The Book of Why

🧠 Free Recall ❓ Quiz 🔍 Cues 🔮 Simulation 🥷 ShadowBox 📐 Encode Cards

Premise: you cannot be data-driven about questions that data alone cannot answer. Every causal conclusion requires causal assumptions, and the honest, rigorous thing to do is state those assumptions explicitly

The Ladder of Causation

There are three fundamentally different levels of cognitive ability when it comes to reasoning about the world, and each answers a different kind of question:

Rung 1 — Association (Seeing)
What is? — Finding patterns and correlations in observed data. Most animals and all current machine learning systems operate here. "Customers who buy X also buy Y."

Rung 2 — Intervention (Doing)
What if I act? — Predicting what happens when you deliberately change something. This requires a causal model, not just data. "What happens to sales if I raise the price?" This is where experiments live.

Rung 3 — Counterfactuals (Imagining)
What if things had been different? — Reasoning about alternative histories. "Would this patient have survived had we given the treatment?" This underpins moral responsibility, regret, credit assignment, and legal reasoning.

You cannot climb the ladder with data alone — no amount of rung-1 statistical sophistication will answer rung-2 or rung-3 questions without additional causal assumptions.

Unstarred

Dunk (exclude from study)

The Book of Why

Review Questions

Todos

Sensemaker