Knowledge Representation Questions and Answers

Answer: C

A skeptical reasoner draws only conclusions that hold in every consistent way of applying the defaults, so where two conflict it concludes neither. A credulous one picks an extension and commits to it. Ordering by assertion is arbitrary, tolerating inconsistency destroys the logic, and refusing entirely discards the unambiguous conclusions too. Choosing between reporting only what is certain and committing to a coherent guess is a decision about what the system's output is for.

Answer: D

An extension is a self consistent set of conclusions obtained by applying as many defaults as possible, and where defaults conflict there are several such sets. The skeptical conclusions are those in every extension. Non-applied defaults and language extensions are different notions. That a logic may have several equally legitimate answers, rather than one, is the structural consequence of allowing defeasible rules and it must be faced rather than engineered away.

Answer: A

A database that reports no matching record, and a program that concludes not p because p cannot be proved, are making the same assumption, which is why logic programming and databases share a semantics. Constraints, transactions and indexes are different mechanisms. Recognising a familiar practice as an instance of a formal assumption is what allows the consequences of that assumption to be transferred from one setting to the other.

Answer: B

Under negation as failure the inability to derive a fact licenses its negation, which is sound only if the knowledge base is complete about that fact. It is the operational form of the closed world assumption and is what a logic programming system implements. The other options misstate the direction or the treatment. An inference rule that is convenient and sound only under a condition rarely checked is a persistent source of confident wrong answers.

Answer: C

If a fact is simply absent because nobody recorded it, concluding its negation asserts something unsupported, and in an open domain that happens constantly. Non-termination is a separate practical problem that makes the rule inapplicable rather than unsound. Variable count and derivation are irrelevant. Checking whether the completeness the rule assumes actually holds, before relying on the rule, is exactly the discipline the open and closed world distinction exists to enforce.

Answer: D

Starting a car requires fuel, a working battery, no potato in the exhaust and indefinitely many further conditions nobody would think to state, so any finite list is incomplete. Default reasoning, assuming nothing unusual holds, is the standard response. The other options describe different difficulties. A problem arising from the open ended nature of the world, rather than from any defect of the formalism, is one that must be managed rather than solved.

Answer: A

Moving a briefcase moves everything inside it, and stating every such consequence for every action is impossible, so indirect effects must be derived from domain constraints instead. It is the companion difficulty to the frame and qualification problems. Non-determinism, ordering and preconditions are different issues. Three related problems, all arising from the impossibility of enumerating what an action does and does not do, together constitute the central difficulty of reasoning about action.

Answer: B

Asserting that the contents of a container are always where the container is lets the indirect effect follow from the constraint rather than from the action's description. Explicit enumeration is what proves impossible, restriction changes the domain, and full recomputation is neither possible nor meaningful. Stating an invariant once, and letting the consequences follow, is the same economy that inheritance provides in a taxonomy.

Answer: C

Domain constraints hold in every situation and therefore constrain the effects of every action, which is how indirect consequences are derived without enumerating them. Permissions, value ranges and scope boundaries are different notions, though value ranges are a special case. Distinguishing a fact that happens to hold from one that must hold in every situation is what allows the second to do inferential work the first cannot.

Answer: D

Representing that one agent knows something another does not requires the content of the belief to be an object in the language, which is reification applied to propositions. Fictional entities, abstract concepts and internal data structures are different notions. A system that must model other agents, whether to cooperate or to compete with them, cannot avoid representing what they believe, which is why the machinery is needed rather than merely interesting.