Paradigms of Artificial Intelligence

by Hoffmann, Achim Gunther

• ISBN13: 9789813083974
• ISBN10: 9813083972
• Format: Paperback
• Copyright: 1998-11-01
• Publisher: Springer Verlag

Summary

This book presents a new methodological analysis of the two competing research paradigms of artificial intelligence and cognitive science: the symbolic versus the connectionist paradigm. It argues that much of the discussion put forward for either paradigm misses the point: Most of the arguments in the debates on the two paradigms concentrate on the question whether the nature of intelligence or cognition is properly accommodated by one or the other paradigm. Opposed to that is the analysis in this book, which concentrates on the question which of the paradigms accommodates the "user" of a developed theory or technique. The user may be an engineer or a scientist - in any case, the "user" has to be able to grasp the theory and to competently use the methods which are developed. Consequently, besides the nature of intelligence and cognition, the "user" must be the focus of the debate on the proper paradigm. From the relationship between the mental capacities of the "user" and certain aspects of the nature of intelligence and cognition, the book derives new objectives for future research which will help to integrate aspects of both paradigms to obtain more powerful AI techniques and to promote a deeper understanding of cognition. . The book presents the fundamental ideas of both, the symbolic as well as the connectionist paradigm. Along with an introduction to the philosophical foundations, an exposition of some of the typical techniques of each paradigm is presented in the first two parts. This is followed by the mentioned analysis of the two paradigms in the third part. The book is intended for researchers, practitioners, advanced students, and interested observers of the developing fields of artificial intelligence and cognitive science. Providing accessible introductions to the basic ideas of both paradigms, it is also suitable as a textbook for a subject on the topic at an advanced level in computer science, philosophy, cognitive science, or psychology.

Table of Contents

Preface	vii
Introduction	1	(10)
Part I. The Symbolic Paradigm
Foundations of the Symbolic Paradigm	11	(14)
Introduction	11	(3)
Physical Symbol Systems	14	(3)
Abstract Levels of System Description	17	(5)
Traditional Levels of Computer System Description	17	(2)
The Knowledge Level	19	(3)
Summary	22	(3)
Knowledge Representation and Reasoning	25	(24)
Logic	25	(13)
Propositional Logic	25	(5)
Predicate Logic	30	(1)
A Formal Language of Predicate Logic	31	(1)
Interpretations	32	(3)
Resolution	35	(3)
Beyond Classical Logic	38	(4)
Default Theories	40	(1)
Further Types of Logic	41	(1)
Reasoning with Uncertainty	42	(5)
Complexity of Reasoning	47	(2)
Expert Systems	49	(10)
Rule-Based Expert Systems	50	(5)
Second-Generation Expert Systems	55	(1)
Practical Difficulties: The Knowledge Acquisition Bottleneck	56	(3)
Symbolic Learning	59	(38)
Introduction	59	(2)
Preliminaries for Learning Concepts from Examples	61	(3)
Representing Training Data	61	(1)
Learning Algorithms	62	(1)
Objects, Concepts, and Concept Classes	62	(1)
Consistent and Complete Concepts	63	(1)
Generalisation as Search	64	(4)
Learning of Classification Rules	68	(6)
Model-Based Learning Approaches: The AQ Family	68	(3)
Non-Boolean Attributes	71	(1)
Problems and Further Possibilities of the AQ Framework	72	(2)
Learning Decision Trees	74	(8)
Representing Functions in Decision Trees	74	(1)
The Learning Process	74	(6)
Representational Aspects in Decision Tree Learning	80	(1)
Overfitting and Tree Pruning	81	(1)
Inductive Logic Programming	82	(12)
A Categorisation of ILP Systems	83	(7)
Some ILP Systems	90	(3)
Discussion on Inductive Logic Programming	93	(1)
Discussion on Learning	94	(3)
Summary of the Symbolic Paradigm	97	(8)
Part II. The Connectionist Paradigm
Foundations of the Connectionist Paradigm	105	(8)
The Emergence of Connectionism	105	(2)
Biological Neural Networks	107	(6)
Connectionist Computing Models	113	(36)
Short History of Artificial Neural Networks	114	(1)
Perceptron: Learning Linear Threshold Functions	115	(6)
Representing Data for Learning in Connectionist Systems	116	(3)
Linear Decision Functions	119	(2)
Limitations of Linear Threshold Functions	121	(2)
Multi-Layer Perceptron	123	(9)
The Generalised Delta Rule	123	(2)
The Mathematical Formulation of the Generalised Delta Rule	125	(5)
Applications of Multi-Layer Perceptrons	130	(2)
Recurrent Networks	132	(2)
Hopfield Networks	134	(5)
The Hopfield Unit Output and Update Functions	135	(1)
Application Areas of Hopfield Networks	136	(3)
Unsupervised Learning in Neural Networks	139	(8)
Adaptive Resonance Theory	139	(5)
Self-Organising Feature Maps	144	(3)
Summary	147	(2)
Integrating Symbols into Connectionist Models	149	(8)
Structured Connectionism	149	(5)
Smolensky's Integrated Connectionist/Symbolic Architecture	154	(1)
Relating Symbolic Rules to Connectionist Networks	155	(2)
Rule-Extraction from Neural Networks	155	(1)
Providing Symbolic Structure to a Neural Network A Priori	156	(1)
Summary of the Connectionist Paradigm	157	(6)
Part III. Methodological Analysis of the Two Paradigms
Formal Foundations	163	(12)
The Notion of Algorithm	163	(7)
The Turing Machine	164	(1)
Production Rules	164	(2)
Cellular Automata	166	(4)
Algorithmic Information Theory	170	(1)
On the Algorithmic Information of Intelligence	171	(4)
Levels of Description	175	(24)
The Symbolic Level	178	(4)
Symbols in the Compositional Sense	179	(2)
Non-Compositional Symbol Systems	181	(1)
Problems with the Symbolic Level	182	(5)
The Phenomenological Critique at Physical Symbol Systems	182	(4)
Practical Experience in the Symbolic Paradigm	186	(1)
Conclusions from the Critiques	186	(1)
The Connectionist Alternative	187	(3)
The Connectionist Claim	187	(1)
The Connectionist Research Program	188	(2)
Characteristics of the Non-Symbolic Levels	190	(6)
The Neural Level	190	(1)
The Subsymbolic Level	191	(5)
Conclusions	196	(3)
The Notion of `Symbols'	199	(14)
The Notion of Symbol	200	(2)
Types of Symbols	200	(1)
Symbols in the Different Views on AI	201	(1)
The Semantics of Symbols	202	(2)
The Reference of Tokens	203	(1)
Tokens and Their Relation to First-Person Symbols	203	(1)
Token Manipulation Processes	204	(2)
Discussion	206	(7)
Explicit versus Implicit Symbol Processing	207	(1)
Understanding a Symbol's Meaning	208	(1)
Operationalising First-Person Symbols by Third-Person Symbols	209	(4)
Computational Limitations of Connectionist Architectures	213	(20)
The Complexity of Connectionist Architectures	214	(4)
Limitations of Functions Computable by Connectionist Systems	218	(5)
Neural Networks and Their Learning Capabilities	219	(2)
On Modelling Biological Neuronal Functionalities	221	(2)
Unsupervised Learning and Self-Organising Systems	223	(6)
Preliminaries	226	(2)
Computational Limitations of Self-Organisation	228	(1)
Implications for Artificial Intelligence and Cognitive Science	229	(4)
Methodological Discussion	233	(10)
Choosing Abstract Description Levels for Complex Systems	234	(1)
Key Factors Determining the Suitability of a Description Level	235	(1)
The Connectionist Metaphor	236	(1)
The Fallacy of Empirical Validations of AI Techniques	237	(3)
Alternative Approaches to Understanding Cognition and Intelligence	240	(3)
Non-Factual Knowledge	243	(10)
Introduction	243	(5)
How to Acquire Non-Factual Knowledge?	248	(2)
Discussion: Non-Factual Knowledge in AI Systems	250	(3)
Conclusions	253	(12)
Brief Summary of the Book	253	(2)
Symbolic AI	253	(1)
Connectionism	253	(1)
Comparative Analysis of Both Paradigms	254	(1)
On a Research Methodology for AI	255	(4)
The Human Subject Must Be the Measure	255	(2)
General Methodological Objectives	257	(2)
Non-Factual Knowledge	259	(2)
Perspectives	261	(4)
APPENDIX
Appendix A. More Details on Logic	265	(16)
A.1 Models	265	(2)
A.2 Herbrand Interpretations	267	(3)
A.3 Resolution	270	(11)
A.3.1 Unification	274	(3)
A.3.2 The Unification Algorithm	277	(1)
A.3.3 The Resolution Process	278	(3)
Appendix B. Technical Details and Proofs for Chapter 13	281	(22)
B.1 Limitations of Functions Computable by Neural Networks	281	(7)
B.1.1 Neural Networks and Their Learning Capabilities	281	(6)
B.1.2 On Modelling Biological Neuronal Functionalities	287	(1)
B.2 Unsupervised Learning and Self-Organising Systems	288	(15)
B.2.1 Preliminaries	288	(5)
B.2.2 Computational Limitations of Self-Organisation	293	(10)
Bibliography	303	(20)
Authors Index	323	(6)
Subject Index	329