1 Introduction to causal AI
This chapter covers
- Defining Causal AI and causal data science
- Describing how Causal AI is robust, explainable, and increases value
- Making machine learning fairer with causal analysis
- Extending probabilistic ML workflows and programming toolsets to causal generative models
- Exploring how the commodification of inference trend in ML empowers causal modeling
This chapter will introduce some key motivations for learning causal AI. I’ll start with what causal AI means and its importance to current data science best practices and the current state-of-the-art of machine learning. I’ll also give some intuition for how algorithmic causality could unlock the next wave of AI.
Next, I’ll present a concrete example of the causal modeling workflow on the MNIST image dataset, which is essentially the “hello world” of machine learning. This machine learning example gives great insight while building intuition with clear extensions to the more classical statistical data examples of most causal inference texts (we’ll have several of those types of examples in this book as well).
Finally, I talk about a trend I call the commodification of inference. This trend motivates this book’s approach of using cutting-edge machine learning frameworks like PyTorch to implement causal models and handle the statistical heavy lifting of causal inference.