4 How Machine Learning and Causal Inference can help each other

In the last decade we have seen an explosion of applications of machine learning in a wide variety of domains. It has become a very popular tool and we keep on seeing new advances every day. These advances, such as automatic translation or self-driving cars, have been so astonishing, especially in areas related to image, video, audio and text, that sometimes it may seem that machine learning is the definitive solution to mimic human intelligence. In fact, this has been the main goal of Artificial Intelligence (AI), an area that combines a wide range of techniques, from logic to robotics. Currently, AI has seen in machine learning a huge potential and has invested a lot of resources in it. However, as with any other tool, machine learning also has its limitations. One of them, as we will see in this chapter, is that it doesn’t handle causality well on its own. So, if AI ever wants to fully develop its potential, at some point it will have to also include causal inference techniques.

4.1 What does supervised learning actually do?

4.1.1 When do we need causal inference and when supervised learning?

4.1.2 The goal of data fitting

4.1.3 When the future and the past behave in the same way

4.1.4 When do causal inference and supervised learning coincide?

4.1.5 Predictive error is a false friend

4.1.6 Validation of interventions

4.2 How does supervised learning participate in causal inference?

4.2.1 Empirical and generating distributions in the adjustment formula

4.2.2 The S-Learner algorithm: a simple approach to evaluate the adjustment formula

4.2.3 The T-Learner algorithm: splitting data to improve

4.2.4 Cross – fitting: avoiding overfitting

4.2.5 Further reading

4.3 Other applications of causal inference in machine learning

4.3.1 Reinforcement learning

4.3.2 Fairness

4.3.3 Spurious correlations

4.3.4 Natural Language Processing

4.3.5 Explainability

4.4 Exercise solution