Artificial Intelligence Questions and Answers

Answer: A

The rational agent formulation asks that a system do the right thing given what it knows, which is measurable, general and does not require any theory of human cognition. Thinking humanly belongs to cognitive science, passing a conversational test is one particular and rather narrow criterion, and reproducing brain structure is a means some approaches adopt rather than a definition. Choosing a definition that can be evaluated is what allowed the field to make measurable progress.

Answer: B

The standard taxonomy crosses thinking against acting and humanly against rationally, giving four programmes with different criteria of success. Approaches, subfields and strength categories are all real distinctions but are not the four definitions. Setting out the alternatives explicitly matters because disagreements in this field are frequently about which definition is being used rather than about any matter of fact.

Answer: C

Turing proposed an imitation game in which a judge converses with a human and a machine through a text channel and tries to tell which is which. It deliberately avoids any requirement about internal mechanism, replacing an unanswerable question about thinking with an operational one about behaviour. Speed, mechanism and learning capability are not part of the criterion. Substituting an answerable question for an unanswerable one is the move the paper is famous for.

Answer: D

Turing argued that the original question is too meaningless to deserve discussion, and replaced it with a concrete test that can actually be conducted. That move set the pattern for much of the field, which prefers measurable performance to definitional argument. Measurement convenience, mechanism and funding were not his reasons. Replacing a philosophical question with an empirical one is a strategy that recurs whenever a discipline wants to make progress rather than argue.

Answer: A

Critics observe that a system might pass by evasion, mimicry of human error and conversational trickery without any of the capabilities the test was meant to detect, and that the test is anthropocentric and unhelpful as a research goal. Judges are readily found, and arithmetic speed and memory are not what it measures. A test that can be gamed measures the gaming as much as the capability, which is a difficulty every benchmark in the field has faced since.

Answer: B

Searle imagines a person following rules to produce Chinese responses without understanding Chinese, arguing that the system displays the behaviour without the semantics, so passing a behavioural test does not establish understanding. The systems reply and other responses contest it vigorously. It says nothing about speed, about the possibility of language processing or about the merits of neural approaches. A thought experiment attacking a criterion rather than a capability is best answered by examining what the criterion was for.

Answer: C

The strong position holds that a suitably programmed machine would genuinely have mental states, while the weak position claims only that machines can act as if they were intelligent. Almost all working research proceeds on the weak assumption, which is sufficient for engineering purposes. Generality, method and data scale are separate distinctions, and the general versus narrow contrast is often confused with this one. Keeping a philosophical claim separate from an engineering one avoids a great deal of fruitless argument.

Answer: D

The summer project at Dartmouth brought together McCarthy, Minsky, Shannon and others and gave the field its name and its founding agenda. Expert systems came two decades later, the winters later still, and neural network research predated it. A field acquiring a name and a community at a single meeting is unusual and it shaped how the subject saw itself for decades afterwards.

Answer: A

Founders predicted within a few years machines that would translate languages, prove theorems and match human intelligence generally, and those predictions were repeatedly unfulfilled. The gap between early successes on toy problems and the difficulty of real ones was underestimated. Statistical learning and formal verification came later and from other traditions. Early success on simplified cases misleading a field about the difficulty of the general case is a pattern worth recognising.

Answer: B

Two such periods, in the mid seventies and the late eighties, followed influential critical reports and the collapse of commercial expectations, and funding and enrolment fell sharply. Each was preceded by overpromising. Geography, hardware and methodological narrowing are not what the term describes. A discipline whose funding follows a cycle of promise and disappointment has a strong incentive to describe its capabilities carefully, which is a lesson the field has had to relearn.