Expert Systems Questions and Answers

Answer: C

Building a causal model of a device demands more effort than eliciting associations, and simulating it costs more at run time, which is why practical systems often combine a shallow layer for common faults with a deep layer for the rest. Explanation actually improves with a model, modularity is unaffected, and stored cases belong to case-based reasoning. The hybrid is the usual engineering answer. Framing the question as which quantity the change acts upon makes the answer follow without calculation.

Answer: D

The classic criteria are that genuine experts exist, that they outperform novices, that a case takes minutes to hours rather than seconds or months, and that the knowledge is largely cognitive. Rapid change makes maintenance impossible, common sense is exactly what these systems lack, and interaction is normally essential. Checking a domain against this list before building was standard practice and remains sound advice. Checking the units of each quantity catches a wrong formula before the arithmetic begins.

Answer: A

Common sense is the one thing rule bases have never captured, since it is enormous, mostly unstated and resistant to enumeration. The other descriptions name favourable conditions: narrow scope, clear inputs and expert agreement all help. The Cyc project's decades of effort at encoding common sense by hand is the measure of how hard the excluded case actually is.

Answer: B

Cyc set out in 1984 to hand-encode millions of common-sense assertions, on the argument that no learning system could bootstrap without them. It is the largest such effort ever attempted and its partial success is still debated. CLIPS and EMYCIN are shells and PROSPECTOR is a geological system. Cyc's premise, that brittleness is a knowledge problem rather than an algorithm problem, is what makes it belong here. The quantity that varies and the quantity that does not are worth naming before any substitution is attempted.

Answer: C

Conventional programs interleave knowledge and control in procedural code, while an expert system keeps the domain knowledge declarative and lets a general engine apply it. That separation is what enables explanation, incremental authoring and shell reuse. Language choice is incidental, speed is usually worse, and validation guarantees nothing of the kind. Every other characteristic of these systems follows from this one. The quantity that varies and the quantity that does not are worth naming before any substitution is attempted.

Answer: D

Because a rule states a relation rather than a step in a procedure, adding one does not require rewriting anything else, which is what let knowledge bases grow through use. Recompilation, implementation language and regression testing are engineering practices rather than the enabling property. The qualification is that interaction through working memory erodes the independence as the base grows. A single measurement chosen to respond to the one asymmetry beats several that respond to shared behaviour.

Answer: A

A decision table lists every combination of a few conditions against the action each requires, which is compact, easy to check for completeness and easy to convert into rules. Frames, networks and blackboards address structure, relations and cooperation. The table's chief virtue is that missing and contradictory combinations are visible at a glance, which is exactly what verification of a rule base struggles to achieve. Framing the question as which quantity the change acts upon makes the answer follow without calculation.

Answer: B

Every combination of conditions appears as a column, so a gap or a contradiction is immediately visible, whereas the same defect hides easily in a list of rules. Expressive power is lower rather than higher, speed is comparable and evaluation order is not the point. Tables are therefore often used as the authoring medium and compiled into rules for execution.

Answer: C

Circular rules can drive an engine into an unterminating loop, or at best waste effort rederiving what is already known. Refractoriness limits the damage but does not remove the defect. Subsumption, redundancy and unreachability are the other three classical faults that static verification looks for. Detecting circularity amounts to finding a cycle in the graph of rule dependencies.

Answer: D

Forward chaining keeps asserting conclusions into working memory until the rules are exhausted, so what accumulates is a set of derived facts. The goal-rooted proof tree belongs to backward chaining, frontiers to search and dependency networks to truth maintenance. Because everything derivable is derived, forward chaining can do a great deal of irrelevant work when only one conclusion was wanted. Reading the definition literally, rather than recalling a formula, is the reliable route through questions of this shape.