Intelligence

Hitoshi Iba

• Publisher: CRC Press
• Year: 2019
• ISBN:
  9780429648434
• Edition: 1
• Topic: Computer Science
• Subtopic: Computer Science General
• Index: Computer Science

This book provides theoretical and practical knowledge on AI and swarm intelligence. It provides a methodology for EA (evolutionary algorithm)-based approach for complex adaptive systems with the integration of several meta-heuristics, e.g., ACO (Ant Colony Optimization), ABC (Artificial Bee Colony), and PSO (Particle Swarm Optimization), etc. These developments contribute towards better problem-solving methodologies in AI. The book also covers emerging uses of swarm intelligence in applications such as complex adaptive systems, reaction-diffusion computing, and diffusion-limited aggregation, etc.

Another emphasis is its real-world applications. We give empirical examples from real-world problems and show that the proposed approaches are successful when addressing tasks from such areas as swarm robotics, silicon traffics, image understanding, Vornoi diagrams, queuing theory, and slime intelligence, etc.

Each chapter begins with the background of the problem followed by the current state-of-the-art techniques of the field, and ends with a detailed discussion. In addition, the simulators, based on optimizers such as PSO and ABC complex adaptive system simulation, are described in detail. These simulators, as well as some source codes, are available online on the author's website for the benefit of readers interested in getting some hands-on experience of the subject.

The concepts presented in this book aim to promote and facilitate the effective research in swarm intelligence approaches in both theory and practice. This book would also be of value to other readers because it covers interdisciplinary research topics that encompass problem-solving tasks in AI, complex adaptive systems, and meta-heuristics.

Chapter 1

Introduction

A sonnet written by a machine would be better appreciated by another machine.

—Alan Turing

1.1 What is AI? – Strong AI vs Weak AI

AI (artificial intelligence) refers to the implementation of intelligence on a computer and is divided into two hypotheses.

Strong AI The viewpoint that true intelligence can be implemented on a computer, also known as general AI.

Weak AI The viewpoint that computers can merely give the impression of intelligence, also known as narrow AI.

In the same manner, artificial life (AL), which is mentioned later, can be defined in terms of strong AL and weak AL.

“Weak AI” is an application of AI technologies to enable a high-functioning system that replicates human intelligence for a specific purpose. In fact, there are breakthroughs in weak AI.

In this book we consider simulation in the sense of “strong AI”. More precisely, the rationale behind this approach is that “the appropriately programed computer really is a mind, in the sense that computers, given the right programs, can be literally said to understand and have other cognitive states.”

Figure 1.1: Turing test.

Realizing such a computer is nontrivial, since the definition of “intelligence” is difficult in the first place. Therefore, if a claim is made that AI (in the strong sense) has been created, what would be the most appropriate way to evaluate it?

To this end, Alan Turing¹ proposed a test of a machine’s capacity to exhibit intelligent behavior, now called the “Turing test”, which, despite being powerful, is the subject of numerous disputes (Fig. 1.1). The question of whether machines can think was considered in great depth by Turing, and his final opinion was affirmative. The Turing test can be translated into modern terms in the form of a game involving the exchanging of messages via a discussion board:

■ One day, two new users, A and B, join the discussion board.

■ When a message is sent to A and B, they both return apt responses.

■ Of A and B, one is human and the other is a computer.

■ However, it is impossible to determine which is which, regardless of the questions asked.

If a program passes this test (in other words, the computer cannot be identified), the program can be said to simulate intelligence (as long as the questions are valid). A similar contest, named the “The Loebner prize” after its patron, the American philanthropist Hugh Loebner, is held online². Although an award of 100,000 US dollars and a solid gold medal has been offered since 1990, so far, not a single machine participating in the contest has satisfied the criteria for winning.

Nevertheless, a number of problems with the Turing test have been pointed out, and various critical remarks have been issued about potential implementation of AI. A notable example is the challenge to the very “definition of intelligence” by John Searle³, who questioned the foundations of the Turing test by creating a counter thought experiment. Searle’s experiment, known as the “Chinese room,” can be summarized as follows.

Figure 1.2: A Chinese room.

A person is confined to a room with a large amount of written material on the Chinese language (Fig. 1.2). Looking inside the room is impossible, and there are only input and output boxes for submitting sentences and obtaining responses. Having no understanding of Chinese, the person cannot distinguish between different Chinese characters (for this purpose, we assume that the person is British and not Japanese). Furthermore, the person is equipped with a comprehensive manual (written in English) containing rules for connecting sets of Chinese characters. Let us consider that a person who understands Chinese is leading a conversation by inserting questions written in Chinese into the input box and retrieving answers from the output box. Searle provides the following argument.

Suppose that the person quickly becomes truly proficient at manipulating Chinese characters in accordance to the instructions, and that the person outside the room also becomes proficient at providing instructions. Then, the answers prepared by the person inside the room would become indistinguishable from answers provided by a Chinese person. Nobody would consider, simply by looking at the provided answers, that the person inside the room does not understand Chinese. However, in contrast to English, in the case of Chinese, the person in the room prepares the answers by formally manipulating characters, without any understanding whatsoever.

It cannot be said that true understanding is achieved simply by looking at the typeface while performing manipulations in accordance with a formal set of rules. However, as demonstrated by the “Chinese room” thought experiment, under specific conditions, human-like behavior can be fabricated by both humans and machines if appropriate formal rules are provided. Searle, therefore, argues that strong AI is impossible to realize.

Various counterarguments have been considered in response to Searle, and questions that would probably occur to most people include

■ Can conversion rules be written for all possible inputs?

■ Can such an immense database actually be searched?

However, these counterarguments are devoid of meaning. The former rejects the realization of AI in the first place, and the latter cannot be refuted in light of the possibility that ultra-high-speed parallel computing or quantum computing may exist in the future. Thus, neither one can serve as the basis of an argument.

O...

Table of contents

• Cover
• Title Page
• Copyright Page
• Preface
• Acknowledgments
• Table of Contents
• Abbreviations
• 1. Introduction
• 2. AI, Alife and Emergent Computation
• 3. Meta-heuristics
• 4. Emergent Properties and Swarm Intelligence
• 5. Complex Adaptive Systems
• 6. Emergence of Intelligence
• 7. Conclusion
• References
• Index
• Color Section