INTROSPECTIVE MULTISTRATEGY LEARNING:
Constructing a Learning Strategy under Reasoning Failure

Table of Contents

DEDICATION
ACKNOWLEDGEMENTS
PREFACE
LIST OF TABLES
LIST OF ILLUSTRATIONS
LIST OF ABBREVIATIONS
SUMMARY

Part One
PRELIMINARIES

CHAPTER I
INTRODUCTION
1.1 Research Motivation
1.2 The Problem
1.3 The Solution
1.4 Learning Goals and the Decomposition of the Problem
1.4.1 The Learning Goal Metaphor
1.4.2 The Decomposition of a Learning Goal
1.4.2.1 What kinds of reasoning failure exist? (Q6)
1.4.2.2 What can cause reasoning failure? (Q4)
1.4.2.3 At what level of granularity should reasoning be represented? (Q7)
1.4.2.4 How to represent mental states and reasoning mechanisms? (Q5)
1.4.2.5 How to explain a reasoning failure? (Q3)
1.4.2.6 How to decide what to learn? (Q2)
1.4.2.7 How to choose or construct a learning strategy? (Q1)
1.4.2.8 How can the research be evaluated? (Q0)
1.5 Overview of the Dissertation

CHAPTER II
CONTENT THEORIES AND PROCESS THEORIES
2.1 The Drug-Bust Examples
2.1.1 A Common Contradiction
2.1.2 A Baffling Situation
2.2 Knowledge and Process
2.3 The Domain of Story-Understanding Failures
2.4 Prolog to Parts Two and Three: The content and the process

CHAPTER III
SYMPTOMS AND CAUSES OF FAILURE: THE CONTENT
3.1 A General Model of Expectation-Driven Reasoning
3.2 Types of Reasoning Failure
3.2.1 Four Basic Cases
3.2.1.1 Contradiction
3.2.1.2 Impasse
3.2.1.3 False expectation
3.2.1.4 The degenerate case
3.2.2 Extending the Analysis
3.2.2.1 Surprise
3.2.2.2 Unexpected success
3.3 Causal Factors in Reasoning Failure
3.3.1 Selection Factors
3.3.2 A Taxonomy of Reasoning Failure Causes
3.3.3 Knowledge States
3.3.3.1 Domain knowledge
3.3.3.2 Knowledge selection
3.3.4 Goal States
3.3.4.1 Goal generation
3.3.4.2 Goal selection
3.3.5 Processes
3.3.5.1 Processing strategy
3.3.5.2 Strategy selection
3.3.6 Environment
3.3.6.1 Input
3.3.6.2 Input selection
3.3.7 Causal Invariants
3.4 Summary and Discussion

CHAPTER IV
MENTAL STATES AND MECHANISMS: THE REPRESENTATION
4.1 Epistemology and Ontology
4.2 Granularity of Representation
4.3 Representing Forgetting: An example
4.3.1 Logic
4.3.2 Conceptual Dependency
4.3.3 Explanation Pattern Theory
4.4 Meta-Explanation Patterns
4.4.1 Trace Meta-XPs
4.4.2 Introspective Meta-XPs
4.5 Vocabulary
4.6 Representing Reasoning Success
4.7 Representing Reasoning Failure to Support Learning
4.7.1 Contradiction
4.7.2 Impasse
4.7.3 False Expectation
4.7.4 Surprise
4.7.5 Unexpected Success
4.8 Summary and Discussion

CHAPTER V
A PROCESS THEORY OF UNDERSTANDING AND LEARNING
5.1 Theoretical Assumptions
5.2 Multistrategy Reasoning
5.3 Process Model of Understanding
5.3.1 Three Sub-processes of Understanding
5.3.2 Understanding Elvis' Behavior
5.3.3 Question-Driven Understanding
5.4 Process Model of Learning
5.4.1 Multistrategy Learning
5.4.2 Process Divisions within the Model of Learning
5.4.3 Generating Changes to the BK
5.4.3.1 Functional justification for introspection as a component of learning
5.4.3.2 Overview of the IML algorithm
Blame assignment (step 2a, Figure 48)
Deciding what to learn (step 2b, Figure 48)
Learning-strategy construction (step 2c, Figure 48)
5.5 Comparison of Learning and Understanding
5.6 Summary

CHAPTER VI
CASE-BASED INTROSPECTION
6.1 Case-based reasoning and introspection
6.2 Blame Assignment: Explaining reasoning failure
6.2.1 The Explanation Pattern Application Algorithm
6.2.2 Returning to the Drug Bust Example
6.3 Deciding What to Learn: Spawning learning goals
6.3.1 Explicit Learning Goals
6.3.2 Learning Goals in the Drug-Bust Example
6.4 Summary and Discussion

CHAPTER VII
LEARNING AS A NON-LINEAR PLANNING TASK
7.1 Blame Assignment and Repair: Tight versus loose coupling
7.1.1 Direct Mappings: Tight coupling
7.1.2 Indirect Mappings: Loose coupling
7.2 Learning-Strategy Construction: Learning as a planning task
7.2.1 Managing Goal Interactions with Nonlinear Planning
7.2.2 Planning for Blocks World Goals
7.2.3 Planning for a Knowledge Differentiation Goal
7.2.4 Planning for a Knowledge Reconciliation Goal
7.3 Strategy Execution: Performing the learning and the aftermath
7.4 The Planning Metaphor in Learning
7.4.1 Generality of the Metaphor
7.4.2 Advantages of the Planning Metaphor
7.5 Summary and Discussion

CHAPTER VIII
META-AQUA
8.1 Meta-AQUA System Architecture
8.2 The Performance Subsystem
8.3 Input Subsystem
8.3.1 The Tale-Spin Story-Generator
8.3.2 The Elvis World
8.3.3 Interface to the Performance System
8.4 Memory
8.4.1 Types and Tokens
8.4.2 Indexing
8.4.3 Remindings in Opportunistic Memory
8.5 Learning Subsystem
8.5.1 Divisions of the Learner
8.5.2 The Implemented Space of Explanation Failures
8.5.3 Learning about higher-order knowledge
8.5.4 Forgetting a Learned Explanation
8.6 Summary

CHAPTER IX
EVALUATION
9.1 The Hypotheses
9.1.1 Hypothesis Number 1: Introspection facilitates learning
9.1.1.1 Fully introspective and semi-introspective learning
9.1.1.2 The null hypothesis
9.1.2 Hypothesis Number 2: IML theory is a sufficient model of introspection
9.2 Computational Evaluation
9.2.1 The Experimental Conditions (Independent Variable)
9.2.2 The Evaluation Criteria (Dependent Variables)
9.2.3 The Empirical Data
9.2.3.1 Run Number Four
9.2.3.2 Quantitative Results
9.3 Psychological Plausibility
9.3.1 Learning to program in LISP
9.3.2 Meta-AQUA Simulation
9.3.3 Learning to Troubleshoot Circuit Boards
9.4 Summary and Discussion

CHAPTER X
FUTURE RESEARCH
10.1 IML Theory and Implementation: Open issues
10.1.1 The Scope of Blame-Assignment.
10.1.2 Additional Goal Interactions.
10.1.3 Enlarging the Toolbox.
10.1.4 Parallelism in Learning Plans
10.1.5 Learning New IMXPs.
10.1.6 Auto-Validation of Learning.
10.1.7 Failure Types as a Cognitive Category
10.1.8 Extending the Computational Evaluation
10.1.8.1 Establishing the hypothesis that introspection facilitates learning
10.1.8.2 Determining the conditions under which the hypothesis holds
10.1.9 Extending the Model of Human Learning
10.1.10 Extending the Representational Vocabulary
10.2 Learning Bias and Category Membership: An extension
10.3 Understanding student explanation failures: An application
10.4 Summary

CHAPTER XI
RELATED RESEARCH
11.1 Artificial Intelligence, Metareasoning, and Introspective Learning
11.1.1 Logic and Belief Introspection
11.1.2 Knowledge-Based Systems, Quasi-Introspection, and Related Theories
11.1.3 Relation of AI Research to IML Theory
11.2 Psychology, Metacognition, and Human Learning
11.2.1 Cognition and Metacognition
11.2.2 Problem Solving and Metacognition
11.2.3 Metamemory
11.2.4 Relation of Psychological Research to IML Theory
11.3 Summary and Discussion

CHAPTER XII
CONCLUSIONS
12.1 Major Points of the Dissertation
12.2 Part One: Motivations and Defining the Problem
12.3 Part Two: Content Theory of Introspective Multistrategy Learning
12.4 Part Three: Process Theory of Introspective Multistrategy Learning
12.5 Part Four: Evaluation and Implementation of the Theory
12.6 Contributions

CHAPTER XIII
EPILOGUE
13.1 The Processes:
Integration of problem solving, understanding, and learning
13.2 The Knowledge:
Integration of world knowledge, metaknowledge and self-knowledge
13.3 The "Turing Test"

APPENDIX A
THE DEGREES OF FREEDOM IN LEARNING

APPENDIX B
META-AQUA OUTPUT IN STORY UNDERSTANDING MODE

APPENDIX C
STORY LISTINGS FOR RUN NUMBER FOUR

APPENDIX D
META-AQUA OUTPUT IN LISP PROGRAMMING MODE

REFERENCES

NAME INDEX
PROGRAM INDEX
SUBJECT INDEX

VITA