Building Intelligent Agents

by Tecuci

• ISBN13: 9780126851250
• ISBN10: 0126851255
• Format: Paperback
• Copyright: 1998-06-23
• Publisher: Elsevier Science

Summary

Building Intelligent Agents is unique in its comprehensive coverage of the subject. The first part of the book presents an original theory for building intelligent agents and a methodology and tool that implement the theory. The second part of the book presents complex and detailed case studies of building different types of agents: an educational assessment agent, a statistical analysis assessment and support agent, an engineering design assistant, and a virtual military commander. Also featured in this book is Disciple, a toolkit for building interactive agents which function in much the same way as a human apprentice. Disciple-based agents can reason both with incomplete information, but also with information that is potentially incorrect. This approach, in which the agent learns its behavior from its teacher, integrates many machine learning and knowledge acquisition techniques, taking advantage of their complementary strengths to compensate for each others weakness. As a consequence, it significantly reduces (or even eliminates) the involvement of a knowledge engineer in the process of building an intelligent agent.

Table of Contents

Preface	x	(6)
About the Contributing Writers	xvi	(4)
Notation	xx
1. Intelligent Agents	1	(12)
1.1 What is an Intelligent agent?	1	(1)
1.2 What is a learning agent?	2	(1)
1.3 Phases in the development of an intelligent agent's knowledge base	3	(1)
1.4 Approaches to knowledge base development	4	(9)
1.4.1 Knowledge acquisition approaches	4	(1)
1.4.2 Machine learning approaches	5	(1)
1.4.3 Multistrategy learning approaches	6	(3)
1.4.4 Complementary nature of machine learning and knowledge acquisition	9	(1)
1.4.5 Integrated machine learning and knowledge acquisition approaches	10	(3)
2. General Presentation of the Disciple Approach for Building Intelligent Agents	13	(20)
2.1 An overview of the Disciple approach	13	(3)
2.2 Building a manufacturing assistant	16	(7)
2.3 Building an assessment agent for higher-order thinking skills	23	(10)
3. Knowledge Representation and Reasoning	33	(46)
3.1 Knowledge representation	33	(17)
3.1.1 Introduction
3.1.1.1 What is a knowledge representation	33	(1)
3.1.1.2 General features of a knowledge representation	34	(1)
3.1.1.3 Hybrid knowledge representation	35	(1)
3.1.2 Semantic network representation of objects	36	(1)
3.1.2.1 Characterization of instances and concepts	36	(1)
3.1.2.2 Intuitive definition of generalization	37	(1)
3.1.2.3 Properties and relations	38	(1)
3.1.2.4 Definition of instances and concepts	39	(3)
3.1.2.5 General characterization of the semantic network	42	(1)
3.1.3 General concepts and rules	42	(1)
3.1.3.1 Representation language for general concepts and rules	42	(2)
3.1.3.2 General concepts	44	(1)
3.1.3.3 Rules	45	(3)
3.1.3.4 Plausible Version Space rules	48	(2)
3.2 Generalization in the representation language of the agent	50	(16)
3.2.1 Formal definition of generalization	50	(1)
3.2.1.1 Term generalization	50	(1)
3.2.1.2 Clause generalization	50	(2)
3.2.1.3 Conjunctive formula generalization	52	(1)
3.2.1.4 Disjunctive formula generalization	53	(1)
3.2.2 Generalization rules	54	(1)
3.2.2.1 Turning constants into variables	54	(1)
3.2.2.2 Turning occurrences of a variable into different variables	55	(1)
3.2.2.3 Climbing the generalization hierarchies	56	(1)
3.2.2.4 Dropping conditions	57	(1)
3.2.2.5 Adding alternatives	57	(1)
3.2.2.6 Generalizing numbers to intervals	57	(1)
3.2.2.7 Using theorems	57	(1)
3.2.3 Other definitions of generalizations	58	(4)
3.2.4 Rules as generalizations of examples of problem solving episodes	62	(4)
3.3 Elementary problem solving methods	66	(13)
3.3.1 Use of transitivity and inheritance	67	(1)
3.3.1.1 Properties of ISA, INSTANCE-OF and IS	67	(1)
3.3.1.2 Inheritance of features	67	(1)
3.3.1.3 Default inheritance	68	(1)
3.3.1.4 Multiple inheritance	69	(1)
3.3.2 Network matching	69	(3)
3.3.3 Example generation	72	(1)
3.3.4 Rule matching	72	(5)
3.3.5 Reasoning with plausible version space rules	77	(2)
4. Knowledge Acquisition and Learning	79	(68)
4.1 Knowledge elicitation	80	(5)
4.1.1 Knowledge elicitation goals and principles	80	(1)
4.1.2 Implicit associations between the knowledge elements	81	(3)
4.1.3 Knowledge elicitation processes	84	(1)
4.2 Rule learning	85	(16)
4.2.1 The rule learning problem	85	(4)
4.2.2 The rule learning method	89	(1)
4.2.2.1 What is an explanation of an example	89	(3)
4.2.2.2 The explanation generation method	92	(2)
4.2.2.3 Analogical reasoning	94	(2)
4.2.2.4 The analogy-based generalization method	96	(3)
4.2.3 Characterization of the learned PVS rule	99	(2)
4.3 Rule refinement	101	(29)
4.3.1 The rule refinement problem	101	(1)
4.3.2 The rule refinement method	102	(1)
4.3.3 Rule refinement through active experimentation with its plausible upper bound	103	(1)
4.3.3.1 The active experimentation process	103	(2)
4.3.3.2 Experimentation and verification	105	(3)
4.3.3.3 Refining the PVS rule with a positive example	108	(7)
4.3.3.4 Refining the PVS rule with a negative example	115	(12)
4.3.4 Rule verification through active experimentation with its plausible lower bound	127	(1)
4.3.5 Rule refinement with external examples	127	(1)
4.3.6 Characterization of the refined rule	128	(2)
4.4 Exception-driven knowledge base refinement	130	(12)
4.4.1 The exception-driven knowledge base refinement problem	130	(1)
4.4.2 The consistency driven knowledge base refinement method	131	(4)
4.4.3 Consistency driven discovery and elicitation of features	135	(2)
4.4.4 Consistency driven discovery and elicitation of concepts	137	(1)
4.4.5 The completeness driven knowledge base refinement method	138	(3)
4.4.6 Completeness driven elicitation of features	141	(1)
4.4.7 Completeness driven discovery and elicitation of concepts	141	(1)
4.5 An analysis of the expert-agent interactions	142	(4)
4.5.1 Types of interactions	142	(2)
4.5.2 The utility of explanations	144	(2)
4.6 Final remarks	146	(1)
5. The Disciple Shell and Methodology	147	(32)
5.1 Architecture of the Disciple shell	147	(4)
5.1.1 Knowledge acquisition and learning component	148	(1)
5.1.2 Problem-solving component	149	(1)
5.1.3 Knowledge base manager	149	(1)
5.1.3.1 Dictionary	150	(1)
5.1.3.2 Knowledge query language	150	(1)
5.2 The methodology for building intelligent agents	151	(3)
5.2.1 An overview of the Disciple methodology	152	(1)
5.2.2 Issues in developing Disciple agents	153	(1)
5.3 Expert-agent interactions during the knowledge elicitation process	154	(7)
5.3.1 Overview of the interaction process	155	(1)
5.3.2 Interactions with the concept browser	156	(2)
5.3.3 Interactions with the concept editor	158	(1)
5.3.4 Interactions with the dictionary browser/editor	158	(3)
5.3.5 Interactions with the association browser	161	(1)
5.4 Expert-agent interactions during the rule learning process	161	(9)
5.4.1 Overview of the interaction process	162	(2)
5.4.2 Interactions with the example editor	164	(2)
5.4.3 Interactions with the rule learner	166	(1)
5.4.4 Interactions with the rule browser	166	(3)
5.4.5 Interactions with the rule editor	169	(1)
5.5 Expert-agent interactions during the rule refinement process	170	(8)
5.5.1 Overview of the interaction process	171	(1)
5.5.2 Interactions with the rule refiner	171	(5)
5.5.3 Interactions with the explanation grapher	176	(2)
5.6 Final remarks	178	(1)
6. Case Study: Assessment Agent for Higher-Order Thinking Skills in History	179	(50)
6.1 Characterization of two types of assessment agents	179	(2)
6.2 Developing a customized Disciple agent	181	(9)
6.2.1 Analyzing the problem domain and defining agent requirements	181	(1)
6.2.1.1 Application domain	181	(1)
6.2.1.2 Use of the reporter paradigm	181	(1)
6.2.1.3 Types of test questions	182	(1)
6.2.1.4 Modes of operation and feedback provided to the student	182	(1)
6.2.1.5 Dynamic and context-sensitive generation of tests	182	(1)
6.2.2 Developing the top level ontology of the knowledge base	183	(5)
6.2.3 Developing domain dependent modules	188	(1)
6.2.3.1 The source viewer	188	(2)
6.2.3.2 The customized example editor	190	(1)
6.2.3.3 The example viewer	190	(1)
6.3 Building the initial knowledge base	190	(3)
6.3.1 Defining the history curriculum	191	(1)
6.3.2 Selecting and representing historical sources	191	(2)
6.3.3 Populating the semantic network with other necessary concepts and instances	193	(1)
6.4 Teaching the agent how to judge the relevancy of a source with respect to a task	193	(9)
6.4.1 Giving the agent an example of a task and a relevant source	193	(1)
6.4.2 Helping the agent understand why the source is relevant	194	(4)
6.4.3 Supervising the agent as it evaluates the relevance of other sources to similar tasks	198	(1)
6.4.3.1 Confirming the agent's evaluation	198	(1)
6.4.3.2 Rejecting agent's evaluation and helping it to understand its mistake	199	(3)
6.5 Developing the assessment engine and the graphical user interface	202	(13)
6.5.1 Augmenting and adjusting the patterns associated with the learned rules	206	(1)
6.5.2 Sample agent-student interaction during an If-relevant test question	207	(1)
6.5.3 Generation of If-relevant test questions with relevant sources	207	(5)
6.5.4 Generation of If-relevant test questions with irrelevant sources	212	(1)
6.5.5 Sample agent-student interaction during a Which-relevant test question	212	(1)
6.5.6 Generation of Which-relevant test questions	213	(1)
6.5.7 Sample agent-student interaction during a Why-relevant test question	213	(1)
6.5.8 Generation of Why-relevant test questions	213	(2)
6.6 Verifying, validating and maintaining the agent	215	(7)
6.7 Developing the integrated assessment agent	222	(5)
6.8 Final remarks	227	(2)
7. Case Study: Statistical Analysis Assessment and Support Agent	229	(27)
7.1 The Natural World-the course	230	(2)
7.2 Assessment in The Natural World	232	(2)
7.3 Sample interactions between the agent and the student	234	(8)
7.3.1 Example 1 -- analysis of cigarette data	234	(4)
7.3.2 Example 2 -- analysis of iris data	238	(2)
7.3.3 Example 3 -- analysis of flies data	240	(2)
7.4 Building of the initial knowledge base	242	(3)
7.5 Teaching the agent to generate test questions	245	(8)
7.6 Final remarks	253	(3)
8. Case Study: Design Assistant for Configuring Computer Systems	256	(18)
8.1 Disciple methodology applied to engineering domains	256	(3)
8.2 Defining the initial knowledge base of the design assistant	259	(5)
8.3 The shared expertise method of problem solving and learning	264	(3)
8.4 Training and using the design assistant	267	(6)
8.4.1 Learning a heuristic design rule from a new creative design	267	(2)
8.4.2 Extending the RDS	269	(1)
8.4.3 Improving the IDS	270	(1)
8.4.4 Handling exceptions	271	(2)
8.5 Final remarks	273	(1)
9. Case Study: Virtual Agent for Distributed Interactive Simulations	274	(23)
9.1 Distributed interactive simulations	274	(2)
9.2 Developing a virtual armored company commander	276	(2)
9.3 Defining the tasks and the top-level ontology of the agent	278	(1)
9.4 Developing the semantic network	278	(4)
9.5 Teaching the agent how to accomplish a defensive mission	282	(11)
9.5.1 Giving the agent an example	282	(1)
9.5.2 Helping the agent to understand why the positioning is correct	282	(2)
9.5.3 Supervising the agent as it generates other defensive placements	284	(3)
9.5.3.1 Rejecting agent's solution and helping it to understand its mistake	287	(4)
9.5.3.2 Confirming the agent's solution	291	(2)
9.6 Final remarks	293	(4)
Selected Bibliography of Machine Learning, Knowledge Acquisition, and Intelligent Agents Research	297	(19)
Index	316