Bayesian networks intro

A Bayesian network is a graph of variables with arrows meaning direct probabilistic influence—plus tables that quantify those links.

Makes assumptions visible: what depends on what. Useful for diagnosis-style reasoning and for talking carefully about causality candidates (Course 16).

Nodes = uncertain variables. Edges = direct dependence. No edge = conditional independence assumptions. Query by updating beliefs when you observe some nodes.

Alarm ← Burglary / Earthquake; Alarm → NeighborCall—classic teaching story (use responsibly; it’s a cartoon).

Draw 4 nodes for “late shipment” with your best dependence arrows; list one independence you assumed.

⚠️ Myth: An arrow always proves causation.

✓ Reality: Graphs encode modeling assumptions; causal claims need more (Course 16).

⚠️ Myth: Bayes nets require “true” probabilities from nature.

✓ Reality: They’re models—calibrate and stress-test.

• Course 06 uncertainty; Course 04 Bayes examples; Course 16 causality
• AIMA / Pearl intro materials (cite specifically)