Advanced Artificial Intelligence

Zhongzhi Shi

• Editorial: WSPC
• Año: 2019
• ISBN:
  9789811200892
• Edición: 2
• Categoría: Ciencia de la computación
• Categoría: Inteligencia artificial (IA) y semántica

The joint breakthrough of big data, cloud computing and deep learning has made artificial intelligence (AI) the new focus in the international arena. AI is a branch of computer science, developing intelligent machine with imitating, extending and augmenting human intelligence through artificial means and techniques to realize intelligent behaviour.

This comprehensive compendium, consisting of 15 chapters, captures the updated achievements of AI. It is completely revised to reflect the current researches in the field, through numerous techniques and strategies to address the impending challenges facing computer scientists today.

The unique volume is useful for senior or graduate students in the information field and related tertiary specialities. It is also a suitable reference text for professionals, researchers, and academics in AI, machine learning, electrical & electronic engineering and biocomputing.

Contents:

• Introduction
• Logic Foundation
• Constraint Reasoning
• Bayesian Network
• Probabilistic Graphic Models
• Case-Based Reasoning
• Inductive Learning
• Statistical Learning
• Deep Learning
• Reinforcement Learning
• Unsupervised Learning
• Association Rules
• Evolutionary Computation
• Multi-Agent Systems
• Internet Intelligence

Readership: Researchers, academics, professionals and senior and graduate students in artificial intelligence, machine learning, electrical and electronic engineering.Constraint Reasoning;Bayesian Network;Probabilistic Graphic Models;Case-Based Reasoning;Inductive Learning;Statistical Learning;Deep Learning;Reinforcement Learning00

Chapter 1

Introduction

Artificial intelligence emphasizes the creation of intelligent machines that work and react like humans. It is usually defined as the science and engineering of emulating, extending, and augmenting human intelligence through artificial means and techniques to make intelligent machines.

1.1Brief History of AI

Artificial intelligence (AI) is usually defined as the science and engineering of imitating, extending, and augmenting human intelligence through artificial means and techniques to make intelligent machines. In 2005, John McCarthy pointed out that the long-term goal of AI is human-level AI (McCarthy, 2005).

In the history of human development, it is a never-ending pursuit to free people from both manual and mental labor with machines. The industrial revolutions enabled machines to perform heavy manual labor instead of people and thus lead to a considerable economic and social progress. To make machines help relieve mental labor, a long cherished aspiration is to create and make use of intelligent machines like human beings.

In ancient China, many mechanical devices and tools were invented to help accomplish mental tasks. The abacus was the most widely used classical calculator. The Water-powered Armillary Sphere and Celestial Globe Tower were used for astronomical observation and stellar analysis. The Houfeng Seismograph was an ancient seismometer used to detect and record tremors and earthquakes. The traditional Chinese theory of Yin and Yang reveals the philosophy of opposition, interrelation, and transformation, having an important impact on modern logic.

Aristotle (384–322 BC) proposed the first formal deductive reasoning system, syllogistic logic, in the Organon. Francis Bacon (1561–1626) established the inductive method in the Novum Organum (or “New Organon”). Gottfried Leibniz (1646–1716) constructed the first mechanical calculator capable of multiplication and division. He also enunciated the concepts of “characteristica universalis” and “calculus ratiocinator” to treat the operations of formal logic in a symbolic or algebraic way, which can be viewed as the sprout of the “thinking machine”.

Since the 19th century, advancement in sciences and technologies such as Mathematical Logic, Automata Theory, Cybernetics, Information Theory, Computer Science, and Psychology laid the ideological, theoretical, and material foundation for the development of AI research. In the book An Investigation of the Laws of Thought, George Boole (1815–1864) developed the Boolean algebra, a form of symbolic logic to represent some basic rules for reasoning in thinking activities. Kurt Gödel (1906–1978) proved the incompleteness theorems. Alan Turing (1912–1954) introduced the Turing Machine — a model of the ideal intelligent computer — and initiated the automata theory. In 1943, Warren McCulloch (1899–1969) and Walter Pitts (1923–1969) developed the MP neuron, a pioneer work of Artificial Neural Networks research. In 1946, John Mauchly (1907–1980) and John Eckert (1919–1995) invented the Electronic Numerical Integrator And Computer (ENIAC), the first electronic computer. In 1948, Norbert Wiener (1894–1964) published a popular book of Cybernetics, and Claude Shannon (1916–2001) proposed the Information Theory.

In the real world, quite a number of problems are complex ones, most of the time without any algorithm to adopt; or even if there are calculation methods, they are still NP problems. Researchers might introduce heuristic knowledge to solve such problems, to simplify complex problems, and to find solutions in the vast search space. Usually, the introduction of domain-specific empirical knowledge will produce satisfactory solutions, though they might not be the mathematically optimal solutions. This kind of problem solving with its own remarkable characteristics led to the birth of AI. In 1956, the term “Artificial Intelligence” was coined, and the Dartmouth Summer Research Project on artificial intelligence, proposed by John McCarthy, Marvin Minsky, etc., was carried on at Dartmouth College with several American scientists of psychology, mathematics, computer science, and information theory. This well-known Dartmouth conference marked the beginning of the real sense of AI as a research field. Through dozens of years of research and development, great progress has been made in the discipline of AI. Many artificial intelligence expert systems have been developed and applied successfully. In domains such as Natural Language Processing, Machine Translation, Pattern Recognition, Robotics, and Image Processing, a lot of achievements have been made, and the applications span various areas of development.

In the 1950s, AI research mainly focused on game playing. In 1956, Arthur Samuel wrote the first heuristic game-playing program with learning ability. In the same year, Alan Newell, Herbert Simon, etc., invented a heuristic program called the Logic Theorist, which proved correct 38 of the first 52 theorems from the “Principia Mathematica”. Their work heralded the beginning of research on cognitive psychology with computers. Noam Chomsky proposed the Syntactics, the pioneer work of Formal Language research. In 1958, John McCarthy invented the Lisp language, an important tool for AI research which can process not only numerical values but also symbols.

In the early 1960s, AI research mainly focused on search algorithms and general problem solving (GPS). Allen Newell et al., published the General Problem Solver, a more powerful and universal heuristic program than other programs at that time. In 1961, Marvin Minsky published the seminal paper Steps Toward Artificial Intelligence which established a fairly unified terminology for AI research and established the subject as a well-defined scientific enterprise. In 1965, Edward Feigenbaum et al., began work on the DENDRAL chemical analysis expert system, a milestone for AI applications, and initiated the shift from computer algorithms to knowledge representation as the focus of AI research. In 1965, Alan Robinson proposed the Resolution Principle. In 1968, Ross Quillian introduced the Semantic Network for knowledge representation. In 1969, International Joint Conferences on Artificial Intelligence (IJCAI) was founded, and since then, the IJCAI has been held biannually in odd-numbered years. Artificial Intelligence, an international journal edited by IJCAI, commenced publication in 1970.

In the early 1970s, AI research mainly focused on Natural Language Understanding and Knowledge Representation. In 1972, Terry Winograd published details of the SHRDLU program for understanding natural language. Alain Colmerauer developed Prolog language for AI programming at the University of Marseilles in France. In 1973, Roger Schank proposed the Conceptual Dependency Theory for Natural Language Understanding. In 1974, Marvin Minsky published the frame system theory, an important theory of Knowledge Representation. In 1977, Edward Feigenbaum published the well-known paper The Art of Artificial Intelligence: Themes and Case Studies in Knowledge Engineering in the 5th IJCAI. He stated that Knowledge Engineering is the art of bringing the principles and tools of AI research to bear on difficult applications problems requiring expert knowledge for their solution. The technical issues of acquiring this knowledge, representing it, and using it appropriately to construct and explain lines of reasoning are important problems in the design of knowledge-based systems.

In the 1980s, AI research developed prosperously. Expert systems were more and more widely used, development tools for expert systems appeared, and industrial AI thrived. Especially in 1982, Japan’s Ministry of International Trade and Industry initiated the Fifth Generation Computer Systems project, which dramatically promoted the development of AI. Many countries also made similar plans for research in AI and intelligent computers. China also started the research of intelligent computer systems as an 863 National High-Tech Program.

During the past more than 60 years, great progress has been made in the field of AI research. Theories of Heuristic Searching Strategies, Non-monotonic Reasoning, Machine Learning, etc., have been proposed. Applications of AI, especially Expert Systems, Intelligent Decision-Making, Intelligent Robots, Natural Language Understandings, etc., also promoted the research of AI. Presently, Knowledge Engineering based on knowledge and information processing is a remarkable characteristic of AI.

However, just as with the development of any other discipline, there are also obstacles in the history of AI research. Even from the beginning, AI researchers had been criticized for their being too optimistic. In the early years of AI research, Herbert Simon and Allen Newell, two of the AI pioneers, optimistically predicted the following:

•Within 10 years, a digital computer will be the world’s chess champion, unless the rules bar it from competition.

•Within 10 years, a digital computer will discover and prove an important new mathematical theorem.

•Within 10 years, a digital computer will write music that will be accepted by critics as possessing considerable aesthetic value.

•Within 10 years, most theories in psychology will take the form of computer programs or qualitative statements about the characteristics of computer programs.

These expectations haven’t been completely realized even till date. A 3-year-old little child can easily figure out a tree in a picture, while a most powerful super computers have only reached a middle level as children in tree recognition. It is also very difficult to automatically understand even stories written for little children.

Some essential theories of AI still need improvements. No breakthrough progresses have been made for some key technologies such as Machine Learning, Non-monotonic Reasoning, Commonsense Knowledge Representation, and Uncertain Reasoning. It is also very difficult for global judgment, fuzzy information processing, multi-granular visual information processing, etc.

Conclusively, AI research is still in the first stage of Intelligence Science, an indispensable cross discipline which is dedicated to joint research on basic theories and technologies of intelligence by Brain Science, Cognitive Science, Artificial Intelligence, and others. Brain Science explores the essence of brain and investigates the principles and models of natural intelligence at the molecular, cellular, and behavioral levels. Cognitive Science studies human mental activities, such as perception, learning, memory, thinking, and consciousness. AI research aims at imitating, extending, and augmenting human intelligence through artificial means and techniques, and finally achieving machine intelligence. These three disciplines work together to explore new concepts, new theories, and new methodologies for Intelligence Science, opening up prospects for a successful and brilliant future in the 21st century (Shi, 2006a).

1.2Cognitive Issues of AI

Cognition is generally referred to as the process of knowing or understanding relative to affection, motivation, or volition. Definitions of cognition can be briefly summarized into five main categories according to the American psychologist Houston:

(1)Cognition is the process of information processing.

(2)Cognition involves symbol processing in psychology.

(3)Cognition deals with problem solving.

(4)Cognition studies mind and intelligence.

(5)Cognition consists of a series of activities, such as perception, memory, thinking, judgment, reasoning, problem solving, learning, imagination, concept forming, language using, etc.

Cognitive psychologist David H. Dodd, etc., held that cognition involves the three aspects of adaptation, structure, and process, i.e., cognition is the process of information processing in certain mental structures for certain objectives.

Cognitive Science is the science of human perceptions and mental information processing, spanning from perceptual input to complex problem solving, including intellectual activities from individuals to the whole society, and investigating characteristics of both human intelligence and machine intelligence (Shi, 1990a). As an important theoretical foundation for AI, Cognitive Science is an interdisciplinary field developed from Modern Psychology, Information Science, Neuroscience, Mathematics, Scientific Linguistics, Anthropology, Natural Philosophy, etc.

The maturing and development of Cognitive Science marked a new stage of research on human-centered cognitive and intelligent activities. Research on Cognitive Science will enable self-understanding and self-control and lift human knowledge and intelligence to an unprecedented level. Moreover, it will lay theoretical foundations for the intelligence revolution, knowledge revolution, and information revolution, as well as provide new concepts, new ideas, and new methodologies for the development of intelligent computer systems.

Promoted by works of Allen Newell and Herbert Simon, research related to cognitive science originated in the late 1950s (Newell and Simon, 1972, 1976; Simon, 1986). Cognitive scientists proposed better models for mind and thinking than the simplified model about humans developed by behaviorism scientists. Cognitive Science research aims at illustrating and explaining how information is processed during cognitive activities. It involves a variety of problems including perception, language, learning, memory, thinking, problem solving, creativity, attention, as well as the impact of environment and social culture on cognition.

In 1991, the representative journal Artificial Intelligence published a special issue on the foundation of AI in its 47th volume, in which trends about AI research are discussed. In this special issue, David Kirsh discussed five foundational questions for AI research (Kirsh, 1991):

(1)Pre-eminence of knowledge and conceptualization: Intelligence that transcends insect-level intelligence requires declarative knowledge and some form of reasoning-like computation — call this cognition. Core AI is the study of the conceptualizations of the world presupposed and used by intelligent systems during cognition.

(2)Disembodiment: Cognition and the knowledge, it presupposes, can be studied largely in abstraction from the details of perception and motor control.

(3)Kinematics of cognition are language-like: It is possible to describe the trajectory of knowledge states or informational states created during cognition using a vocabullry very much like English or some regimented logic–mathematical version of English.

(4)Learning can be added later: The kinematics of cognition and the domain knowledge needed for c...