The well-known 20th century scientist J. B. S. Haldane is famously
quoted as having said, “Teleology is like a mistress to a biologist: he cannot live without her, but he’s unwilling to be seen with her in public.” Haldane’s point was that teleology, the study of purposes and primal causes, is critical to the understanding of natural phenomena, particularly the choices that biological creatures make; however, as teleology is also tied up with creationism and is therefore perceived as being against the Darwinist tradition of modern biology, it is officially taboo and unacceptable.
A similar situation occurs with the study of causality, a
critical issue in science and even in everyday life, but one that has
confounded philosophers and laypeople alike for centuries. Causality
stands all but abandoned by classical statistics, as shown by the
dictum “correlation does not imply causation.” Statistics has been confined to correlation and other issues that are amenable to
mathematically precise formulation. As successors of statistics,
machine learning (ML) and artificial intelligence (AI) are similarly only applied to correlations, and are of little use for understanding causal relationships.
Judea Pearl, the primary author of this book, is possibly the scholar and scientist most known for his work to remedy this deplorable state of affairs, by bringing causality out from the dark shadows of
ignorance and indifference to the bright daylight of scientific
scrutiny and understanding. Pearl is of course also well known for
his work in probabilistic AI, specifically Bayesian networks, but he
has also, in the latter half of his career, done seminal work on
giving a formal structure to causal reasoning [1,2]. Though Pearl’s work has perhaps not yet had the large impact one would like to see, it has spawned some important related work .
The present book finds its place in precisely this context. It is a
very readable and engaging account of the quest, largely by Pearl himself, to restore causality to its rightful place as a pillar
of wholesome scientific reasoning. There are few formal prerequisites
for reading and understanding this book; the writing is, in the
tradition of Ian Stewart and Martin Gardner, very effective at
conveying sophisticated concepts to nonspecialists or even
talented and well-read amateurs.
The book, spread over ten chapters, gives a round-up of key concepts, including the “ladder of causation” (a concept from Pearl) as well as the history of statistical research, with Karl Pearson as the early antagonist and Sewall Wright as the early protagonist for causal reasoning. The sixth chapter deals with paradoxes and is particularly interesting to read; it clearly shows how paradoxes arise in our minds when probabilistic reasoning takes the place of causal reasoning. The explanation of the Monty Hall problem (which itself has famously confounded even some esteemed mathematicians) is a special highlight.
The primary value of the book is perhaps in showing the value of
causal understanding for counterfactual reasoning, the consideration
of scenarios that have not yet come to pass (and perhaps never will).
This is something that people do effortlessly in their daily lives,
but which AI and ML systems are completely incapable of. The book
considers some issues related to the current fads of big data and AI,
and sounds an appropriate note of caution about the limits of the
current approaches, suggesting how AI that incorporates causal
reasoning could cross hurdles that no current approach can.
More reviews about this item: Amazon, Goodreads