Knowledge Representation Issues — Beginner-Friendly Notes
1. What Knowledge Representation (KR) Is
KR = how to store what the agent knows in a form the agent can reason with to act. Without good KR, reasoning (Module 3-4) fails.
Good KR must be:
- Expressive: can say what matters (facts, relations, uncertainty, time).
- Concise: brief and maintainable.
- Efficient: reasoning is fast enough.
- Correct: means what you intend (clear semantics).
2. Five Core Issues (exam expects these)
1. Adequacy — What Can You Say?
- Representational adequacy: can you encode needed facts? (e.g., "every student has a mentor" needs quantifiers, not just propositions).
- Inferential adequacy: can you derive new facts efficiently (e.g., from "birds fly" + "Tweety is bird" infer "Tweety flies")?
2. Granularity — Level of Detail
Too fine → explosion (model every atom); too coarse → misses nuance (treat "bank" as one meaning). Choose right abstraction.
3. Commonsense & Incompleteness
Real knowledge is incomplete, vague, default ("birds fly" but penguins don't) — KR must handle exceptions, defaults (non-monotonic reasoning).
4. Reasoning Trade-off — Expressive vs Tractable
More expressive logics (first-order, temporal) are computationally harder (often undecidable/slow). Propositional is fast but weak. You trade power for speed.
5. Ontology & Commitment
What exists in your world? Objects, relations, time, beliefs? Different KR commit to different views (e.g., logic commits to true/false facts; objects + relations commit to entities).
3. Desiderata (extra scoring points)
- Declarative vs Procedural: declarative "what is true" (logic) is easier to maintain; procedural "how to deduce" is faster but brittle.
- Learnability: can agent acquire new knowledge easily (learning agents)?
- Explainability: can you explain why you concluded something (important for rule-based/expert systems).
4. One-Line Summary
Knowledge representation must balance expressive adequacy (saying what matters, inferring efficiently) with granularity, handling of commonsense exceptions, and computational tractability across different ontological commitments.
5. Quick Self-Check
- Why is "birds fly, except penguins" a KR challenge?
- Give a trade-off between propositional logic and first-order logic.