2.2 a. According to the agent program, if the agent’s location contains dirt, it will suck the dirt and gain one point. Otherwise, the agent will move to another square to see if it is dirty (which is the only way of getting another point in the future). In every case, the agent selects the action that can maximize its performance measure, so this agent is rational.

b. If both squares are clean, the agent has to stop moving left and right in order to prevent costing points. So we can improve the agent by installing memory on it so that it can remember the square it just left is clean. The agent can thus be rational.

2.3 a. False. The simple vacuum-cleaner agent is rational but can not sense the state of its adjacent square.

b. True. If the memory is required in this task environment, such as when each movement costs one point in vacuum world, then pure reflex agent can not be rational.

c. True. If every action results in the same result, no matter what the agents do, they get the same reward. Then every agent is rational.

d. False. The input of the agent function is percept history, yet the input of the agent program is only the current percept which is the only thing it can get from the environment.

e. False. When the agent needs to keep putting every percept into the percept sequence and can not delete them, the memory of agent will eventually be full. Thus the agent will fail.

2.4

PEAS description:
| |Performance |Environment |Actuators |Sensors |
| |Measure | | | |
|Playing soccer |Win/lose |Field |Head, upper body and |Eyes, ears |
| | | |legs | |
|Exploring the subsurface |Life that is found. |The subsurface oceans |Steering system, |Camera, GPS |
|oceans of Titan |Speed. Safety | |accelerator, brake | |

Property:

| |Playing soccer |Exploring the subsurface oceans of Titan |
|Observable |Partial |Partial |
|Single-agent |No |Yes |
|Deterministic |No |No |
|Episodic |No |No |
|Static |No |No |
|Discrete |No |No |
|Known |No |No |

2.6 a. Yes. Sometimes we need multiple ways to implement a function, such as we want the agent to calculate a/b (10/3) and return the integer part (3) or compute to two decimal places (3.33). In this case, there are more than one agent program implements the agent function.

b. Yes. If the function is to achieve an unreachable goal, such as drawing a triangle whose area is negative, it can not be implemented by any agent program.

c. Yes. One agent program only returns one action. Implementing multiple agent functions requires the agent program returns multiple actions, which is impossible. So each agent program implements exactly one agent function.