good (?) for use in specific contexts (set task, domain, problem - i.e., very context sensitive)

Decision trees:

• what they are
• how they are built
• how they relate to rules
• how they relate to use in Expert Systems

conventional programming technique
(Huntsinger - E.S. not AI anymore - once know how...question is, how is how?)

Knowledge Base Support - what need

Procedural Knowledge -Can only use in isolation if: (fundamental conditions)

1. diagnosis always starts with same info
2. D.T. constructed and no modification
3. no explanations required
4. all info always available on demand

Structural Knowledge (attribute/value - no structure, organization)

Empirical Knowledge (not first principles)

Meta-Knowledge (note: with these additions, no longer automatic generation)

Support for why all four types of knowledge are necessary

• knowledge acquisition
  • knowledge placement
  • knowledge content
• knowledge application (methods of control during run of d.t.)
  • problem focus
  • generating explanations
  • handling unknown information
• knowledge maintenance
  • replacing knowledge
  • deficient knowledge

Based on reliability? cost? timeliness of correction?

Would one need more than one? or can this info be put into the overall control strategy & organ.

(like design - sponsors for plans... like classification - structures for match)

Author indicates to stick with decision trees and add meta-knowledge to control use of structural and emperical knowledge.

Different techniques - depends on "starting point" See, there is AI in it if we add ............