Artificial Intelligence Expert Systems RuleBased Systems

1
Artificial Intelligence / Expert
SystemsRule-Based Systems

• Justin Gaudry
• June 7, 2007

2
Rules

• Ifthen structure
• Antecedent -gt Consequent
• Hypothesis -gt Conclusion
• Antecedent tests an object with a possible value
• Consequent is action to take place when
  antecedent conditions are met or fact to be added
  to current set

3
Types of Rules

• Recommendation
• Gives advice
• Maybe you should
• Could be this
• Directive
• Perform an action
• Relational
• Demonstrates relationship between two objects or
  object and characteristic

4
Production Systems

• Most common form of expert system
• Use rules to provide recommendations, determine
  course of action, or solve a problem

5
Production Systems

• Knowledge base set of rules
• Knowledge of the system represented in the rules
• Working memory set of facts
• Starts with premises, adds new intermediate
  conclusions and perhaps removes old ones
• Inference engine aka interpreter
• Procedure by which rules are identified or
  triggered and action fired

6
Conflict Resolution

• Resolution necessary when several rules are
  triggered at the same time
• Some priority hierarchy must be developed
• Heuristic priority levels
• Rule with highest preassigned value wins
• If-elseif
• Longest-matching strategy
• Rule with most positive factors in antecedent
  wins
• More specific
• Temporal strategy
• Rule with most recent factors added to working
  memory wins

7
Chaining

• Forward chaining
• Using deductive reasoning to derive new
  intermediate conclusions from existing facts and
  make them new facts
• Data-driven reasoning
• Uncertain goal
• Present to future
• Breadth-first search facilitated
• Explanation not facilitated

8
Chaining

• Backward chaining
• Start from conclusion and show logical path
  backward to true antecedents
• Goal-driven reasoning
• Must have a clear and known goal
• Present to past
• Depth-first search facilitated
• Explanation facilitated

9
Forward Chaining

• Add premises to working memory
• While more rules can be fired (and possibly
  satisfactory conclusion has not been reached)
• Fire rule
• Add new facts to working memory
• Endwhile

10
Backward Chaining

• Add conclusion to facts to be proved.
• While more to be proved and more rules can be
  found
• If current fact is a conclusion to a rule
• Add rule hypothesis (es) to facts to be proved
• Endif
• Endwhile

11
Example Rules

• If something has spots, it is a Hereford.
• A Hereford is a cow.
• If something is a Hereford, it does not bark.
• If something has spots, it is an animal.
• If something barks, it is a dog.
• Dogs are not cows.

12
Example Facts

• Fido has spots.
• What new information can be added?
• Fido just barked.
• What new information can be added? Removed?
• Conclusion Fido is a dog.