Session: Causal AI for Sustainable AI

The transition to sustainable societies requires not only accurate predictions but also actionable insight into the causal mechanisms underlying environmental, economic, and social systems. Traditional machine learning methods—while powerful at forecasting—often fail to distinguish between correlation and causation, leading to brittle models that can reinforce existing biases or perform poorly when systems change. This course introduces Causal Artificial Intelligence (Causal AI) as a frontier methodology that enables robust, generalizable, and policy-relevant inference in the sustainability domain. We will explore how causal reasoning can guide decision-making in climate mitigation, biodiversity preservation, resource allocation, energy transitions, and sustainable finance.

We begin by establishing the theoretical foundations of causal inference, including counterfactual reasoning, structural causal models, potential outcomes frameworks, and graphical models such as Directed Acyclic Graphs (DAGs). The course then transitions to advanced methods such as instrumental variables, causal discovery algorithms, causal representation learning, and Do-calculus, with emphasis on interpretability and stability under interventions.

Throughout the course, sustainability is not treated as a static context but as a dynamic, multi-layered challenge requiring interdisciplinary knowledge. Students will engage with real-world situations including carbon markets and social development indicators. Example applications include estimating the causal effect of green subsidies on emissions reduction, modelling climate risk exposure in financial portfolios, and the relationship between health, hunger and population dynamics. By the end of the course, students will be equipped not only with the technical toolkit to model causality in complex systems, but also with the conceptual insight to apply these tools responsibly in the service of a more sustainable future.