Intelligent agents
------------------

We would like to design intelligent agents, we need first to agree
on what it is?

Def 1: An agent is anything that
    + senses its environment        (perceive)
    + acts upon that environment    (act)

Exp.
    + Human     (look,hear,smell,touch ...   -   move)
    + Robot     (sense through sensors/cameras, steer the wheels)
    + Softbot   (inputs from environment, outputs)

    + Cleaning robot (senses/looks for dirts, clean & move around)

Picture of generic agents

Now, our goal is to design rational agents

Def. 2: (First attempt to define rational agents)
A rational agent - an agent that does the right thing
    + the right action: the action that makes it most successful
    -> evaluate the agent's success (how and when)

Two points to decide:
    + objective performance measure to determine how
        successful is an agent?
    + when to evaluate ?

-----------------------------------------
Example of irrational agents?
    + Does not do what it is supposed to do
        - A cleaning agent that sleeps the whole day/creates dirt
        - A teacher who does not come to the lecture

Need: Criteria to say that an agent is rational
-----------------------------------------

Exp.:

HOW successful is an agent?
    Performance measure
        - cannot be fixed
        - subjective
    ---> objective performance measure!
    + for a cleaning robot: the amount of dirt the robot can
    clean in a certain amount of time, say 8 hours

WHEN to evaluate the agent?
    + for a cleaning robot: after 8 hours

Another view of agent:
----------------------
    An agent could be viewed as a player in a game with the
    environment in which each action of the agent will be
    responded by the environment and vice versa.


Problems in interacting with the environment:
--------------------------------------------

    Environment is unpredictable  ---> agent cannot know the ACTUAL outcomes
    of its action
    --> what happens if an agent does the right action but does not get the
    expected result?


Rationality depends on:
    + performance measure (defines the degree of success)
    + percept sequence (everything that the agent has perceived so far)
    + what it knows about the environment
    + the actions it can perform

IDEALLY: agent should know what to do next given a percept sequence
(ideal rational agent)

---> We can create a lookup table for each percept sequence  -- one action

---> A rational agent is a mapping (ideal rational agent corresponds to
ideal mapping)

STOP - PROBLEM: infinite many possible percept sequences --> impossible to
have an ideal mapping --> there exists no ideal rational agents!

Fortunately, in many cases, we do not need to enumerate ALL possible sequences.

Exp. Agent computing z such that z^2 = x given x (table in the book)