Definition
Plain language
Figuring out which things cause which, rather than just which things tend to happen together.
As stated in the literature
The task of inferring causal structure (a directed graph) among variables; observation alone identifies only a Markov equivalence class, while interventions can resolve remaining ambiguity.
Why it matters: Knowing what causes what, not just what correlates, is essential for predicting the effect of an action before you take it.
For example, it tries to determine whether smoking actually causes a cough or whether both just tend to appear together for some other reason.
Heard on the show
“Now, here's the thing that makes causal discovery hard.”Episode 091 — When Better Fine-Tuning Can't Help: A Geometric Impossibility in LLM Causal Reasoning