Sunday, October 25, 2015
"Superforecasting," by Philip Tetlock
Some years ago, Philip Tetlock published "Expert Political Judgement,"a book that reported the results of a research project showing that, on average, expert political scientists and economists were no better at forecasting important events than monkeys randomly throwing darts. All averages report incomplete information, and some years later Tetlock started a new research project trying to show that some people are better at forecasting than others. The new book, subtitled "The Art and Science of Prediction," shows the results of this new project. Tetlock and his colleagues found a new category of super-forecasters, individuals who were much better than others at predicting important events. These individuals are more modest than typical media pundits. They are not famous and they are not even on demand by broadcasters or newspapers, because they are bad at providing headlines. Tetlock is very good at showing with grace the errors of New York Times columnists. One wonders what would he think of our more local pundits. Instead, super-forecasters are good at assigning precise probability ranges, which they are able to update in the face of new information. They are also good at team-working and at being open to adversarial views. Another thing they do well is to be attentive to the outside view. For example, in front of a question such as would a given family buy a new pet, instead of giving priority to information about that particular family (the inside view) super-forecasters would look at what percentage of similar families had adopted pets in the recent past (the outside view). Tetlock also addresses the criticisms of behavioural sciences giants such as Daniel Kahneman or Massim Taleb. He shows respect for them, and their criticisms made him qualify his views. Kahneman argued that super-forecasting is cognitively exhausting, and that makes him pessimistic about whether their example is going to expand. But Tetlock says that precisely this point should make us reflect on ways to make institutionally lest costly to introduce good forecasting practices where they are needed. Taleb argued that most important phenomena are impossible to predict (such as "black swans"), but the irreducible uncertainty that exists calls as Taleb argues for "anti-fragile" institutions, and where to build such anti-fragility requires some degree of forecasting. In addition, Tetlock argues that although it is true that individual events (where will the next terrorist attack take place) are impossible to predict, what is possible is to predict slightly broader phenomena, such as the degree of risk of a terror attack. A wonderful book.