Probabilistic Modelling is Sufficient for Causal Inference
Bruno Mlodozeniec, David Krueger, Richard E. Turner
TL;DR
The paper argues that probabilistic modelling is sufficient for answering interventional and counterfactual causal questions, challenging the view that bespoke causal frameworks are necessary. It grounds the claim with concrete aspirin headache examples and a twin-model Bayesian-network formulation to handle interventions and counterfactuals by writing down the joint distribution of all variables. It then connects these probabilistic constructions to the standard causal toolbox (do-notation, do-calculus, SCMs) as syntactic devices rather than essential primitives, clarifying identifiability and the role of Markov equivalence. The discussion highlights advantages such as accessibility, flexibility, and applicability to non-graphical models and probabilistic programming, while cautioning that careful modelling assumptions remain crucial.
Abstract
Causal inference is a key research area in machine learning, yet confusion reigns over the tools needed to tackle it. There are prevalent claims in the machine learning literature that you need a bespoke causal framework or notation to answer causal questions. In this paper, we want to make it clear that you \emph{can} answer any causal inference question within the realm of probabilistic modelling and inference, without causal-specific tools or notation. Through concrete examples, we demonstrate how causal questions can be tackled by writing down the probability of everything. Lastly, we reinterpret causal tools as emerging from standard probabilistic modelling and inference, elucidating their necessity and utility.