Lessons being learned from an open-source causal AI suite [Hybrid]

Critical data science and decision-making questions across a wide variety of domains are fundamentally causal questions.  Answering these questions via causal analysis requires combining human domain expertise and sophisticated analytical frameworks in ways that often challenge practitioners.  In this presentation, I will describe some of Microsoft’s contributions towards an open-source causal AI suite that aims to broaden access to causal methods, including the DoWhy library for scaffolding best practices of a trustworthy causal analysis; the EconML library for estimation methods based on the latest advances in causal machine learning; the Causica library for end-to-end causal discovery and effect inference; and the ShowWhy no-code interfaces for causal machine learning tasks.  I will describe some of the applications of our tools and the lessons we are learning from their usage, including three topics that represent what we see as particularly critical open research challenges: how we elicit and capture domain knowledge; methods for validation, refutation, and sensitivity analyses; and support for unstructured and high-dimensional text and image data within a causal analysis.