Expert Systems Questions and Answers

Answer: C

Both statements are accurate: a justification-based system links each belief to its supporting beliefs, while an assumption-based system labels each belief with the assumption sets under which it holds. The difference is that the ATMS can hold several contexts open at once, at the price of labels that may grow combinatorially. Choosing between them turns on how many alternative worlds the task requires.

Answer: D

MYCIN chained backward from candidate organisms and used certainty factors for uncertainty, so the first two statements hold. It was never used clinically, for reasons of liability, integration with hospital practice and the labour of keeping its knowledge current, so the third fails. That a system can be validated as competent and still not be adopted is one of the more durable lessons of the era. Naming the cost of a choice alongside its benefit is what turns a fact into a design decision. The quantity that varies and the quantity that does not are worth naming before any substitution is attempted. Checking the units of each quantity catches a wrong formula before the arithmetic begins. Most errors here come from carrying a result across from a neighbouring case where it does not hold.

Answer: A

Conflict resolution chooses within the conflict set, and because different choices can lead down different lines of reasoning it forms part of the system's semantics. Determining which rules are eligible is pattern matching, performed before resolution, so the second statement misattributes it. Keeping matching and resolution apart is necessary to understand what Rete accelerates and what it does not.

Answer: B

All three are genuine features: defaults express typical values, inheritance shares structure down a hierarchy, and attached demons compute values on access. Together they make frames a richer representation than rules for describing objects, which is why the two are usually combined rather than chosen between. The overridable default is also what makes frame reasoning non-monotonic.

Answer: C

Tacit knowledge and the techniques developed to reach it are correctly described, but acquisition dominated rather than trailed the cost, which is why it was named a bottleneck. Machine learning was eventually the more effective response, replacing elicitation with induction from data. The third statement inverts the historical record and is the one to reject.

Answer: D

The first three pairings are correct, while PROSPECTOR assessed geological sites for mineral deposits rather than forecasting weather. These four systems between them cover medicine, chemistry, configuration and geology, which is why they are the standard set. Learning them as a group, with the reasoning method each used, answers most historical questions on this topic in one stroke.

Answer: A

Only the last pairing fails: a shell is an application with its domain knowledge removed, not a completed one. The first three describe the standard architecture correctly. Because the shell's whole value lies in what it lacks, the definition is worth stating in that negative form. EMYCIN, obtained by emptying MYCIN, is the historical demonstration of the point.

Answer: B

The last pairing is wrong: a rule that can never fire is unreachable, whereas subsumption means one rule's conditions are implied by another's. The other three are stated correctly. These four defects, plus unreachability, are what static verification of a knowledge base looks for, and being able to name each from its description is the usual form of the question.

Answer: C

A training set belongs to machine learning, where a model is induced from examples, whereas the classical architecture consists of a knowledge base, an inference engine, working memory, an explanation facility and a user interface. The distinction matters historically, because learning from data eventually addressed the acquisition bottleneck that hand-built knowledge bases could not escape. A single measurement chosen to respond to the one asymmetry beats several that respond to shared behaviour. Reading the definition literally, rather than recalling a formula, is the reliable route through questions of this shape.

Answer: D

Unification is the pattern-matching operation of logic programming, which finds a substitution making two terms identical, and it has nothing to do with choosing among eligible rules. Recency, specificity and refractoriness are genuine strategies, prefering the newest data, the most detailed rule and an instantiation that has not already fired. The three are usually applied in a fixed order.