Glosarium AI
Kamus lengkap Kecerdasan Buatan
Counterfactual Inference
Process of estimating the outcome of a hypothetical scenario by modifying a past cause while keeping all other conditions constant. It allows answering the question 'what would have happened if...?' by isolating the effect of a specific variable.
Conditional Independence Principle
Assumption that the potential outcome is independent of treatment assignment, conditional on a set of observed variables. This principle is the basis for covariate adjustment to estimate causal effects from non-experimental data.
Inverse Probability of Treatment Weighting (IPTW)
Adjustment technique where each observation is weighted by the inverse of the probability of having received the treatment it received. It recreates a pseudo-population where treatment assignment is independent of covariates, allowing for an unbiased estimation of counterfactual effects.
Pearl's Back-Door Criteria
A set of formal rules to identify the minimal set of variables to adjust for to block all non-causal back-door paths between a treatment and an outcome. Applying these criteria is essential to ensure the validity of counterfactual inferences.
G-computation Adjustment Formula
Parametric method for estimating causal effects that models the distribution of the outcome conditional on the treatment and covariates. It allows for the calculation of potential outcomes by standardizing over the covariate distribution to estimate the effect of a counterfactual intervention.
Average Causal Effect (ATE)
The expected difference between the potential outcomes if the entire population received the treatment and if it received the control. It is a fundamental counterfactual measure that quantifies the average impact of an intervention at the population level.
Average Treatment Effect on the Treated (ATT)
The average difference between the observed outcome for units that received the treatment and the counterfactual outcome they would have had if they had not been treated. This measure is particularly relevant for evaluating the impact of a policy on the population it actually affected.
Counterfactual Reasoning
Cognitive and computational process of constructing and evaluating alternative hypothetical worlds to understand causality. In AI, it is formalized by structural models to predict the consequences of unobserved actions.
Causal Regression Equation
A model that links an outcome variable to its direct causes, where each coefficient represents the causal effect of the corresponding cause, all other things being equal. It is the causal counterpart of the standard regression equation and serves as the basis for counterfactual simulations.