Logic and knowledge bases

Logic systems store facts and rules, then infer new conclusions. A knowledge base (KB) is that store. Classical AI used this heavily before statistical learning dominated many tasks.

Rules still matter in products: business policies, safety constraints, medical protocols as guardrails—even when an LLM drafts text. Knowing logic’s strengths and brittleness prevents nostalgia and hype.

1. Encode facts (“Ada is a bird”) and rules (“birds typically fly”)
2. Ask a query (“does Ada fly?”)
3. Inference engine derives answers—or says unknown

Messy reality breaks brittle rules (penguins; exceptions; conflicting sources).

Tax software encodes rules; it does not “vibe” deductions. Exceptions still need human judgment.

Write 3 facts and 1 rule about your kitchen inventory. What query would fail if an exception appears?

⚠️ Myth: LLMs made knowledge bases obsolete.

✓ Reality: Hybrid systems use both—LLMs for language, KBs/tools for enforceable facts.

⚠️ Myth: If it is logical, it is fair.

✓ Reality: Rules encode human choices—including biased ones.

• Russell & Norvig AIMA (logic chapters)
• Course 09 RAG (modern retrieval as soft KB)
• Course 10 tool-use (APIs as living KBs)