structural causal model

A mathematical model of a causal system made of variables, structural equations that determine each variable from its direct causes plus an independent noise term, and a probability distributionprobability distribution, n.: A complete description of how likely each possible outcome is — for a fair die, one-sixth for each face. A language model produces one of these over its whole vocabulary at every step. over the noise. Its causal graph, usually a directed acyclic graph (DAG), draws an arrow from each cause to its effect. An SCM answers three kinds of query: observational, through the joint distribution it induces (as a Bayesian networkBayesian network, n.: A diagram of causes and effects with probabilities attached — say, how rain and a sprinkler can each make grass wet — that lets a computer update its beliefs about hidden causes when it sees evidence. does); interventional, by replacing an equation — written do(X = x) — and propagating the change; and counterfactualcounterfactual, n.: A statement about what would have happened if something had been different — 'had I left earlier, I would have caught the train'. AI uses counterfactuals to explain decisions and to measure causes., by inferring the noise from evidence and then intervening. Graphical criteria such as d-separation read off conditional independences, and the do-calculusdo-calculus, n.: A set of mathematical rules, devised by Judea Pearl, for working out the effect of deliberately changing something using only observational data plus a diagram of what causes what. determines when interventional quantities can be computed from observational data. SCMs are the formal backbone of Pearl-style causal inferencecausal inference, n.: Working out whether one thing actually causes another, rather than merely occurring alongside it — for example, whether a new feature made users stay longer or just happened at the same time., causal discovery algorithms and counterfactual fairnesscounterfactual fairness, n.: A fairness test that asks: had this person been of another race or sex, with everything that would have followed from that, would the system have decided the same way?, and in interpretabilityinterpretability, n.: The effort to understand how an AI system arrives at its outputs — what information it uses and how it combines it — instead of treating it as a sealed black box. the basis of causal abstractioncausal abstraction, n.: A mathematical way to check whether a simple, human-readable recipe really describes what a neural network does inside, by testing whether matching internal parts behave the same way when swapped., which tests whether a network implements a hypothesized high-level causal model.

We wrote down a structural causal model of the pipeline before deciding which variables to adjust for.