Questions: Intelligent Agents and Task Environments

Question 1

How is the task environment of a computer program playing Texas hold 'em poker against other players best classified?

• A. Fully observable, single-agent, deterministic, episodic
• B. Partially observable, multiagent, stochastic, sequential
• C. Fully observable, multiagent, deterministic, sequential
• D. Partially observable, single-agent, deterministic, episodic

Answer: (B)

Opponents' cards are hidden (partially observable), there are other agents with their own goals (multiagent), the deal is random (stochastic), and earlier bets affect later decisions within a hand (sequential).

Question 2

A thermostat turns the heater on whenever the sensed temperature is below 20 degrees C and off otherwise. Which kind of agent is it?

• A. Simple reflex agent
• B. Goal-based agent that plans ahead
• C. Utility-based agent
• D. Learning agent

Answer: (A)

It chooses its action from the current percept alone using a condition-action rule. It keeps no model of the world's history, does not search over future action sequences, and does not change its rules from experience.

Question 3

A rational agent, by definition, never takes an action that turns out badly.

• T. True
• F. False

Answer: False

Rationality is judged by expected performance given the information available at the time. An agent that crosses a street after looking both ways is acting rationally even if an unforeseeable event harms it. Demanding perfect outcomes would require omniscience.

Question 4

Write a PEAS description for an automated taxi driver.

Model answer: Performance measure: safety, reaching the correct destination, trip time, legality, passenger comfort, fuel or energy cost. Environment: roads, other vehicles, pedestrians, traffic signals, weather, customers. Actuators: steering, accelerator, brake, signals, horn, display or voice to passengers. Sensors: cameras, lidar or radar, GPS, speedometer, odometer, engine and battery sensors, microphone or touchscreen.

PEAS forces designers to state what success means and what the agent can perceive and do. The taxi example shows a partially observable, multiagent, stochastic, sequential, dynamic and continuous environment, which is among the hardest kinds.