Research map / Probabilistic and scientific modeling / Causal discovery

Position: Probabilistic Modelling is Sufficient for Causal Inference

ICML 2025 · Probabilistic and scientific modeling › Causal discovery · approach: Causal effect identification · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Argues that standard probabilistic modelling (writing down the joint over all settings of interest) suffices for interventional and counterfactual causal inference, with do-notation and SCMs presented as syntactic shorthand within that framework.

Problem

Confusion in the ML community over whether causal inference requires a bespoke causal framework or notation beyond probability and statistics (e.g., Pearl's claims that a do-operator is needed).

Approach

Shows that interventional and counterfactual questions can be answered by writing a joint probabilistic model over observed and intervened/counterfactual settings ("twin model"), with do-calculus, SCMs and adjustment formulas reinterpreted as syntactic shorthand for such joints; illustrated with a log-normal aspirin example.

Key results

• Under a log-normal aspirin model, the interventional expectation E[Y*|theta] = (t*)^(-c) exp(b mu_Z + (b^2 sigma_Z^2 + sigma_Y^2)/2), showing the effect depends only on c.
• Naive conditional E[Y|T] can increase with dose even when c=0, illustrating confounding (Simpson's paradox-like behaviour).

Surprising findings

• Naive conditioning on the observed dose yields a positive dose-duration association even when aspirin has no effect (c=0).

Limitations

• Argument is conceptual and illustrated with a single toy example; no empirical benchmark.
• Probabilistic approach requires specifying all assumptions explicitly, which can be difficult in practice.

Experimental setup

Builds on: Structural Causal Models do-calculus Causal Bayesian Networks twin network

Closest papers by meaning

• Causality can systematically address the monsters under the bench(marks) (NeurIPS 2026 · similarity 0.85)
• Position: Causal Machine Learning Requires Rigorous Synthetic Experiments for Broader Adoption (ICML 2025 · similarity 0.85)
• Prediction-Powered Causal Inferences (NeurIPS 2025 · similarity 0.85)
• Can Large Language Models Infer Causation from Correlation? (ICLR 2024 · similarity 0.85)
• Generalized Bayes for Causal Inference (NeurIPS 2026 · similarity 0.84)
• Bivariate Causal Discovery using Bayesian Model Selection (ICML 2024 · similarity 0.83)