3: Symbolic AI Knowledge Representation

1
3 Symbolic AI Knowledge Representation

• Outline
• Knowledge
• Representation schemes
• Structured object representations
• Semantic networks
• Frames
• Production systems
• Expert systems
• Learning outcomes
• Reading

2
Knowledge

• Knowledge vital to intelligence
• 2 types of knowledge
• Declarative
• Procedural
• Also
• Domain-specific
• Domain-general

3
Representing knowledge

• Need to get knowledge from (human?) source
• Represent it in a form usable by a machine
• Need to represent
• Facts
• Relationships between facts
• Property inheritance
• All men are mortal
• Socrates is a man
• Socrates is mortal
• Procedural knowledge

4
Knowledge representation schemes

• Need to be
• Expressive
• Different types of knowledge
• Different levels of knowledge
• Specific facts
• Generic information
• Clear
• Effective
• Infer new knowledge from old
• Efficient
• Explicit
• Explanation/justification

5
Semantic networks
Collins Quillian (1969)
isa
6
Semantic nets - reasoning

• Networkdata
• Reasoning traverse arcs to identify
  relationships
• Collins Quillian (1969)
• Sentence verification task
• Canaries sing 1310 ms
• Canaries have wings 1380 ms
• Canaries have skin 1470 ms

7
Semantic nets - problems

• Only simple ideas/relationships can be expressed
• quantification and intentional concepts are hard
  to represent in this formalism.
• Some birds fly
• All the birds sing some of the songs
• Some of the birds sing all the songs
• Mike thinks that Janes belief that Bernard will
  like their new home is false.

8
Frames

• More structured than semantic nets
• Nodes replaced by groups of information
• Frame slot and filler structure
• Slot property
• Filler value
• Can rep. Classes of entity or specific instances

9
Frames
NOVEL INSTANCE_OF BOOK PURPOSE
ENTERTAINMENT
THE_STAND INSTANCE_OF NOVEL AUTHOR
STEPHEN_KING NO_OF_PAGES 1009 YEAR_PUBLISHED
1978 ISBN_CODE 0-340-35895-5 STYLE HORROR
STEPHEN_KING INSTANCE_OF HUMAN SEX
MALE YEAR_BORN 1947
HORROR MEMBER_OF FICTION
10
Problem with multiple inheritance
NIXON INSTANCE_OF HUMAN SEX MALE MEMBER_OF
REPUBLICAN MEMBER_OF QUAKER
QUAKER PACIFIST YES
REPUBLICAN PACIFIST NO
11
Production systems
INFERENCE ENGINE

• RULE SET
• if ltconditiongt
• then ltactiongt
• if ltcondition1gt
• ltcondition2gt
• ltcondition3gt
• then ltaction1gt
• ltaction2gt

WORKING MEMORY Fact1 Fact2 Fact3
12
Production systems - control

• Global
• Forward-chaining (data-driven)
• Backward-chaining (goal-directed)
• Local (conflict resolution)
• Textual order
• Refractoriness
• Recency
• Specificity

13
Production systems pros and cons

• Natural and plausible
• Record of problem-solving
• Modular
• Sensitive to change
• -
• Knowledge items not related
• Can get unwieldy
• Inflexible syntax

14
(No Transcript)
15
Childrens subtraction errors

• Childrens errors are due to incorrect strategies
  rather than incorrect number facts
• Can model errors by adding/modifying/deleting
  rules

• 63
• 44
• 21

• 70
• 47
• 37

• 70
• 47
• 30

• 96
• 42
• 44

• 72
• 57
• 20

Young OShea (1981)
16
MYCIN an expert system

• Diagnosis and treatment of bacterial blood
  infections
• Consults with Doctor in English
• Collects info about patient
• Collects info about lab tests
• Reasons with set of production rules
• Can explain reasoning behind decisions
• Can acquire new knowledge with the aid of
  TEIRESIAS

17
Learning Outcomes

• Demonstrate an understanding of issues in
  knowledge representation
• Understand different knowledge representation
  schemes
• Show an awareness of some of the applications of
  systems constructed according to these schemes

18
Reading

• See list
• FOR NEXT WEEK, READ
• Searle, J. R. (1981). Minds, brains, and
  programs.
• Available in
• Study pack
• Boden, M.A. (1990). The philosophy of artificial
  intelligence. OUP.
• Haugeland, J. (1997). Mind design II. MIT Press.