Theory of Intelligence
MACHINE LEARNING•Advanced Production Ml•Specialized Learning Settings•2.4 Causal Inference

Introduction

Correlation is not causation: why prediction models can't answer 'what if?' questions, how confounders mislead, and how experiments and causal methods estimate real effects.

timer1 min read•format_list_bulletedConcept 1 of 3•forumLesson Discussion

Introduction

A model can predict very accurately that people who carry umbrellas are more likely to get wet. That doesn't mean taking away umbrellas would keep people dry. Both are caused by rain.

Machine learning models are excellent at finding associations: "when X is high, Y tends to be high". But many business and policy questions are causal:

  • "If we give this customer a discount, will they stay?"
  • "Does this medicine improve recovery?"
  • "If we change the recommendation algorithm, will people watch more?"

These ask what happens if we intervene — change something on purpose. Answering them needs causal inference, not just prediction.

Discuss theoretical concepts, mathematical formulas, and architectural foundations.

Supports inline code with `backticks`
No comments yet. Be the first to start the discussion!
arrow_backPrevious Lesson
2.3 Bayesian Machine Learning
Next Lessonarrow_forward
2.5 Federated Learning