This document discusses different types of knowledge and methods for knowledge acquisition. It describes declarative and procedural knowledge, as well as the knowledge acquisition paradox where experts have difficulty verbalizing their knowledge. Various knowledge acquisition methods are outlined, including observation, problem discussion, and protocol analysis. Knowledge representation techniques like rules, semantic networks, frames, and predicate logic are also introduced.
2. Types of Knowledge
• Declarative Knowledge - tells us the facts
– Facts, knowledge about objects and relationships
– Descriptive representation of knowledge
– It is often shallow knowledge
· Procedural Knowledge - tells us what to do
– Knowledge about procedures involved in solving
problems
· Declarative Knowledge - tells us facts and
procedural knowledge tells us what to do
3. Knowledge Acquisition Paradox
· The more competent a Domain Expert (DE)
becomes, the less able they are to describe
the knowledge they use to solve problems
· Don’t be your own expert
· Don’t believe everything experts say
4. Difficulties of Knowledge Acquisition
• Difficulty in verbalizing
– Reasoning process too broad
– Use of combined and compiled knowledge
– Unaware of the individual steps taken to reach a
solution
• Difficulties in transferring to a machine
– Machine works at a more basic level, but the
expert seldom operates at a basic level
5. Difficulties of Knowledge Acquisition
• Difficulties in structuring knowledge
– Losing a significant amount of knowledge
when structuring implicit knowledge
• Domain Expert’s unwillingness
– Unavailable
– Uncooperative
– No knowledge of computers and Expert
Systems
6. Knowledge Acquisition Methods 1
· On-site observation
· Watch the expert solving real problems on the
job
• We are not the experts, so we research the particular
area BEFORE sitting down with the Domain
Expert(s)
• Ex: Sometimes a Doctor brings a Student with
them/Student learns from the Expert
7. Knowledge Acquisition Methods 2
· Problem discussion - observe at first
· Explore the kinds of data, knowledge, and procedures
needed to solve specific problems
· How does the problem differ from prototypical problems in
the domain?
· How is this problem different from others?
· What different approach do you use?
· Types of data required and kinds of solutions adequate for
the problem?
· What kinds of knowledge are needed to solve the problem?
· What constitutes an adequate explanation or justification of
a problem solution?
8. Knowledge Acquisition Methods 3
· Problem Description
· Have the expert describe a prototypical problem
for each category of answer in the domain
· Protocol Analysis (Problem Analysis)
· Present the expert with a series of realistic
problems to solve aloud, probing for the rationale
behind the reasoning steps (solve the problem
verbally)
· Widely used in psychology
· Ex: Dermatology-Psoriasis
· Expert Syst. to diagnose Psoriasis
· Color?
· How long rash lasts?
· Where is the rash?
9. Knowledge Acquisition Methods 4
• Repertory Grid Analysis
– Identify important objects
– Identify important attributes
• Specific objects
– Example: Rash/Color/Duration/Level of itching/Local
or whole body?
– For each attributes, establish a bipolar
scale with differentiable characteristics and
their opposites
– Ex: Computer Language
10. Repertory Grid Analysis
• Assisting in selecting a computer language
– Identify objectives
• LISP, C (Procedural Lang), C++(OOP Lang)
– Attributes
• Availability, Ease of Programming, Training
Time
• Orientation
– Traits
• high, low, symbolic, numeric
11. Reasoning Methods
• Deductive Reasoning
• Inductive Reasoning
• Forward Reasoning (Chaining)
– Reasoning starts with raw facts
• Backward Reasoning (Chaining)
– Reasoning starts with hypothesis as in
statistics, them moves to prove or disprove
hypothesis
12. RGA Input for Selecting a Computer
Language
Attributes: Trait or Opposite
Availability: Widely Available or Not Available
Ease of Programming: High or Low
(C++) (C)
Training Time: Low or High
Orientation: Symbolic or Numeric
Example: The Animal Problem – Done in LISP – “Symbol Oriented
Can store colors – Red/Blue/Orange/Green
1 variable can be 26 Char long/1 char long
13. Automatic Knowledge Acquisition
Techniques
• Methods
– Rule Induction - DE provides some examples
similar to Data Mining, then apply
Statistical/Mathematical Techniques such as
Multivariate Regression
– Artificial Neural Net (ANN) - Qualitative
Approach-Statistical & Mathematical
Methods/Dev. Intelligent Machine/Data Mining
14. Automatic Knowledge Acquisition
Techniques
• Methods
– Case-based Reasoning - asking DE to provide
case/Law - Attorney
• Work by previous cases/Dev. argument from
previous cases
• Use previous as base argument
– Example: Help Desk
» Printer not functioning
» Refer to previous case from “n” weeks ago
15. Automatic Knowledge Acquisition
Techniques
• Methods
– Model-based Reasoning
• Applicable to design of an engineering application
• Give me specifications of some hardware
• Used often in NASA
• Build a model using DE knowledge
16. Knowledge Representation
• Logic is used heavily in AI
– Prepositional Logic
– Predicate Logic
– Rules (easiest to represent)
– Semantic Nets
– Frame
– Object
17. Propositional Logic
• It is raining
– RAINING
• Proposition/Propositional Logic - Is this true or
false? Is it raining now?
• It is sunny
• We can deduce whether a certain
proposition (fact) is true or false
18. Proposition Logic
• Propositional logic cannot drive the
association
• Socrates is a man (true or false)
– SOCRATESMAN
• Plato is a man (true or false)
• We can not draw any conclusions about
similarities between Socrates and Plato
– By separate propositional logic cannot
reach a conclusion
• Variable = Substituting a value
• Constant = Have to assign value
19. Predicate Logic
• More like a variable/can hold different values
• Socrates is a man (true or false)
– PREDICATE(VALUE)
• Socrates is a man
– MAN(SOCRATES)
• Plato is a man
– MAN(PLATO)
• Now the structure of representation reflects
the structure of knowledge itself
20. Predicate Logic
• Marcus is a man
– MAN(Marcus)
• Marcus is a Pompeian
– POMPEIAN (Marcus)
• All Pompeians were Romans
– Vx POMPEIAN(x) -> ROMAN(x)
21. Predicate Logic
• All Romans were either loyal to Caesar
or hated him
• Vx ROMAN(x) -> loyalto (x, Caesar) v hate (x, Caesar)
• It is difficult to represent knowledge in predicate logic
23. Semantic Networks (Nets)
• Semantic net is a knowledge presentation
method based on a network structure
• It consists of
– points called nodes connected by
– links called arcs
• Nodes – object, concepts, events
• Arcs - relationships between nodes
24. Semantic Nets
• Common arcs used for representing
hierarchies include isa and has-part
• Processing Natural Language
– Example: Text Mining
• Uses Natural Language for summarizing article
or newspaper
25. Example:
The Queen Mary is an ocean liner
Every ocean liner is a ship
Ship
isa
Ocean Liner
isa
Queen Mary
26. SHIP
Isa (hierarchical relationship)
Ocean Liner Oil Tanker Engine Hull
Swimming Queen Mary Liver Pool Boiler
Pool
Has-part (component relationship)
isa
27. Bill gives Judy a gift
Judy
Give
(verb)
Gift
Bill
Recieves
Object
Gives
Node
Node
28. Bill told Laura that he gave Judy a gift
Judy
Give
(verb)
Gift
Bill
Recieves
Object
Gives
Tell
Laura
Speaker
Listener
Time
Past
29. Frame 1
• Similar to Object
• Hierarchical Representation
– Introduce details as necessary
– Polymorphism
– Multi-inheritance
• A data structure for representing a stereotyped
situation
• A network of nodes and relations organized in a
hierarchy
• The topmost nodes - general concepts (abstract
class)
• The lower nodes - more specific instances (more
specific classes)
30. Frame 2
• The concepts at each node is described by a
set of attributes and values of those attributes
• Attributes are called slots
• Each slot can have procedures (codes)
• Typical procedures
– if added procedure
– if deleted procedure
– if needed procedure
31. Frame 3
• OOP
– Class
– Attribute
– Method
• AI
– Node
– Slots
– Procedures
32. Report
isa isa
Progress Report Technical Report
isa
DSS Project Process Report
33. A Node in a Frame System
Value 1
Slot 1
Slot 2 Value 2
Value 3
Slot 3
Procedure 1
Procedure 2
Procedure 3
34. Comparisons of KR Methods
• Rules
• When get too large become unmanageable
– IF… THEN… ELSE
– Advantage
• Simple syntax, easy to understand, simple
interpreter, high modularity, flexible
– Disadvantage
• Hard to follow hierarchies, inefficient for large
systems, not all knowledge can be expressed
as rules
35. Comparisons of KR Methods
• Semantic Nets
– Advantage
• Easy to follow hierarchy, easy to trace
association, flexible
– Disadvantage
• Meaning attached to nodes might be
ambiguous
• Exception handling is difficult
• Difficult to program
36. Comparisons of KR Methods
• Frames
– Advantage
• Expressive power, easy to set up slots for new
properties and relations
• Easy to create specialized procedures
• Easy to include default information and detect
missing values
– Disadvantage
• Difficult to program
• Difficult for inference