Causal AI: A Theoretical Introduction

This 7-hour course provides a theoretical introduction to Causal Artificial Intelligence (Causal AI) / causal inference.

In many data-driven settings, decisions are still based on correlations. However, when the goal is to understand the effect of actions or interventions, correlation-based models alone are not sufficient. In these settings, we need to reason about cause and effect.

Causal inference provides a structured way to do this. Over the past decades, several causal inference frameworks have been developed. One of the most widely used is the structural causal model framework associated with Judea Pearl, which introduces tools such as causal graphs, structural causal models, and do-calculus to map causal quantities to associational ones that can be estimated from observed data.

This course introduces you to that Pearlian framework. The goal is to develop a clear understanding of its core components and how they fit together, as this framework forms the foundation for modern applied causal analysis.

The course introduces the main components of causal inference within the Pearlian framework and explains how they relate to one another.

Foundations of causal inference

• The distinction between association and causation

• The role and limitations of randomized experiments

• The Ladder of Causation

Causal graphs

• Causal directed acyclic graphs (DAGs)

• Graphical concepts such as d-separation

• The role of causal graphs in causal analyses

Identification of causal effects

• The definition of causal quantities

• The backdoor and frontdoor criteria

• Do-calculus

Estimation and extensions

• An overview of estimation approaches

• The connection between causal quantities and statistical estimation

• An introduction to causal discovery methods

By the end of the course, you will understand what causal inference is about, how causal quantities are defined, under which assumptions these quantities can be estimated from observational data, and how the Pearlian framework can be used to structure causal analyses.

This theoretical foundation provides a natural starting point for more applied courses within the Causal Academy.