Definition
Plain language
Notation for actually reaching in and forcing a variable to a value, instead of just observing it.
As stated in the literature
Pearl's do(·) notation denoting an external intervention that sets a variable, cutting its incoming edges; the formal tool that distinguishes interventional from observational distributions and lets interventions disambiguate Markov-equivalent causal graphs.
Also called: do(A)
Why it matters: It formally separates correlation from causation, which is essential for reasoning about what actually causes what rather than what merely co-occurs.
For example, it captures the difference between noticing that umbrellas and rain go together and actually forcing everyone to carry umbrellas to see if that changes the weather.