1. Computer Science Department California Polytechnic State University San Luis Obispo, CA, U.S.A. Franz J. Kurfess Knowledge Processing

2. Some of the material in these slides was developed for a lecture series sponsored by the European Community under the BPD program with Vilnius University as host institution Acknowledgements

3. 3 Franz Kurfess: Knowledge Processing Use and Distribution of these Slides ❖ These slides are primarily intended for the students in classes I teach. In some cases, I only make PDF versions publicly available. If you would like to get a copy of the originals (Apple KeyNote or Microsoft PowerPoint), please contact me via email at fkurfess@calpoly.edu. I hereby grant permission to use them in educational settings. If you do so, it would be nice to send me an email about it. If you’re considering using them in a commercial environment, please contact me first.fkurfess@calpoly.edu 3

4. 4 Franz Kurfess: Knowledge Processing Overview Knowledge Processing 4 ❖ Motivation ❖ Objectives ❖ Chapter Introduction ❖ Knowledge Processing as Core AI Paradigm ❖ Relationship to KM ❖ Terminology ❖ Knowledge Acquisition ❖ Knowledge Elicitation ❖ Machine Learning ❖ Knowledge Representation ❖ Logic ❖ Rules ❖ Semantic Networks ❖ Frames, Scripts ❖ Knowledge Manipulation ❖ Reasoning ❖ KQML ❖ Important Concepts and Terms ❖ Chapter Summary

5. 5 Franz Kurfess: Knowledge Processing Bridge-In ❖ Getting to the airport in time to catch a plane 5

6. 6 Franz Kurfess: Knowledge Processing Catching My Plane ❖ Scenario for an examination of knowledge representation and reasoning: ❖ a traveller wants to know when she needs to leave her hotel in order to catch a plane ❖ background knowledge ❖ situation knowledge ❖ acquisition of additional knowledge for decision making ❖ reasoning methods ❖ verification and validation 6

7. 7 Franz Kurfess: Knowledge Processing Computer Scenario ❖ traveler posts a query to a computer 7

8. 8 Franz Kurfess: Knowledge Processing Google: F 8

9. 9 Franz Kurfess: Knowledge Processing Bing: F 9

10. 10 Franz Kurfess: Knowledge Processing Wolfram Alpha: F 10

11. 11 Franz Kurfess: Knowledge Processing Ask.com: F 11

12. 12 Franz Kurfess: Knowledge Processing Human Scenario ❖ traveler asks a human ❖ e.g. hotel receptionist 12

13. 13 Franz Kurfess: Knowledge Processing Pre-Test 13

14. 14 Franz Kurfess: Knowledge Processing Motivation ❖ the representation and manipulation of knowledge has been essential for the development of humanity as we know it ❖ the use of formal methods and support from machines can improve our knowledge representation and reasoning abilities ❖ intelligent reasoning is a very complex phenomenon, and may have to be described in a variety of ways ❖ a basic understanding of knowledge representation and reasoning is important for the organization and management of knowledge 14

15. 15 Franz Kurfess: Knowledge Processing Objectives ❖ be familiar with the commonly used knowledge representation and reasoning methods ❖ understand different roles and perspectives of knowledge representation and reasoning methods ❖ examine the suitability of knowledge representations for specific tasks ❖ evaluate the representation methods and reasoning mechanisms employed in computer- based systems 15

16. 16 Franz Kurfess: Knowledge Processing Chapter Introduction ❖ Knowledge Processing as Core AI Paradigm ❖ Knowledge Processing in Humans ❖ Relationship to KM ❖ Terminology 16

17. 17 Franz Kurfess: Knowledge Processing Relationship to KM 17 KP/AIHumans/KM representation methods suited for KP by computers representation of knowledge in formats suitable for humans reasoning performed by computers essential reasoning performed by humans mostly limited to symbol manipulation support from computers very demanding in terms of computational power emphasis often on documents can be used for “grounded” systems larger granularity interpretation (“meaning”) typically left to humans mainly intended for human use

18. 18 Franz Kurfess: Knowledge Processing Knowledge Processes Chaotic knowledge processes Human knowledge and networking Information databases and technical networking Systematic information and knowledge processes [Skyrme 1998] 18

19. 19 Franz Kurfess: Knowledge Processing Knowledge Cycles Create Product/ Process Knowledge Repository Codify Embed Diffuse Identify Classify Access Use/Exploit Collect Organize/ Store Share/ Disseminate [Skyrme 1998] 19

20. 20 Franz Kurfess: Knowledge Processing Knowledge Representation ❖ Types of Knowledge ❖ Factual, subjective, heuristic, deep, shallow ❖ Knowledge Representation Methods ❖ Rules, Frames, Semantic Networks ❖ Blackboard Representations ❖ Object-based Representations ❖ Case-Based Reasoning ❖ Knowledge Representation Tools ❖ Expert System Shells ❖ Ontology development tools ❖ Repositories 20

21. 21 Franz Kurfess: Knowledge Processing Types of Knowledge The field that investigates knowledge types and similar questions is epistemology ❖ Factual Knowledge ❖ Subjective Knowledge ❖ Heuristic Knowledge ❖ Deep and Shallow Knowledge ❖ Other Types of Knowledge 21

22. 22 Franz Kurfess: Knowledge Processing Factual Knowledge ❖ verifiable ❖ through experiments, formal methods, sometimes commonsense reasoning ❖ often created by authoritative sources ❖ typically not under dispute in the domain community ❖ often incorporated into reference works, textbooks, domain standards 22

23. 23 Franz Kurfess: Knowledge Processing Subjective Knowledge ❖ relies on individuals ❖ insight, experience ❖ possibly subject to interpretation ❖ more difficult to verify ❖ especially if the individuals possessing the knowledge are not cooperative ❖ different from belief ❖ both are subjective, but beliefs are not verifiable 23

24. 24 Franz Kurfess: Knowledge Processing Heuristic Knowledge ❖ based on rules or guidelines that frequently help solving problems ❖ often derived from practical experience working in a domain ❖ as opposed to theoretical insights gained from deep thoughts about a topic ❖ verifiable through experiments 24

25. 25 Franz Kurfess: Knowledge Processing Deep and Shallow Knowledge ❖ deep knowledge enables explanations and plausibility considerations ❖ possibly including formal proofs ❖ shallow knowledge may be sufficient to answer immediate questions, but not for explanations ❖ heuristics are often an example of shallow knowledge 25

26. 26 Franz Kurfess: Knowledge Processing Other Types of Knowledge ❖ procedural knowledge ❖ knowing how to do something ❖ declarative knowledge ❖ expressed through statements that can be shown to be true or false ❖ prototypical example is mathematical logic ❖ tacit knowledge ❖ implicit, unconscious knowledge that can be difficult to express in words or other representations ❖ a priori knowledge ❖ independent on experience or empirical evidence ❖ e.g. “everybody born before 1983 is older than 20 years” ❖ a posteriori knowledge ❖ dependent of experience or empirical evidenceevidence ❖ e.g. “X was born in 1983” 26

27. 27 Franz Kurfess: Knowledge Processing Roles of Knowledge Representation (KR) ❖ KR as Surrogate ❖ Ontological Commitments ❖ Fragmentary Theory of Intelligent Reasoning ❖ Medium for Computation ❖ Medium for Human Expression [Davis, Shrobe, Szolovits, 1993] 27

28. 28 Franz Kurfess: Knowledge Processing KR as Surrogate ❖ a substitute for the thing itself ❖ enables an entity to determine consequences by thinking rather than acting ❖ reasoning about the world through operations on the representation ❖ reasoning or thinking are inherently internal processes ❖ the objects of reasoning are mostly external entities (“things”) ❖ some objects of reasoning are internal, e.g. concepts, feelings,... [Davis, Shrobe, Szolovits, 1993] 28

29. 29 Franz Kurfess: Knowledge Processing Surrogate Aspects ❖ Identity ❖ correspondence between the surrogate and the intended referent in the real world ❖ Fidelity ❖ Incompleteness ❖ Incorrectness ❖ Adequacy ❖ Task ❖ User [Davis, Shrobe, Szolovits, 1993] 29

30. 30 Franz Kurfess: Knowledge Processing Surrogate Consequences ❖ perfect representation is impossible ❖ the only completely accurate representation of an object is the object itself ❖ incorrect reasoning is inevitable ❖ if there are some flaws in the world model, even a perfectly sound reasoning mechanism will come to incorrect conclusions [Davis, Shrobe, Szolovits, 1993] 30

31. 31 Franz Kurfess: Knowledge Processing Ontological Commitments ❖ terms (formalisms, methods, constructs) used to represent the world ❖ by selecting a representation a decision is made about how and what to see in the world ❖ like a set of glasses that offer a sharp focus on part of the world, at the expense of blurring other parts ❖ necessary because of the inevitable imperfections of representations ❖ useful to concentrate on relevant aspects ❖ pragmatic because of feasibility constraints [Davis, Shrobe, Szolovits, 1993] 31

32. 32 Franz Kurfess: Knowledge Processing Ontological Commitments Examples ❖ logic ❖ views the world in terms of individual entities and relationships between the entities ❖ enforces the assignment of truth values to statements ❖ rules ❖ entities and their relationships expressed through rules ❖ frames ❖ prototypical objects ❖ semantic nets ❖ entities and relationships displayed as a graph [Davis, Shrobe, Szolovits, 1993] 32

33. 33 Franz Kurfess: Knowledge Processing KR and Reasoning ❖ a knowledge representation indicates an initial conception of intelligent inference ❖ often reasoning methods are associated with representation technique ❖ first order predicate logic and deduction ❖ rules and modus ponens ❖ the association is often implicit ❖ the underlying inference theory is fragmentary ❖ the representation covers only parts of the association ❖ intelligent reasoning is a complex and multi-faceted phenomenon [Davis, Shrobe, Szolovits, 1993] 33

34. 34 Franz Kurfess: Knowledge Processing KR for Reasoning ❖ a representation suggests answers to fundamental questions concerning reasoning: ❖ What does it mean to reason intelligently? ❖ implied reasoning method ❖ What can possibly be inferred from what we know? ❖ possible conclusions ❖ What should be inferred from what we know? ❖ recommended conclusions [Davis, Shrobe, Szolovits, 1993] 34

35. 35 Franz Kurfess: Knowledge Processing KR and Computation ❖ from the AI perspective, reasoning is a computational process ❖ machines are used as reasoning tools ❖ without efficient ways of implementing such computational process, it is practically useless ❖ e.g. Turing machine ❖ most representation and reasoning mechanisms are modified for efficient computation ❖ e.g. Prolog vs. predicate logic [Davis, Shrobe, Szolovits, 1993] 35

36. 36 Franz Kurfess: Knowledge Processing Computational Medium ❖ computational environment for the reasoning process ❖ reasonably efficient ❖ organization and representation of knowledge so that reasoning is facilitated ❖ may come at the expense of understandability by humans ❖ unexpected outcomes of the reasoning process ❖ lack of transparency of the reasoning process ❖ even though the outcome “makes sense,” it is unclear how it was achieved 36

37. 37 Franz Kurfess: Knowledge Processing KR for Human Expression ❖ a knowledge representation or expression method that can be used by humans to make statements about the world ❖ expression of knowledge ❖ expressiveness, generality, preciseness ❖ communication of knowledge ❖ among humans ❖ between humans and machines ❖ among machines ❖ typically based on natural language ❖ often at the expense of efficient computability [Davis, Shrobe, Szolovits, 1993] 37

38. 38 Franz Kurfess: Knowledge Processing Knowledge Acquisition ❖ Incorporating Knowledge into a Repository ❖ human mind ❖ human-readable ❖ book, magazine, etc ❖ computer-based ❖ Knowledge Acquisition Types ❖ Knowledge Elicitation ❖ conversion of human knowledge into a format suitable for computers ❖ Machine Learning ❖ extraction of knowledge from data 38

39. 39 Franz Kurfess: Knowledge Processing Acquisition of Knowledge ❖ Published Sources ❖ Physical Media ❖ Digital Media ❖ People as Sources ❖ Interviews ❖ Questionnaires ❖ Formal Techniques ❖ Observation Techniques ❖ Knowledge Acquisition Tools ❖ automatic ❖ interactive 39

40. 40 Franz Kurfess: Knowledge Processing Knowledge Elicitation ❖ knowledge is already present in humans, but needs to be converted into a form suitable for computer use ❖ requires the collaboration between a domain expert and a knowledge engineer ❖ domain expert has the domain knowledge, but not necessarily the skills to convert it into computer-usable form ❖ knowledge engineer assists with this conversion ❖ this can be a very lengthy, cumbersome and error- prone process 40

41. 41 Franz Kurfess: Knowledge Processing Machine Learning ❖ extraction of higher-level information from raw data ❖ based on statistical methods ❖ results are not necessarily in a format that is easy for humans to use ❖ the organization of the gained knowledge is often far from intuitive for humans ❖ examples ❖ decision trees ❖ rule extraction from neural networks 41

42. 42 Franz Kurfess: Knowledge Processing Knowledge Fusion ❖ integration of human-generated and machine- generated knowledge ❖ sometimes also used to indicate the integration of knowledge from different sources, or in different formats ❖ can be both conceptually and technically very difficult ❖ different “spirit” of the knowledge representation used ❖ different terminology ❖ different categorization criteria ❖ different representation and processing mechanisms ❖ e.g. graph-oriented vs. rules vs. data base-oriented 42

43. 43 Franz Kurfess: Knowledge Processing Knowledge Representation Mechanisms ❖ Logic ❖ Rules ❖ Semantic Networks ❖ Frames, Scripts 43

44. 44 Franz Kurfess: Knowledge Processing Logic ❖ syntax: well-formed formula ❖ a formula or sentence often expresses a fact or a statement ❖ semantics: interpretation of the formula ❖ “meaning” is associated with formulae ❖ often compositional semantics ❖ axioms as basic assumptions ❖ generally accepted within the domain ❖ inference rules for deriving new formulae from existing ones 44

45. 45 Franz Kurfess: Knowledge Processing KR Roles and Logic ❖ surrogate ❖ very expressive, not very suitable for many types of knowledge ❖ ontological commitments ❖ objects, relationships, terms, logic operators ❖ fragmentary theory of intelligent reasoning ❖ deduction, other logical calculi ❖ medium for computation ❖ yes, but not very efficient ❖ medium for human expression ❖ only for experts 45

46. 46 Franz Kurfess: Knowledge Processing Rules ❖ syntax: if … then … ❖ semantics: interpretation of rules ❖ usually reasonably understandable ❖ also known as production rules ❖ allows the generation of new symbol strings from existing ones ❖ initial rules and facts ❖ often capture basic assumptions and provide initial conditions ❖ generation of new facts, application to existing rules ❖ forward reasoning: starting from known facts ❖ backward reasoning: starting from a hypothesis 46

47. 47 Franz Kurfess: Knowledge Processing KR Roles and Rules ❖ surrogate ❖ reasonably expressive, suitable for some types of knowledge ❖ ontological commitments ❖ objects, rules, facts ❖ fragmentary theory of intelligent reasoning ❖ modus ponens, matching, sometimes augmented by probabilistic mechanisms ❖ medium for computation ❖ reasonably efficient ❖ medium for human expression ❖ mainly for experts 47

48. 48 Franz Kurfess: Knowledge Processing Semantic Networks ❖ syntax: graphs, possibly with some restrictions and enhancements ❖ semantics: interpretation of the graphs ❖ initial state of the graph ❖ propagation of activity, inferences based on link types 48

49. 49 Franz Kurfess: Knowledge Processing KR Roles and Semantic Nets ❖ surrogate ❖ limited to reasonably expressiveness, suitable for some types of knowledge ❖ ontological commitments ❖ nodes (objects, concepts), links (relations) ❖ fragmentary theory of intelligent reasoning ❖ conclusions based on properties of objects and their relationships with other objects ❖ medium for computation ❖ reasonably efficient for some types of reasoning ❖ medium for human expression ❖ easy to visualize 49

50. 50 Franz Kurfess: Knowledge Processing Frames, Scripts ❖ syntax: templates with slots and fillers ❖ semantics: interpretation of the slots/filler values ❖ initial values for slots in frames ❖ complex matching of related frames 50

51. 51 Franz Kurfess: Knowledge Processing KR Roles and Frames ❖ surrogate ❖ suitable for well-structured knowledge ❖ ontological commitments ❖ templates, situations, properties, methods ❖ fragmentary theory of intelligent reasoning ❖ conclusions are based on relationships between frames ❖ medium for computation ❖ ok for some problem types ❖ medium for human expression ❖ ok, but sometimes too formulaic 51

52. 52 Franz Kurfess: Knowledge Processing Knowledge Manipulation ❖ Reasoning ❖ KQML 52

53. 53 Franz Kurfess: Knowledge Processing Reasoning ❖ generation of new knowledge items from existing ones ❖ frequently identified with logical reasoning ❖ strong formal foundation ❖ very restricted methods for generating conclusions ❖ sometimes expanded to capture various ways to draw conclusions based on methods employed by humans ❖ requires a formal specification or implementation to be used with computers 53

54. 54 Franz Kurfess: Knowledge Processing KQML ❖ stands for Knowledge Query and Manipulation Language ❖ language and protocol for exchanging information and knowledge 54

55. 55 Franz Kurfess: Knowledge Processing KQML Performatives ❖ basic query performatives ❖ evaluate, ask-if, ask-about, ask-one, ask-all ❖ multi-response query performatives ❖ stream-about, stream-all ❖ response performatives ❖ reply, sorry ❖ generic informational performatives ❖ tell, achieve, deny, untell, unachieve ❖ generator performatives ❖ standby, ready, next, rest, discard, generator ❖ capability-definition performatives ❖ advertise, subscribe, monitor, import, export ❖ networking performatives ❖ register, unregister, forward, broadcast, route. 55

56. 56 Franz Kurfess: Knowledge Processing KQML Example 1 ❖ query (ask-if :sender A :receiver B :language Prolog :ontology foo :reply-with id1 :content ``bar(a,b)'' ) ❖ reply (sorry :sender B :receiver A :in-reply-to id1 :reply-with id2 ) agent A (:sender) is querying the agent B (:receiver), in Prolog (:language) about the truth status of ``bar(a,b)'' (:content) 56

57. 57 Franz Kurfess: Knowledge Processing KQML Example 2 ❖ query (stream-about :language KIF :ontology motors `:reply-with q1 :content motor1) ❖ reply (tell :language KIF :ontology motors :in-reply-to q1 : content (= (val (torque motor1) (sim- time 5) (scalar 12 kgf)) (tell :language KIF :ontology structures :in-reply-to q1 : content (fastens frame12 motor1)) (eos :in-repl-to q1) agent A asks agent B to tell all it knows about motor1. B replys with a sequence of tells terminated with a sorry. 57

58. 58 Franz Kurfess: Knowledge Processing Post-Test 58

59. 59 Franz Kurfess: Knowledge Processing Evaluation ❖ Criteria 59

60. 60 Franz Kurfess: Knowledge Processing KP/KM Activity ❖ select a domain that requires significant human involvement for dealing with knowledge ❖ identify at least two candidates for ❖ knowledge representation ❖ reasoning ❖ evaluate their suitability ❖ human perspective ❖ understandable and usable for humans ❖ computational perspective ❖ storage, processing 60

61. 61 Franz Kurfess: Knowledge Processing KP/KM Activity Outcomes 2007 ❖ Images with Metadata ❖ Extracting contact information from text ❖ Qualitative and quantitative knowledge about cheese making ❖ Visualization of astronomy data ❖ Surveillance/security KM ❖ Marketing ❖ Face recognition ❖ Visual marketing 61

62. 62 Franz Kurfess: Knowledge Processing Important Concepts and Terms 62 ❖ automated reasoning ❖ belief network ❖ cognitive science ❖ computer science ❖ deduction ❖ frame ❖ human problem solving ❖ inference ❖ intelligence ❖ knowledge acquisition ❖ knowledge representation ❖ linguistics ❖ logic ❖ machine learning ❖ natural language ❖ ontology ❖ ontological commitment ❖ predicate logic ❖ probabilistic reasoning ❖ propositional logic ❖ psychology ❖ rational agent ❖ rationality ❖ reasoning ❖ rule-based system ❖ semantic network ❖ surrogate ❖ taxonomy ❖ Turing machine

63. 63 Franz Kurfess: Knowledge Processing Summary Knowledge Processing ❖ there are different types of knowledge ❖ knowledge acquisition can be conceptually difficult and time-consuming ❖ popular knowledge representation methods for computers are based on mathematical logic, if... then rules, and graphs ❖ computer-based reasoning depends on the knowledge representation method, and can be computationally very challenging 63

64. 64 Franz Kurfess: Knowledge Processing 64