Knowledge Representation and Reasoning

Let's start by going back to our discussions about agents and recall that

Intelligent behavior requires knowledge about the world
• Procedural, e.g., functions
• Using knowledge = executing the procedure
• Declarative, e.g., facts
• Using knowledge = performing inference
• Deliberative agents
• Can represent and reason with knowledge
• Exhibit logical rationality

Quite simply, Knowledge representation (KR) is a surrogate and it is a powerful surrogate at that! Why?

Because ----

A declarative knowledge representation
• Encodes facts that are true in the world into sentences
• Reasoning is performed by manipulating sentences according to sound
rules of inference
• The results of inference are sentences that correspond to facts that are
true in the world
• The correspondence between facts that hold in the world and sentences
that describe the world ensures semantic grounding of the representation
• Allows agents to substitute thinking for acting in the world
• Known facts: The coffee is hot; coffee is a liquid; a hot liquid will burn
your tongue;
• Inferred fact: Coffee will burn your tongue

It is also important to remember that KR is a set of ontological commitments (p.s the concept of ontology dates back to Aristotle)
• What does an agent care about?
Entities
• coffee, liquid, tongue
Properties
• being hot, being able to burn
Relationships
• Coffee is a liquid
• KR involves abstraction, simplification
• A representation is
• a (logical) model of the world
• like a cartoon
• All models are wrong, but some are useful

Finally, KR is a theory of intelligent reasoning
• How can knowledge be encoded ?
• Syntax
• What does the encoded knowledge mean?
• Semantics (entailment)
• Inferences that are sanctioned by the semantics
• What can we infer from what we know?
• Inferences that can be performed (by rules of inference)
• Soundness, completeness, efficiency
• How can we manage inference?
• What should we infer from among the things we can infer?

KR (Knowledge Representation) and Reasoning (KRR) is a very powerful AI method and thus takes several chapters in the AIMA book.

Course Book Reference for this submodule

Part III Knowledge Representation

Chapter 7 Logical Agents ... 208
Chapter 8 First-Order Logic ... 251
Chapter 9 Inference in First-Order Logic ... 280
Chapter10 Knowledge Representation ... 314

Online Lectures on KRR

Lecture KRR Part I

Lecture KRR Part II

Additional Material

Genesereth, M. R., & Nilsson, N. J., Logical Foundations of Artificial Intelligence. Palo Alto, CA: Morgan Kaufmann (1987).