Expert Systems Questions and Answers

Answer: A

A prototype of perhaps fifty rules was run against a handful of carefully chosen cases, and its behaviour on those decided whether the domain and the representation were viable. Rule counts, engine speed and vocabulary coverage measure effort rather than competence, and a system can accumulate all three while still answering badly. Judging by case performance early, before large investment, is the practical discipline that the rapid prototyping style of development was built around.

Answer: B

Expert disagreement leaves the knowledge engineer arbitrating a question neither the formalism nor the engine can settle, and whichever view is encoded becomes the system's opinion. Conflict resolution will pick one rule, so no double answer appears, explanation reports whatever fired, and shells load contradictory rules without complaint. Measuring the extent of disagreement among experts is therefore part of validation, since a system cannot reasonably be held to a standard the experts themselves do not meet.

Answer: C

Annotating each rule with who supplied it, when and on what basis makes it possible to review the parts of the base a change in the domain has affected, instead of re-examining everything. Storage format and rule count do nothing for currency, and running the engine tests consistency rather than truth. Provenance is cheap to record at authoring time and almost impossible to reconstruct afterwards, which is why the discipline matters so much more in a knowledge base than in ordinary code.

Answer: D

In regulated settings such as tax, benefits and compliance, the rules are written down already, they change only by legislation, and every decision must be justifiable line by line. That is exactly the shape a rule base fits, and no learned model matches it for auditability. Build cost, hardware needs and user extensibility are not what keeps them alive. The survival of the technology in these niches, long after the general enthusiasm faded, is a good illustration of matching a method to the structure of its problem.