Backward chaining
Backward chaining is a goal-driven inference strategy in which reasoning starts from a desired conclusion and works backward through rules to find facts th...
Goal-Driven Reasoning in Expert Systems
Backward chaining is a goal-driven inference strategy in which reasoning starts from a desired conclusion and works backward through rules to find facts that would support it. Rather than generating all possible conclusions from available data, backward chaining focuses narrowly on proving a specific goal, making it efficient when only particular conclusions are needed.
The process begins with a goal fact. The inference engine searches the rule base for rules whose consequent matches the goal. If such a rule is found, its preconditions become new subgoals. The system then tries to prove each subgoal, either by finding it in working memory as a known fact or by finding another rule that could derive it. This recursion continues until all subgoals are proven (the original goal is established) or the search fails.
Backward chaining is closely associated with Prolog, a logic programming language where queries are resolved by exactly this mechanism. When a Prolog program is asked "Is Socrates mortal?" it finds the rule "mortal(X) :- man(X)" and then asks whether "man(Socrates)" is a known fact.
A concrete example from medical diagnosis: the goal is to determine whether a patient has pneumonia. A rule states "IF fever AND cough AND abnormal chest X-ray THEN pneumonia." Backward chaining creates three subgoals: verify fever, verify cough, and verify abnormal X-ray. If all three are confirmed from patient data, the conclusion is reached.
Backward chaining naturally avoids exploring irrelevant parts of the rule base. If only ten of a thousand rules are relevant to the stated goal, backward chaining touches only those ten.
Exam tip: Remember that backward chaining is used in Prolog and many expert system shells. A common exam question asks whether forward or backward chaining is more appropriate for a given scenario — choose backward when the goal is known but data is uncertain.