Knowledge-Based AI (CS 7637) - Course Navigation

Overview

Knowledge-Based AI (KBAI) is the study of artificial intelligence systems that use structured knowledge representations to solve problems through human-like reasoning, learning, and memory. This course explores cognitive systems that exhibit human-level intelligence through the interaction of these three fundamental components.

Core Philosophy: KBAI represents a unified theory where reasoning, learning, and memory are intimately connected. We learn so we can reason; the results of reasoning lead to additional learning; and memory provides the knowledge foundation for both.

Course Organization

This course is organized into 9 comprehensive modules covering the fundamental topics, reasoning strategies, learning methods, and advanced AI concepts.

Learning Path

Fundamentals → Core Reasoning → Learning Methods → Logic & Planning
     ↓              ↓                 ↓                    ↓
Language & Common Sense → Advanced Reasoning → Applied Problem Solving
     ↓                          ↓                          ↓
          Metacognition & Advanced Topics

Module Guide

1. Fundamentals of KBAI

Lessons 1-3 | Introduction, Cognitive Systems, Semantic Networks

Core concepts: Knowledge representations, cognitive architectures, semantic networks as knowledge structures, characteristics of AI agents and problems.

Start here if you’re new to KBAI or want to understand the foundational principles.

Key Topics:

• Four schools of AI: Acting/Thinking, Human-like/Optimally
• Cognitive system architecture: Reaction, Deliberation, Metacognition
• Semantic networks: Nodes, links, and structured representations
• Guards & Prisoners problem

2. Core Reasoning Strategies

Lessons 4-6 | Generate & Test, Means-Ends Analysis, Production Systems

Problem-solving methods that form the basis of AI reasoning. Learn how to map percepts to actions through systematic approaches.

Start here if you understand representations and want to learn reasoning methods.

Key Topics:

• Generate & Test: Smart generators vs. smart testers
• Means-Ends Analysis: Goal-driven problem solving
• Problem Reduction: Decomposing complex problems
• Production Systems: Cognitive architectures with rules
• Chunking: Learning from impasses

3. Learning Methods

Lessons 8-11, 19, 23 | Case-based Reasoning, Classification, Version Spaces

How AI agents learn from experience through recording cases, concept learning, and systematic generalization/specialization.

Start here if you want to understand learning mechanisms in KBAI.

Key Topics:

• Learning by Recording Cases: k-nearest neighbor
• Case-Based Reasoning: Retrieve, adapt, evaluate, store
• Incremental Concept Learning: Variabilization, specialization, generalization
• Classification: Prototype vs. exemplar concepts
• Version Spaces: Hypothesis refinement
• Learning by Correcting Mistakes: Error detection and repair

4. Logic and Planning

Lessons 12-13 | Formal Logic, Planning Strategies

Logical representations and planning methods for goal achievement using operators and state spaces.

Start here for systematic approaches to goal-driven problem solving.

Key Topics:

• Predicate logic: Conjunctions, disjunctions, implications
• Truth tables and logical equivalences
• Rules of inference: Modus ponens, resolution
• Partial order planning: Open preconditions, conflict detection
• State space search

5. Language and Common Sense Reasoning

Lessons 7, 14-16 | Frames, Understanding, Scripts

How AI systems understand language and reason about everyday situations using structured knowledge.

Start here for natural language understanding and common sense reasoning.

Key Topics:

• Frames: Slots, fillers, default values, stereotypes
• Thematic role systems: Agent, object, location, time
• Common sense reasoning: Primitive actions, implied actions
• Scripts: Stereotypical event sequences with tracks
• Story understanding

6. Advanced Reasoning

Lessons 17-18 | Explanation-Based Learning, Analogical Reasoning

Advanced learning and reasoning methods including cross-domain analogy and explanation-driven learning.

Start here for sophisticated reasoning techniques.

Key Topics:

• Explanation-Based Learning: Abstraction and transfer
• Analogical Reasoning: Retrieval, mapping, transfer, evaluation
• Spectrum of similarity: Semantic, pragmatic, structural
• Design by analogy
• Cross-domain analogy

7. Applied Problem Solving

Lessons 20-22 | Constraint Propagation, Configuration, Diagnosis

Practical AI applications using constraints, configuration, and diagnostic reasoning.

Start here for real-world AI applications.

Key Topics:

• Constraint propagation: Visual reasoning from 2D to 3D
• Configuration: Design through constraint satisfaction
• Diagnosis as abduction: Hypothesis generation and testing
• Connection to classification and planning

8. Metacognition and Advanced Topics

Lessons 24-26 | Meta-reasoning, Visuospatial Reasoning, Design & Creativity

Advanced topics including reasoning about reasoning, visual problem solving, and creative AI systems.

Start here for cutting-edge KBAI topics.

Key Topics:

• Meta-reasoning: Reasoning about deliberation and reaction
• Strategy selection and knowledge gaps
• Visuospatial reasoning: Ravens Progressive Matrices
• Design and creativity
• Symbol grounding problem
• Systems thinking

Key Projects & Examples

Throughout the course, several recurring examples illustrate KBAI concepts:

Ravens Progressive Matrices

The primary project: Building AI agents that solve visual analogy problems from intelligence tests. This project integrates knowledge representation, reasoning, and learning.

Complexity progression:

• 2×1 matrices: Simple transformations
• 2×2 matrices: Pattern completion
• 3×3 matrices: Complex rule systems

Classic AI Problems

• Guards & Prisoners: State space search, semantic networks
• Blocks World: Means-ends analysis, planning, problem reduction
• Baseball Pitcher: Production systems, action selection, chunking

Recommended Learning Paths

Path 1: Foundation-First (Recommended for beginners)

1. Fundamentals → 2. Core Reasoning → 3. Learning Methods → 4. Logic & Planning → 5. Language & Common Sense → 6. Advanced Reasoning → 7. Applied → 8. Metacognition

Path 2: Problem-Solving Focus

1. Fundamentals → 2. Core Reasoning → 4. Logic & Planning → 7. Applied → 3. Learning Methods → 6. Advanced Reasoning

Path 3: Learning-Centric

1. Fundamentals → 3. Learning Methods → 6. Advanced Reasoning → 2. Core Reasoning → 4. Logic & Planning

Path 4: Applications-First

1. Fundamentals → 7. Applied → 2. Core Reasoning → 5. Language & Common Sense → 3. Learning Methods

Unifying Principles

Seven key principles run throughout this course:

1. Knowledge Representations are central to KBAI
2. Reasoning, Learning, Memory are intimately connected
3. Cognitive architectures separate content from behavior
4. Analogy and abstraction enable transfer and generalization
5. Generate and test underlies many AI methods
6. Meta-reasoning enables self-improvement
7. Human cognition inspires and validates AI design

Core Concepts Cross-Reference

Study Tips

1. Follow the cognitive connection sections - They link AI techniques to human cognition
2. Practice with Ravens problems - The project integrates all concepts
3. Compare knowledge representations - Understand when to use semantic networks vs. frames vs. production rules
4. Connect reasoning and learning - Notice how reasoning drives what/when/why to learn
5. Build incrementally - Start with simple problems, add complexity gradually

Prerequisites

• Basic programming skills (for projects)
• Familiarity with data structures
• Logical thinking and problem-solving ability
• Interest in cognitive science helpful but not required

Course Resources

• Main Project: Raven’s Progressive Matrices AI agent
• Classic Problems: Guards & Prisoners, Blocks World, Baseball Pitcher
• Key Readings: Spread throughout lessons on cognitive architectures, chunking, scripts, analogical reasoning

See Also

• Start with Fundamentals
• Begin Problem Solving
• Explore Learning
• Jump to Advanced Topics

This course explores how knowledge-based AI can achieve human-level intelligence through the unified interaction of reasoning, learning, and memory.