Unit 1 -- Oct 10, 2011

Terminology

  Perception -> Action Cycle  -- see, then do
     agent with sensors and actuators interacts with environment

  Environment ->  Deterministic or Stochastic
                  Discrete or Continuous
                  Benign or Adversarial

      Observable -> Fully or Partially
        Full has implicit memory in the environment
        Partial may use internal Memory of previous State

   schip's own definitons:
    The scope of each of the classes is important, so...

     * Fully vs Partially Observable -- Is the CURRENT state fully known?
        Observability covers just the current system state and doesn't
         preclude being uncertain about future states.
		Homework edit: If the Agent needs internal memory to know the
		  full system state then it's Partially Observable. If _everything_
		  is maintained in the Environment, then it's Fully Observable.

     * Discrete vs Continuous -- Are the ACTION's steps or real-valued?
        If you get a result that can take:
          A sub-set of values, it's Discrete;
          Any real-number value, it's Continuous.
        Note that DvsC can apply to sensor inputs as well, but it's not
         clear if that makes the whole environment Continuous...

     * Deterministic vs Stochastic -- Are the RESULTs known before
                                         taking an Action?
        If you can know the full Result of a Action before executing it,
         then the Result is Deterministic. Otherwise Stochastic, e.g.:
         a coin toss or dice throw.

     * Benign vs Adversarial -- Will SUBSEQUENT States change in order
                                 to counter your Actions?
        I.e.: Is the Environment out to get you?  This can be a feature of
         the Environment itself, but often requires a second Agent who is
         in competition. An example from the video is Weather, it's in the
         Environment and may interfere with your actions but it's not
         actively plotting against you (in most cases....), so it's Benign.
         Whereas a opponent in chess is trying to counter your moves and
         is thus Adversarial.

    

  AI as uncertainty management:
          "what to do when you don't know what to do"
      Reasons for uncertainty --
        Sensor limits
        Adversaries
        Stochastic environment
        Laziness & Ignorance

Did not mention Rationality at all until the summary....!?