1. Knowledge n Knowledge is a collection of specialized facts, procedures and judgment rules High Low Degree of Abstraction Quantity Knowledge Information Data

2. Knowledge Sources n Documented (books, manuals, etc.) n Undocumented (in people's minds) – From people, from machines n Knowledge Acquisition from Databases n Knowledge Acquisition Via the Internet

3. Knowledge Levels n Shallow knowledge (surface) n Deep knowledge n Can implement a computerized representation that is deeper than shallow knowledge n Special knowledge representation methods (semantic networks and frames) to allow the implementation of deeper-level reasoning (abstraction and analogy): important expert activity n Represent objects and processes of the domain of expertise at this level n Relationships among objects are important

4. Major Categories of Knowledge n Declarative Knowledge n Procedural Knowledge n Metaknowledge

5. Declarative Knowledge Descriptive Representation of Knowledge n Expressed in a factual statement n Shallow n Important in the initial stage of knowledge acquisition

6. Procedural Knowledge n Considers the manner in which things work under different sets of circumstances – Includes step-by-step sequences and how-to types of instructions – May also include explanations – Involves automatic response to stimuli – May tell how to use declarative knowledge and how to make inferences

7. n Descriptive knowledge relates to a specific object. Includes information about the meaning, roles, environment, resources, activities, associations and outcomes of the object n Procedural knowledge relates to the procedures employed in the problem- solving process

8. Metaknowledge Knowledge about Knowledge In ES, Metaknowledge refers to knowledge about the operation of knowledge-based systems Its reasoning capabilities

9. Knowledge Modeling n The knowledge model views knowledge acquisition as the construction of a model of problem- solving behavior-- a model in terms of knowledge instead of representations n Can reuse models across applications

10. Knowledge Representation n Logical representation – first order predicate calculus, Prolog, declarative knowledge n Procedural representation – a set of instructions for solving a problem, such as a production system n Network representation – knowledge is in a graph structure, such as conceptual dependency and conceptual graphs we will study in this chapter n Structured representation – an extension of networks, such as scripts or frames we will study in this chapter

11. Group Work n What type of knowledge representation would be appropriate to contain a rules to generate the following sequences

12. Group Work n What type of knowledge representation would be appropriate to solve “ analogy ” problems?

13. Knowledge Representation Once acquired, knowledge must be organized for use

14. Introduction n A good knowledge representation naturally represents the problem domain n An unintelligible knowledge representation is wrong n Most artificial intelligence systems consist of: – Knowledge Base – Inference Mechanism (Engine)

15. n Knowledge Base – Forms the system's intelligence source – Inference mechanism uses to reason and draw conclusions n Inference mechanism: Examines the knowledge base to answer questions, solve problems or make decisions within the domain

16. n Many knowledge representation schemes – Can be programmed and stored in memory – Are designed for use in reasoning n Major knowledge representation schemas: – Production rules – Frames

17. Representation in Logic and Other Schemas n General form of any logical process n Inputs (Premises) n Premises used by the logical process to create the output, consisting of conclusions (inferences) n Facts known true can be used to derive new facts that are true

18. n Symbolic logic : System of rules and procedures that permits the drawing of inferences from various premises n Basic Forms of Computational Logic – Propositional logic (or propositional calculus) – Predicate logic (or predicate calculus)

19. Propositional Logic n A proposition is a statement that is either true or false n Once known, it becomes a premise that can be used to derive new propositions or inferences n Rules are used to determine the truth (T) or falsity (F) of the new proposition

20. n Symbols represent propositions, premises or conclusions Statement: A = The mail carrier comes Monday through Friday. Statement: B = Today is Sunday. Conclusion: C = The mail carrier will not come today. n Propositional logic: limited in representing real-world knowledge

21. Predicate Calculus n Predicate logic breaks a statement down into component parts, an object, object characteristic or some object assertion n Predicate calculus uses variables and functions of variables in a symbolic logic statement n Predicate calculus is the basis for Prolog (PROgramming in LOGic) n Prolog Statement Examples – comes_on(mail_carrier, monday). – likes(jay, chocolate). (Note - the period “. ” is part of the statement)

22. Lists Written Series of Related Items n Normally used to represent hierarchical knowledge where objects are grouped, categorized or graded according to – Rank or – Relationship

23. Decision Tables (Induction Table) Knowledge Organized in a Spreadsheet Format n Attribute List n Conclusion List n Different attribute configurations are matched against the conclusion

24. Decision Trees n Related to tables n Similar to decision trees in decision theory n Can simplify the knowledge acquisition process n Knowledge diagramming - very natural

25. O-A-V Triplet n Objects, Attributes and Values n O-A-V Triplet n Objects may be physical or conceptual n Attributes are the characteristics of the objects n Values are specific measures of the attributes

27. Default Logic n Deals with uncertainties n Incomplete information

28. Knowledge Maps n Visual representation n Cognitive maps

29. Semantic Networks n Semantics nets were introduced by Quilian in the late 1960s for representing knowledge as a network of associations By following links, simple questions can be answered Studies with human recall supported this model

30. Semantic Networks n Graphic Depiction of Knowledge n Nodes and Links Showing Hierarchical Relationships Between Objects n Nodes: Objects n Arcs: Relationships – is-a – has-a

31. n Semantic networks can show inheritance n Semantic Nets - visual representation of relationships n Can be combined with other representation methods

32. Semantic Network Example Joe Boy Kay Woman Food Human Being School Has a child Needs Goes to Is a

33. Conceptual Graphs n Graph Structure –Finite, connected, bipartite –Arcs are not labeled –Conceptual relation nodes are introduced between concepts –The bipartite nature of the graph means concepts can only link to conceptual relations and vice versa –In drawings, concepts are shown in boxes and conceptual relations in ellipses n Concepts may be concrete (dog, child, etc.) or abstract (love, beauty, etc.)

34. Arity of Relations n Examples of 1-ary, 2-ary, and 3-ary relations

35. Graph of a Sentence n “ Mary gave John the book ” –As in conceptual dependency, the verb plays a central role in the structure –The verb “ give ” in this sentence has an agent, an object, and a recipient

36. Group Work n What does the following conceptual graph represent

37. Types and Individuals n In the first case, the type is dog and the individual is “ emma ” n A specific but unnamed dog is given a unique number (#) n An alternative representation is to use a dog specified by a # and add a conceptual relation for a name

38. Three Names n “ Her name was McGill and she called herself Lil, but everyone knew her as Nancy ” (song lyric) Who was the artist? What was the name of the song?

39. Itchy Dog n If the same, unspecified individual is present in two or more nodes, a variable can be introduced that may eventually be bound to the same value What is the English sentence for this structure?

40. Type Lattice n Concepts often form a lattice of types, such as a class golden retriever a type of dog, a type of carnivore, a type of animal, and so forth n  is a supertype of all types,  is the absent type n Answering queries about a pair of concepts may involve finding the minimum common supertype

41. Generalization and Specialization n A concept node can be replaced with a restriction

42. Join of Concepts n If two graphs contain an identical node, they can be joined together by having only one copy of the identical node n Join is a form of restriction since the resultant graph is more specific than the original graphs

43. Simplification n A join may result in duplicate information n The simply operation allows the removal of duplication information

44. Inheritance n Inheritance is a form of generalization n Generalization does not guarantee that the resultant graph is true even if the original graphs are true

45. Propositional Nodes n “ Tom believes that Jane likes pizza ” The verb believes takes a propositional node as its object

46. “There are no pink dogs” n In some cases a propositional node may stand alone, as seen here: This is similar to modal logics that introduce a level of believability, such as necessary, probably, possible, or other levels, such as negative shown here

47. Group Work n What does the following conceptual graph represent

48. Conceptual Graphs and Logic n Conceptual graphs are equivalent to predicate calculus in expressive power Here is an algorithm to change a conceptual graph into a predicate calculus expression

49. Production Rules n Condition-Action Pairs – IF this condition (or premise or antecedent) occurs, – THEN some action (or result, or conclusion, or consequence) will (or should) occur – IF the stop light is red AND you have stopped, THEN a right turn is OK

50. n Each production rule in a knowledge base represents an autonomous chunk of expertise n When combined and fed to the inference engine, the set of rules behaves synergistically n Rules can be viewed as a simulation of the cognitive behavior of human experts n Rules represent a model of actual human behavior

51. Forms of Rules n IF premise, THEN conclusion – IF your income is high, THEN your chance of being audited by the IRS is high n Conclusion, IF premise – Your chance of being audited is high, IF your income is high

52. n Inclusion of ELSE – IF your income is high, OR your deductions are unusual, THEN your chance of being audited by the IRS is high, OR ELSE your chance of being audited is low n More Complex Rules – IF credit rating is high AND salary is more than $30,000, OR assets are more than $75,000, AND pay history is not "poor," THEN approve a loan up to $10,000, and list the loan in category "B. ” – Action part may have more information: THEN "approve the loan" and "refer to an agent"

53. Knowledge and Inference Rules Common Types of Rules n Knowledge rules, or declarative rules, state all the facts and relationships about a problem n Inference rules, or procedural rules, advise on how to solve a problem, given that certain facts are known n Inference rules contain rules about rules (metarules) n Knowledge rules are stored in the knowledge base n Inference rules become part of the inference engine

54. Advantages of Rules n Easy to understand (natural form of knowledge) n Easy to derive inference and explanations n Easy to modify and maintain n Easy to combine with uncertainty n Rules are frequently independent

55. n Complex knowledge requires many rules n Builders like rules (hammer syndrome) n Search limitations in systems with many rules Limitations of Rules

57. Frames Definitions and Overview n Frame: Data structure that includes all the knowledge about a particular object n Knowledge organized in a hierarchy for diagnosis of knowledge independence n Form of object-oriented programming for AI and ES. n Each Frame Describes One Object n Special Terminology

58. Frames (2) n Frames, like scripts, are used in stereotypical situations –When a new situation is encountered, a frame may be recalled from memory –The frame provides a complete framework –Details may vary from situation to situation –Frames can provide default values –Frames can be arranged in a hierarchy

59. Frames and NLP n Much of the inference required for NLP involves making assumptions about what is typically true about a situation n Encode this stereotypical information in a frame n Looks like themes, but on a higher level of abstraction

61. Components of a Frame

62. Frame for a Hotel Room

64. Inheritance - 1 n A hierarchy for birds

65. Inheritance - 2 n Multiple inheritance for “ Opus ”

66. Inheritance - 3 n A new class to resolve ambiguity

67. Transitivity of Subclasses n Fixing one problem –Penguins don ’ t fly –Introduce a flightless bird class n Results in other problems –If subclasses are transitive, we infer a penguin is a bird –This adds an extra link that introduces problems with multiple inheritance Flightless bird is introduced to handle an exception

68. A Summary of Frames n Frames organize knowledge into structures n Frames are recalled on an as needed basis n Procedures can be attached to frames where the procedure may process one of the slots in the frame in some way, such as detecting changes n Frames support class inheritance n Frames can supply default knowledge n In essence, frames extended semantic networks by providing organization and structure

69.  A means of identifying common situations in a particular domain  A means of generating expectations – We precompile information, rather than recomputing from first principles.  Elements include  Entry Conditions: These must be satisfied before events in the script can occur  Props: Slots representing objects involved in events  Roles: Persons involved in the events  Tracks: Variations on the script. Different tracks may share components of the same script.  Scenes: The sequence of events that occur. Events are represented in conceptual dependency form. Scripts

70. Travel by plane: – Roles: Actor, Clerk, Source, Destination, Airport, Ticket, Money, Airplane – Constraints: Person(Actor), Value(Money, Price(Ticket)),... – Preconditions: Owns(Actor, Money), At(Actor, Source) – Effects: not(Owns(Actor, Money)), not(At(Actor, Source)), At(Actor, Destination) – Decomposition: GoTo(Actor, Airport) BuyTicket(Actor, Clerk, Money, Ticket),... Scripts

71. The Restaurant Script

72. The Robbery Script

73. Advantages and Disadvantages Advantages of Scripts: n Ability to predict events. n A single coherent interpretation may be build up from a collection of observations. Disadvantages: n Less general than frames. n May not be suitable to represent all kinds of knowledge.

74. Group Work n Write a “ shopping ” script –What are the major activities –Once you have listed the major activities, a subgroup will be asked to detail the steps in each activity

75. Basic Conceptual Dependency Relations

76. Additional Notation

77. The Primitive Actions

78. Two Example Sentences

79. Group Work n Diagram the sentence: John took a plane to New York. n Diagram the sentence: John wondered who ate the cheese.

80. Some More Examples

81. Issues in Knowledge Representation n We have examined several ways to represent knowledge –Predicate calculus –Procedural, as in an expert system –Network, as in semantic nets, conceptual dependency and conceptual graphs –Structured, as in frames and scripts n Particular problems arise with each type, we examine problems with more recent types –Hierarchies and inheritance –Exceptions

82. Considerations for Evaluating a Knowledge Representation n Naturalness, uniformity and understandability n Degree to which knowledge is explicit (declarative) or embedded in procedural code n Modularity and flexibility of the knowledge base n Efficiency of knowledge retrieval and the heuristic power of the inference procedure

83. n No single knowledge representation method is ideally suited by itself for all tasks n Multiple knowledge representations: each tailored to a different subtask n Production Rules and Frames works well in practice n Object-Oriented Knowledge Representations – Hypermedia

84. Multiple Knowledge Representations n Rules + Frames n Others Knowledge Representation Must Support n Acquiring knowledge n Retrieving knowledge n Reasoning

85. Experimental Knowledge Representations n Cyc n NKRL n Spec-Charts Language

86. The Cyc System n Attempt to represent a substantial amount of common sense knowledge n Bold assumptions: intelligence needs a large amount of knowledge n Need a large knowledge base n Cyc over time is developing as a repository of a consensus reality - the background knowledge possessed by a typical U.S. resident n There are some commercial applications based on portions of Cyc

87. NKRL n Narrative Knowledge Representational Language (NKRL) n Standard, language-independent description of the content of narrative textual documents n Can translate natural language expressions directly into a meaningful set of templates that represent the knowledge

88. Knowledge Interchange Format (KIF) To Share Knowledge and Interact

89. The Spec-Charts Language n Based on Conceptual Graphs: to Define Objects and Relationships n Restricted Form of Semantic Networks n Evolved into the Commercial Product - STATEMATE

90. Knowledge Representation and the Internet n Hypermedia documents to encode knowledge directly n Hyperlinks Represent Relationships n MIKE (Model-based and Incremental Knowledge Engineering n Formal model of expertise: KARL Specification Language n Semantic networks: Ideally suited for hypermedia representation n Web-based Distributed Expert System (Ex-W-Pert System) for sharing knowledge-based systems and groupware development

91. Knowledge Representation and NLP n Quilian, an early researcher in semantic nets, suggested they could be the basis of natural language understanding Graphs proved to be too general and additional standardization of structure had to be introduced, as in conceptual dependency and conceptual graphs

92. Representing Uncertainty: An Overview Dealing with Degrees of Truth, Degrees of Falseness in ES n Uncertainty – When a user cannot provide a definite answer – Imprecise knowledge – Incomplete information

93. Several Approaches Related to Mathematical and Statistical Theories n Bayesian Statistics n Dempster and Shafer's Belief Functions n Fuzzy Sets Uncertainty

94. Uncertainty in AI Approximate Reasoning, Inexact Reasoning

95. Relevant Information is Deficient in One or More n Information is partial n Information is not fully reliable n Representation language is inherently imprecise n Information comes from multiple sources and it is conflicting n Information is approximate n Non-absolute cause-effect relationships exist n Can include probability in the rules n IF the interest rate is increasing, THEN the price of stocks will decline (80% probability)