Review: The Book of Why: The New Science of Cause and Effect

Correlation does not prove causation, but that should not stop us.

Until recently, progress in many fields has been retarded by the admonition that correlation of two or more items, measurements, or observations does not prove causation. Logically, this is still true, but The Book of Why: The New Science of Cause and Effect1 shows that we can make progress by carefully using the correlations as if we were certain of cause and effect. What’s more, Pearl’s technology permits quantitative predictions.

Artificial intelligence is one extreme example. AI searches for correlations, both positive and negative, from a training set. Then, it tests the correctness by running the algorithm on a test set of data. If the correlation holds and useful data or actions are supported, then it is used. In cases where the test runs fail; one can often learn something of value by examining the failure.

Pearl introduces the ladder of causation,2 where simple AI looks for correlations in the first rung, which he calls “seeing.” This is where most AI is today. However, if one models the question by adding “intervention,” one goes to the second rung. Here, one can ask and answer simple questions such as “How can I make XXX happen?” The third rung deals with counterfactuals such as "What would have happened if I acted differently?" This is the level of knowledge and understanding.

Logical diagrams are introduced in the book. These are accompanied by mathematical formalism, which is essential to take advantage of the technology, especially at rungs two and three. I found these to be tedious but understandable. The book is salted with interesting application examples from real life, which holds the attention of the reader. However, it would probably take a semester or more to become truly proficient at using it in one’s research.

Is it useful? You bet. At the 2018 meeting of Sigma Xi, a keynote address by Professor Timothy Davis of Texas A&M University used several causation diagrams to illustrate his success in developing algorithms for a variety of applications, including music and art.

References

  1. Pearl, J. and Mackenzie, D. The Book of Why: The New Science of Cause and Effect. Hachette Book Group: New York, NY, 2018.
  2. Ibid., 28.

Robert L. Stevenson, Ph.D., is Editor Emeritus, American Laboratory/Labcompare; e-mail: [email protected]

Related Products

Comments