PRODIGY: Knowledge represented as Control Rules

Description

• Represented as first-order logic condition and a control action to be taken when condition is met
• Condition expressed in terms of meta-predicates such a CURRENT-NODE or CANDIDATE-NODE
• Control action indicates whether to SELECT, REJECT, or PREFER a candidate or set of candidates
• Used to select:
  • which node in search tree to expand
  • which goal to work on
  • which operator to apply
  • which objects to use
• Increase efficiency of search, improve solution quality, and direct search along normally avoided paths
• Have access to internal state of problem solver, so can access any knowledge source
• May be entered by hand or acquired by learning
• May be general or domain-specific
• May be heuristic preferences or absolute selections
• Permit emulation of many search heuristics
• Casual commitment strategy relies on presence of control rules for important decisions