Logic and Rule-Based Reasoning

Facts, rules, inference

Logic-based AI represents knowledge as facts and rules and uses inference to derive new facts. Forward chaining starts from known facts and applies rules until nothing new follows (good for monitoring and alerts). Backward chaining starts from a goal and works back to see whether facts support it (good for diagnosis questions). Expert systems used thousands of such rules. Rule systems are transparent (every conclusion has a trace) and still widely used in business rules engines, compliance checks and configuration.

Represent facts, derive new ones

Symbolic AI stores knowledge explicitly and reasons over it with rules.

Forward chaining over three rules, run

I ran this with plain Python 3 (standard library only), with fixed random seeds where randomness is used. From fever, cough and recent travel, the engine derives respiratory_infection and then recommend_test in the first pass. It cannot derive isolate because test_positive is unknown, and it says so rather than guessing.

facts = {"has_fever", "has_cough", "recent_travel"}
rules = [  # (conditions, conclusion)
    ({"has_fever", "has_cough"}, "respiratory_infection"),
    ({"respiratory_infection", "recent_travel"}, "recommend_test"),
    ({"recommend_test", "test_positive"}, "isolate"),
]
changed = True; round_no = 0
while changed:
    changed = False; round_no += 1
    for conds, concl in rules:
        if conds <= facts and concl not in facts:
            facts.add(concl); changed = True
            print(f"round {round_no}: {sorted(conds)} => {concl}")
print("final facts:", sorted(facts))
print("'isolate' derived:", "isolate" in facts, "(needs test_positive, which is unknown)")

Output:

round 1: ['has_cough', 'has_fever'] => respiratory_infection
round 1: ['recent_travel', 'respiratory_infection'] => recommend_test
final facts: ['has_cough', 'has_fever', 'recent_travel', 'recommend_test', 'respiratory_infection']
'isolate' derived: False (needs test_positive, which is unknown)

Keep rules testable

Write test cases for rule sets just like code; conflicting or unreachable rules are common as rule bases grow.

Quick check: What does forward chaining do?

• Searches a game tree
• Starts from a goal and works backwards
• Trains a neural network
• Applies rules to known facts repeatedly to derive new facts