KNOWLEDGE SYSTEMS

1
KNOWLEDGE SYSTEMS

• IS4185 Session 5
• Prof. Mark Nissen

2
Agenda

• GE CATS-1
• Knowledge System History
• Expertise
• Knowledge Systems
• Exercises

3
GE CATS-1

• David Smith, locomotive guru
• Retiring after 40 years
• GE very dependent on Smith
• Problems mentor-protégé approach?
• CATS-1?
• Development time?
• Installation usage?

4
Knowledge System History

• Knowledge vs. expert system?
• GPS - General Problem Solver
• Goals, states operators
• Distance reduction
• Failure - weak knowledge
• Expert systems - narrow domains
• Strong domain knowledge
• RB representation matching OK
• Much more practical

5
Expertise

• Task-specific knowledge, often tacit
• Long time to develop, incrementally
• Self-contained, often heuristic
• Expertise profile
• Expertsnon-experts - 1100
• Expertnovice ability - 301
• Expertaverage ability - 31
• ROI for expert system project?
• Expert attributes (p. 444)

6
Knowledge Systems

• Emulate performance of human experts
• Turing Test (dont ask, dont tell)
• Knowledge system roles
• Expert, consultant, advisor, instructor
• Colleague, partner, assistant, agent
• Use knowledge for inference (data-dr)
• Hard problems, unclear solution approach, complex
  space, unstructured, high ignorance
• DS7 - Intelligence Density

7
Knowledge Acquisition System Structure
DSS
Docs other sources
Decision Maker
DGMS
Knowledge Engineer
KB Infer Engine
Expert(s)
8
Examples

• Diagnosis - car wont start
• Auto source selection (p. 452)
• General categories (T 12.2)
• TARA - foreign currency traders
• Wines on disk
• MYCIN, XCON, etc.

9
Benefits

• DM time quality
• Leveraging knowledge expertise (KM)
• Hazardous duty remote locations
• Freeing experts for tough problems
• Integrating experts capabilities (D Blue)
• Training
• Reliability and institutional memory

10
Success Factors

• 80/20 rule
• Genuine business case
• Expert available/willing, articulate how
• Qualitative, complex problem
• Narrow problem scope
• Reasonable expectations
• Exceptional performance
• User acceptance, immediate value
• (Re)Use shells pearls, ML

11
Exercises

• Software source selection
• Multi-attribute decision making - LDW?
• Rule-based inference - VPExpert?
• Integrate LDW VPExpert?
• Restaurant source selection
• Person (volunteer)
• restaurantrow (http//www.restaurantrow.com/)
• sfgate (http//www.sfgate.com/eguide/food)
• How intelligent is behavior?