Artificial Intelligence Expert Systems Knowledge Representation

1
Artificial Intelligence / Expert
SystemsKnowledge Representation

• Justin Gaudry
• May 17, 2007

2
Knowledge

• Theoretical or practical understanding of a
  subject or a domain
• Expert one with knowledge in both facts and
  rules (and perhaps experience) in a limited domain

3
Knowledge Representation

• Psychology
• How people store and process information
• Artificial Intelligence
• How to store knowledge so that programs can
  process it to achieve the appearance of
  intelligence
• How to store the real world internally

4
Importance and Difficulty

• Granularity
• Expressiveness and efficiency
• The ability to process information with enough
  detail to produce inferences yet enough
  generality to be efficient
• Abstraction
• Representation of only that information which is
  needed for a given purpose

5
Representation Requirements

• Handle qualitative knowledge
• Allow new knowledge to be inferred from a set of
  facts and rules
• Allow representation of general principles as
  well as specific situations
• Capture complex semantic meaning
• Allow for meta-level reasoning

Luger and Stubblefield, Artificial Intelligence
6
Qualitative Knowledge

• Quantitative explicit logistical description
• Qualitative relationship between objects

7
Inferences

• Do not store an inflexible description of every
  situation
• Provide abstract descriptions of classes of
  objects and situations (rules) then specific
  domain-specific instances (facts)
• Ax chase(X, cat) and bites(X,Y) -gt dog(X)

8
General and Specific Representation

• Variables
• Type and scope rules in traditional languages are
  too restrictive
• Allow dynamic binding
• Rules of inference
• (P(x) -gt Q(x) and P(x)) -gt Q(x) modus ponens
• P(x,y) and P(y,z) -gt P(x,z) hyp. syllogism
• Not true in all cases, transitive property

9
Semantic Meaning

• Highly structured interrelated knowledge
• Represent data and relationships
• Causal relationship between events over time
• Semantic network promotes efficiency over
  predicate calculus

10
Meta-Level Reasoning

• Meta-X x about x
• Meta-photography photography about photography
• Meta-poetry poetry about writing poetry
• Meta-knowledge knowledge about ones own
  knowledge
• Not only solve problems but explain how and why
  certain decisions were made
• George believes that Tom knows that Marys car
  is red.

11
Schemes

• Logical representation
• Expressions in formal logic predicate calculus
• Procedural representation
• Sets of instructions for solving a problem
  rule-based systems
• Network representation
• Graph where nodes represent objects or concepts
  and edges represent relationships semantic nets
• Structured representation
• Each node has slots with attached values frames

12
Predicate Calculus

• Predicate represents some property or
  relationship among its arguments
• Specific to a particular domain yet formal logic
  rules apply to all situations
• dog(fido) cat(fifi)
• chase(dog, cat) scratch(cat, dog)
• bites(dog, X) bites(cat, X)
• pet(dog) pet(cat)
• (Predicate calculus proofs done in class)

13
Rule-Based Systems

• Implementation of a predicate calculus
• Set of rules and associated facts
• Rules expressed as if-then
• If ltantecedentgt
• Then ltconclusion1gt
• ltconclusion2gt
• ltconclusion3gt
• If X is a cat then X scratches dogs and bites Y
• (Simplification and conjunction rules discussed)

14
Semantic Networks

• All features of knowledge base represented in
  graph
• Can lead to incorrect conclusions without rule
  constraints (is-a, has-char) vs. (works-for,
  is-a)
• human(ed) human(mary) robot(hal)
• autonomous(human) autonomous(robot)
• biological(human) mechanical(robot)
• owns(mary, hal) worksFor(hal, mary)
• worksFor(ed, mary) (net drawn in class)

15
Frames

• Extension of network relationship
• Each instance or class of object represented with
  frame
• Each frame has slots and slot values assigned to
  it
• Slots represent property or relationship
• All information about object immediately available

16
Frames

• Procedures
• Associated with frames with CAN relationship
• Behavior or action which takes place upon request
• Demon (not daemon)
• Automatic procedure which runs when value changes
  or event takes place

17
Frames

• Simple object-oriented example
• Rectangle hasChar Length
• Rectangle hasChar Width
• Rectangle can calcArea
• Rectangle can setLengthD
• Rectangle can setWidthD
• Demon verify value gt 0

18
Search Spaces

• Also known as State Spaces
• Represented by graph with paths that represent
  actions (state transitions)
• Each state is a situation which might be reached
  by the system
• Finite-state automata (machine)
• The Ferrymans Dilemma
• Ferryman, wolf, sheep, cabbage, boat
• (To be continued on Tuesday)