Knowledge Engineering

When possible you will want to provide additional layers of rules that can provide values for attributes the user cannot provide directly.
	Avoid knowledge base designs that are one rule deep and a hundred rules wide: an expert system that approximates the richness of an interview with a human expert will not describe each value of each goal variable with a single complex rule.
	Example: Suppose your knowledge base has the following rule that can determine a value for the recommended action goal variable: If the fuel tank is empty then the recommended action is refill the tank Could your expert determine whether or not the fuel tank is empty when the user does not know? The expert might suggest adding the following rule: If the starter cranks and a gasoline smell is not present then the fuel tank is empty with 90% confidence If the user answers the "is the fuel tank empty" question, this rule isn't needed. Otherwise, questions about the starter action and the gasoline smell will be generated as the inference engine backward chains. Responses to these questions might allow the original rule to fire, concluding that the recommended action is to refill the tank.
Continued...