Expert Systems Questions and Answers

Answer: A

DENDRAL generated candidate molecular structures consistent with mass spectrometry data and is generally counted the first expert system. Its plan-generate-test organisation, using heuristics to prune an enormous space, became a standard pattern. XCON, MYCIN and PROSPECTOR worked in configuration, medicine and geology. DENDRAL also produced Meta-DENDRAL, an early attempt at learning the rules automatically. Most errors here come from carrying a result across from a neighbouring case where it does not hold.

Answer: B

Experts find it hard to articulate knowledge they apply automatically, so eliciting and encoding it dominated the cost of building these systems. Interviewing, protocol analysis and repertory grids were developed in response, and machine learning was eventually the more effective answer. Engine speed, uncertainty and interface design were real problems but not the bottleneck the term names. Naming the cost of a choice alongside its benefit is what turns a fact into a design decision.

Answer: C

Protocol analysis captures the expert's running commentary while solving a real case and then codes it into rules, which surfaces knowledge an interview would miss because the expert never thinks to mention it. Repertory grids, interviews and card sorting elicit different aspects. The methods are complementary, and a serious acquisition effort uses several rather than relying on any one of them. A limit that survives making every external component ideal is an intrinsic one, and worth knowing for that reason.

Answer: D

The explanation facility answers how a conclusion was reached and why a question is being asked, typically by walking the trace of fired rules. Its availability was a major reason users came to trust these systems, and it also served as a debugging aid for the knowledge engineer. Acquisition, memory and conflict resolution are different components. The facility is a direct benefit of keeping knowledge declarative. A limit that survives making every external component ideal is an intrinsic one, and worth knowing for that reason.

Answer: A

A HOW explanation traces the rules that fired to reach a given conclusion, back to the facts the user supplied. Its companion, WHY, justifies the question currently being asked by naming the rule that needs the answer. Confidence, timing and search effort are reported by other means if at all. Distinguishing the two directions of explanation is a routine examination point.

Answer: B

A WHY explanation names the rule whose condition the current question is trying to establish, and repeated WHYs walk further up the goal stack. It makes an otherwise arbitrary interrogation feel purposeful. Contrast with expectations, failed rules and contradictions are handled by debugging tools rather than by the explanation facility. The pairing of HOW with WHY covers backward and forward views of the same trace. The shared properties explain why these belong to one family, and the differences explain why more than one exists.

Answer: C

Certainty factors combine a measure of belief and a measure of disbelief into a single number between minus one and one, propagated by simple combination formulae. They were adopted because the joint probabilities Bayes would need were unavailable. Fuzzy grades and belief functions are rival schemes. Certainty factors were later shown to be sound only under strong independence assumptions, which is why probability displaced them. Framing the question as which quantity the change acts upon makes the answer follow without calculation.

Answer: D

The combination formulae treat every piece of evidence as independent, so correlated findings are double counted and confidence is overstated. Heckerman's analysis showed certainty factors are a probabilistic calculus only under assumptions the domains violate. They represent absence of confidence perfectly well, they are cheap, and they were adopted precisely to avoid eliciting priors. Bayesian networks answered the same need soundly. The quantity that varies and the quantity that does not are worth naming before any substitution is attempted.

Answer: A

A certainty factor runs from -1 for complete disbelief through 0 for no evidence to +1 for certainty, and it is the difference between a belief measure and a disbelief measure, each of which lies in the unit interval. The signed range lets confirming and disconfirming evidence be expressed on one scale. A unit interval, a percentage or an unbounded scale would each lose part of that convenience. Framing the question as which quantity the change acts upon makes the answer follow without calculation.

Answer: B

PROSPECTOR used an inference network with likelihood ratios to combine geological evidence, and famously identified a molybdenum deposit. Its odds-based updating was closer to sound probability than certainty factors were, though it too made independence assumptions. MYCIN, XCON and DENDRAL used other schemes. PROSPECTOR is the natural bridge from expert systems to the Bayesian networks that followed. Reading the definition literally, rather than recalling a formula, is the reliable route through questions of this shape.