In fields such as data science, explainable artificial intelligence (AI), and big data analytics, “why?” is the ultimate question that needs to be answered to make sense of real data. “Why?” has been the simplest question to advance science since the Age of Enlightenment. Now we want to ask the very same question of our computers, which are capable of crunching huge amounts of data and may discover new insights in data that are hidden to human scrutiny. This book postulates that an answer to the “why?” question calls for an inquiry into the causal relationships that make the generation of data under analysis possible. Thanks to the writing sagacity of Dana Mackenzie and the genius of Judea Pearl, this book shows the breakthrough of causal modeling to the scholar, the practitioner, and the curious reader.
This book has already been appraised in several reviews, and Judea Pearl is the 2011 recipient of the Turing Award for his fundamental contributions to AI, thanks to the development of probabilistic and causal reasoning. No additional words are needed to appreciate the quality of the book; therefore, I will focus on the book from the perspective of explainable AI (XAI) and data science. Roughly speaking, data science is a corpus of scientific methodologies and multidisciplinary techniques for extracting useful knowledge from data. Data scientists are equipped with powerful analytical tools to make sense from data. Among such tools, those offered by statistics and machine learning are the most widely used. More often than not, however, the kind of knowledge that is most interesting to acquire is of a causal nature, that is, the kind of knowledge that is required to answer “why?” questions, for example: “Why does a certain type of cancer occur in people?” “Why do some children show a certain level of intelligence?” In XAI, answering “why?” questions is even more crucial because XAI systems must be able to explain the reasons behind the decisions made by the machine. (XAI systems are in demand because of the increasing impact of intelligent systems in people’s lives, which call for “algorithmic transparency and accountability” .) In XAI, causal chains are necessary for providing explanations to users ; therefore, causal modeling is of utmost importance in this field.
From the first pages, those dreaming of causal inference by just looking at data may receive a cold shower: according to Pearl, no causal inference can be done with data only, but a causal model must accompany data in order to answer “why?” questions; data is only sufficient to find correlations without any hint of causation. To convince the reader, Pearl and Mackenzie guide the reader through the fascinating history of statistics, starting with Galton’s regression lines. Through this historical excursus, often enriched with lesser-known anecdotes and personal traits of the most relevant people that shaped statistics in its first century, the authors show why causal inference was not in the mainstream of statistical research (on the contrary: it was banned) and why it exploded in the so-called “causal revolution” of the last 30 years.
As a consequence of this historical delay, causal modeling and reasoning are still not as widespread as they should be, especially in the new field of data science in which they should be considered the most precious tools. Pearl postulates that statistical, association-based (that is, non-causal) models lay at the first of three levels (or rungs) of the causation ladder. Therefore, intelligent systems (which are at this basic level) can hardly achieve human-level intelligence, which is capable of imagination, retrospection, and understanding, all mental activities that are only possible through counterfactuals, the most powerful tool of causal reasoning.
Pearl and Mackenzie gently introduce causal reasoning through causal diagrams, the latter being innocent-looking graphs that unleash an unexpected representation power that can be used to reason in terms of the higher levels of causation, namely interventions and counterfactuals. Pearl shows how causal diagrams help in solving long-stated questions and apparent paradoxes that could not find definitive answers via statistical methods. To this aim, concrete examples in the most disparate fields of science abound in the book, including epidemiology, medicine, psychology, sociology, and so on. The book ends with some personal reflections from Pearl on big data, AI, and other “big” questions, including the feasibility and opportunity of human-like thinking machines and the impact of causal reasoning on this futuristic technology.
This book is for falling in love with causal reasoning. It is therefore nontechnical, but instead oriented toward a wide audience. After reading this book, the next step is to read Pearl’s Causality  in order to acquire the full paraphernalia of causal modeling and reasoning. Both books are must-reads for all people who want to put mind over data and make real sense of it.
More reviews about this item: Amazon, Goodreads