Intelligent Agents Questions and Answers

Answer: D

The task environment is Partially observable, Stochastic and Episodic.

Russell and Norvig classify a part-picking robot's environment this way. Its camera cannot see everything (partially observable). Parts arrive in varied positions and grasps can fail (stochastic). Each part is picked and sorted independently of the previous ones, so the task divides into separate episodes (episodic). It is also dynamic and continuous.

The fully observable and deterministic options describe a far tidier world than a conveyor belt, and 'sequential' would mean that each decision affects future ones, which is not the case when each part is handled on its own.

Answer: C

The statement that is not true is Its environment as Internet is fully observable and sequential.

For an internet book-shopping agent, the internet is a partially observable environment: the agent can see only the pages it has fetched, never the whole web, and prices and stock change while it searches. Russell and Norvig classify such a task as partially observable.

The other statements fit the agent's PEAS description. Its interface for receiving user requests and pages acts as its sensors. Following links and filling in forms are its actuators. Its performance measure is obtaining the requested books at minimum cost and in minimum time.

Answer: A

The four types are (i), (ii), (iii) and (v): simple reflex, model-based, goal-based and utility-based agents.

Russell and Norvig describe four basic kinds of agent program in order of increasing generality: simple reflex agents (act on the current percept), model-based reflex agents (keep an internal model of the world), goal-based agents (choose actions that achieve goals) and utility-based agents (choose actions that maximise expected utility). Any of these can be made into a learning agent.

'Action based' is not one of the types, so options containing (iv) are wrong.

Answer: B

The two middle agents are the Matchmaker and Broker.

In multi-agent frameworks, middle agents help requester agents find provider agents. A matchmaker returns the list of suitable providers so that the requester contacts them directly, while a broker takes the request, deals with a provider on the requester's behalf and returns the result.

Dispatchers, planners and schedulers are other kinds of agents or components, not the standard pair of middle agents.

Answer: B

Conflicts in multiagent systems are resolved through Negotiation.

When autonomous agents have conflicting goals or compete for resources, negotiation lets them reach an agreement acceptable to all by exchanging proposals and counter-proposals, through mechanisms such as auctions, bargaining and the contract net protocol.

Co-operation describes agents working towards shared goals, where there is no conflict to resolve. Competition is the source of the conflict, not its resolution, and seclusion is not a recognised mechanism.

Answer: D

The one that is not an agent communication language is FIPA-QIF.

The recognised agent communication languages come from two efforts. FIPA, the Foundation for Intelligent Physical Agents, defined FIPA-ACL. The Knowledge Sharing Effort (KSE) defined KQML, the Knowledge Query and Manipulation Language, together with KIF, the Knowledge Interchange Format, used to express the content of messages.

There is no FIPA language called QIF, so option D is the odd one out.

Answer: B

Statements (i), (ii) and (iii) are true, so the answer is (i), (ii) and (iii).

An agent perceives its environment through sensors (i) and acts upon it through actuators (ii). A rational agent chooses the action expected to maximise its performance measure, and it should learn as much as possible from what it perceives so as to improve that performance (iii).

Statement (iv) is false. Rationality is not omniscience: an omniscient agent would know the actual outcome of its actions, but a rational agent only maximises expected performance given what it knows. Options with (iv) are therefore wrong.

Answer: B

An agent's behaviour is described by its agent function.

The agent function maps every possible percept sequence to an action. It is the abstract, mathematical description of what the agent does, implemented in practice by an agent program.

The percept sequence is the input to that function, not the behaviour itself. Sensors and actuators are how the agent perceives and acts, and the environment is where it operates. None of these alone specifies which action the agent takes.