Russell and Norvig, Chapter 6: Agents That Reason Logically

6.1 A Knowledge-Based Agent [Fig6.1]
- knowledge base
  - declared
  - learned
- inference mechanism

6.2 The Wumpus World Environment [Fig6.2]
- Percept = [Stench, Breeze, Glitter, Bump, Scream]
  - agent cannot perceive its location
- Actions = [goforward, turnleft, turnright, grab, shoot, climb]
- agent dies upon entering room with pit or live wumpus
- agent's goal is to find gold, return to [1,1], and climb
- score
  - (-1) for each action attempted
  - (-10000) for dying
  - (+1000) for leaving cave with gold
- enhancements
  - randomly chosen wumpus worlds
  - multiple, communicating agents
  - moving wumpi
  - multiple gold pieces
  - natural language
  - learning
  - vision
  - speech
- reasoning
  - sample inferences
    - location of wumpus, pits, okay squares

6.3 Representation, Reasoning and Logic
- representation
  - syntax: form used to represent sentences
  - semantics: mapping from sentences to facts in the world
  - e.g., Wumpus world simulator <-> your HW3 solution
  - facts "follow" from facts
  - sentences "entail" sentences
- inference
  - checking validity of sentences which are guaranteed true regardless
    of their interpretation
- logics
  - propositional logic
  - first-order logic

6.4 Propositional Logic
- models: worlds in which a particular sentence is true
  - for propositional logic, a row in the truth table
- a logic is monotonic as long as entailed sentences are preserved
  as more knowledge is added

6.5 An Agent for the Wumpus World
- example: finding the wumpus
  - ~S11, S12, S21
  - ~S11 => ~W11 ^ ~W12 ^ ~W21            % represent in Prolog?
  - S12 => W11 v W12 v W13 v W22
  - S21 => W11 v W21 v W31 v W22
  - conclude: (W13 v W22) ^ (W31 v W22)   % implied XOR
- representing the rule:
  - stench(x+1,y) ^ stench(x,y+1) ^ ~stench(x,y) -> wumpus(x+1,y+1)
  - (n-1)*(n-1) propositions: s(2,1) ^ s(1,2) ^ ~s(1,1) -> w(2,2)
  - need first-order logic

Exercises
- 6.15 GSAT