4.14 Logical agents and propositional logic

Work through these

• Separate a knowledge base from the inference procedure that reads it

  One part holds what is known, the other derives what follows. Keeping them apart is what lets the same facts be reused by different reasoning methods, and it is the idea expert systems are built on.

• Define entailment, and distinguish it from implication and from provability

  Entailment is a relation between meanings: every world where the knowledge holds is a world where the conclusion holds. Implication is a connective inside a sentence, and provability is what one procedure can reach.

• Decide entailment by enumerating models

  Listing every assignment of true and false and checking the ones where the knowledge holds. It is correct, it is exponential, and it is the baseline every faster method is measured against.

• Convert sentences to conjunctive normal form and apply resolution

  Rewriting everything as a conjunction of clauses so that a single inference rule can do all the work. Resolution is complete for propositional logic, which is why the conversion is worth the effort.

• State soundness and completeness, and say why an agent needs both

  Soundness means nothing false is derived; completeness means nothing true is missed. An unsound agent acts on nonsense and an incomplete one walks into danger it had the facts to avoid.

• Work the Wumpus world: deduce which squares are safe

  A small grid where percepts about breeze and stench let an agent prove where it may step. It is the example that shows deduction doing something an agent could not do by pattern matching.