In this simple example system, the Control Rules are made explicit. They are very simple, but at least you can see how important the Control Rules are to making the system work.

Knowledge Base (list of productions or rules)

• R1. IF the animal chews its cud, THEN it is a ruminant.
• R2. IF the animal is a ruminant and has humps, THEN it is a camel.
• R3. IF a camel has one hump, THEN it is a dromedary.
• R4. If a camel has two humps, THEN it is a bactrian.
• R5. IF and animal has one or two humps, THEN it has humps.

Working Memory

Control Rules:

• CR-1. If the Left-Hand-Side of a rule matches working memory, it is marked Applicable.
• CR-2. If more than one rule is Applicable, then remove `Applicable' from any rule whose action would add a duplicate fact to working memory.
• CR-3. Execute the lowest-numbered rule marked Applicable. If none, then quit.
• CR-4. Remove Applicable feature from all rules. Return to CR-1.

Cycle 1.

R1 and R5 are applicable. CR-2 does not apply. By CR-3, R1 applies to WM.

Cycle 2.

R1 and R4 are applicable. But R1 is redundant, so R5 applies.

Cycle 3.

R1, R2 and R5 are applicable. By redundancy, R2 is applied.

Cycle 4.

R1, R2, R4 and R5 are applicable. R4 is applied.

Cycle 5.

R1, R2, R4, and R5 are applicable, but none apply by CR-2 because all are redundant. Thus, CR-3 makes it Quit.

This system is a forward-chaining (or data-driven) production system, since it is the information present in the WM (matching by Left-Hand Sides) that keeps controlling the choice of the next rule. In a backward chaining system, the rules are chosen depending on what appears on their Right-Hand Side so that some specific conclusion can be worked toward (eg, when doing a mathematical proof).

By R. Port (Sept/99) After Stillings, et al `Cognitive Science, An Introduction, 2d Ed' MITP, p. 165ff.