Causal Inference and Counterfactual Reasoning
From correlation to causation
Causal questions
Causal AI asks what happens if we intervene on a variable, not just how variables correlate. Causal reasoning supports policy evaluation, robustness, fairness analysis, and scientific discovery.
Causal vocabulary
What does an intervention represent in causal inference?
Interventions correspond to actively fixing a variable, often written as .
Correct answer: Setting a variable externally
Why is correlation insufficient for causation?
Observational association alone cannot determine effect direction or intervention outcomes.
Correct answer: Because correlated variables can be linked by confounding, reverse causation, or selection bias.
Counterfactual reasoning
Counterfactuals ask what would have happened in an alternate world, given what actually occurred. This is central to explanation, blame assignment, and policy analysis, but it requires strong structural assumptions.
Levels of causal questions
Associational
- What is observed?
- Example:
Interventional
- What if we change ?
- Example:
Important limitation
Causal claims require assumptions, design, or experimental leverage; they cannot be extracted reliably from raw correlation alone.
Which expression is typically associated with intervention?
The -operator denotes intervention rather than conditioning.
Correct answer:
Name one application of causal AI.
Causal methods can help determine whether disparities persist under intervention or arise from confounding.
Correct answer: Fairness analysis