Hannah Correia: Leveraging flexible models with interpretable ML and eXplainable AI approaches for causal reasoning in complex systems

In the face of urgent planetary health challenges such as the effects of climate change on soil and crop health and socio-environmental threats to human health, traditional approaches to causal inference, while rigorous, often require extensive data collection, long-term experimental studies, or strong assumptions that may delay the ability to guide practical and timely decision-making. This talk explores how flexible machine learning approaches, commonly used for prediction and traditionally considered unsuitable for causal estimation due to concerns about interpretability and confounding, can be repurposed to generate potential interventions even in the absence of perfect causal understanding. Using eXplainable AI and interpretable machine learning, I will demonstrate how these tools can help uncover relationships in complex systems and guide focused research on early-stage interventions in agriculture and public health. We will also explore future directions in unifying diverse causal frameworks, emphasizing that causal reasoning should be seen as a continuum where even highly flexible models can contribute meaningfully to our understanding of causality, particularly in complex systems, provided correct model specification is prioritized.