Causality – Hey, We Have Machine Learning, So Why Even Bother?
Our journey starts here.
In this chapter, we’ll ask a couple of questions about causality.
What is it? Is causal inference different from statistical inference? If so – how?
Do we need causality at all if machine learning seems good enough?
If you have been following the fast-changing machine learning landscape over the last 5 to 10 years, you have likely noticed many examples of – as we like to call it in the machine learning community – the unreasonable effectiveness of modern machine learning algorithms in computer vision, natural language processing, and other areas.
Algorithms such as DALL-E 2 or GPT-3/4 made it not only to the consciousness of the research community but also the general public.
You might ask yourself – if all this stuff works so well, why would we bother and look into something else?
We’ll start this chapter with a brief discussion of the history of causality. Next, we’ll consider a couple of motivations for using a causal rather than purely statistical approach to modeling and we’ll introduce the concept of confounding.
Finally, we’ll see examples of how a causal approach can help us solve challenges in marketing and medicine. By the end of this chapter, you will have a good idea of why and when causal inference can be useful. You’ll be able to explain what confounding is and why it’s important.
In this chapter, we will cover the following:
- A brief history of causality
- Motivations to use a causal approach to modeling
- How not to lose money… and human lives