Sources of Uncertainty — Beginner-Friendly Notes

1. Why Uncertainty Matters

Real AI never has perfect knowledge. Logic assumes true/false certainty; real environments are uncertain — agent must reason with probabilities (Module 4) not just logic (Module 3).

2. Six Sources (memorise for exam — give one example each)

1. Incomplete Knowledge (model gaps)

Theory is missing — e.g., medical knowledge incomplete, so symptoms don't guarantee disease.

2. Partial Observability

Sensors don't reveal whole state — e.g., self-driving cannot see car behind truck.

3. Noisy Sensors / Data

Measurements have error — e.g., GPS ±5m, camera blur, typos in records.

4. Non-determinism / Randomness

World is stochastic — e.g., dice, weather, market — same action → different outcomes.

5. Adversarial Uncertainty (other agents)

Other decision-makers hide intent — e.g., poker opponent cards, rival bidding.

6. Bounded Rationality & Laziness

Agent lacks time/compute to model fully → deliberately ignore details (abstraction adds uncertainty) — e.g., ignore wind when planning delivery route.

3. Why List This Before Probability?

To justify why probability is needed: each source makes crisp true/false reasoning fail — need degrees of belief $P(H|E)$.

4. How AI Handles Each Source (preview)

• Probability theory (next) handles randomness + partial observability.
• Bayesian networks (topic 4) structure dependencies.
• Fuzzy logic (topic 5) handles vagueness ("warm").
• Certainty factors (topic 6) patch rule-based systems.

5. One-Line Summary

Uncertainty in AI stems from incomplete knowledge, partial/noisy sensing, stochasticity, adversarial agents and computational laziness, necessitating probabilistic rather than purely logical reasoning.

6. Quick Self-Check

1. Is "sensor noise" partial observability or noisy sensing? (Both, but noise is accuracy, partial is hidden.)
2. Give one source that is not randomness but still causes uncertainty.