Introduction to Artificial Intelligence Questions and Answers

Answer: A

Specifying the performance measure, the environment, the actuators and the sensors characterises a task fully enough that an agent can be designed for it. Writing the PEAS description first forces the designer to state what success means before choosing any technique. The other expansions describe no standard framework. A checklist that must be completed before design begins is a modest but effective way of preventing the commonest error, which is optimising the wrong thing.

Answer: B

Full observability means the agent's percepts at each moment reveal everything relevant about the state, so no internal memory of past percepts is needed. Determinism, the presence of other agents and static behaviour are separate dimensions of the same classification. Partially observable environments require the agent to maintain a belief state. Classifying an environment along these dimensions is what selects the class of agent architecture appropriate to it.

Answer: C

In a stochastic environment the same action in the same state can lead to different outcomes, so the agent must reason about probabilities rather than certainties. Partial observability, static behaviour and multi-agent settings are separate dimensions, though the last is often modelled as stochastic when the other agents cannot be predicted. Identifying which dimension a difficulty belongs to determines whether probability, memory or game theory is the appropriate tool.

Answer: D

A dynamic environment does not wait for the agent, so a slow decision may be acted upon in a state that no longer exists, which forces real time constraints on the design. Chess with a clock is semidynamic, since the environment is static but the score changes with time. Stochasticity, moving obstacles and changing objectives are related but separate matters. Whether an agent may take as long as it likes is one of the most consequential facts about a task.

Answer: A

A simple reflex agent maps the current percept directly to an action through a set of rules, with no memory and no model, which is fast and adequate only in fully observable environments. Model based agents keep internal state, goal based agents consider consequences, and learning agents improve with experience. The four form a ladder of increasing capability and cost, and choosing the simplest that suffices for the environment is the design principle.

Answer: B

If two different states produce the same percept but require different actions, a rule keyed only on the percept must choose wrongly in one of them, and no set of such rules can help. Internal state carrying a memory of previous percepts is the remedy. Speed, goals and learning are separate limitations of the architecture. Recognising when an architecture is provably inadequate, rather than merely inefficient, saves a great deal of fruitless tuning.

Answer: C

The agent keeps an internal state updated by a model of how the world evolves and how its own actions affect it, which lets it act sensibly when the percept alone is insufficient. Goals belong to the next architecture up, storing every percept is neither necessary nor practical, and action distributions belong to policies. Adding just enough memory to overcome partial observability, rather than storing everything, is the design idea.

Answer: D

A goal based agent combines its model with an explicit goal and searches or plans to find actions that bring the goal about, which makes it flexible when the goal changes. Reflex agents would need every rule rewritten. Speed is if anything worse, internal state belongs to the model based agent, and learning is a further addition. Separating knowledge of the world from the objective is what makes an agent reconfigurable rather than rebuilt.

Answer: A

A goal is binary, achieved or not, whereas a utility function ranks outcomes, allowing the agent to trade speed against safety or cost against quality, and to act sensibly when goals conflict or are uncertain. Multiple goals and models are handled by both, and learning is an orthogonal addition. Replacing a predicate with a number is a small change that greatly enlarges the range of situations an agent can handle sensibly.

Answer: B

The learning element improves the performance element using feedback from the critic, while the problem generator suggests exploratory actions that may be informative though not immediately optimal. The fourth component is easily forgotten and is what prevents an agent from becoming stuck in a locally good habit. The other lists describe hardware, decision components and a processing pipeline. An explicit place for exploration in the architecture is the design's most instructive feature.