Rationality for Mortals: How People Cope with Uncertainty (Evolution and Cognition Series) [Paperback]

This book has its roots in two earlier fields that sought to discover whether or how humans make "rational" decisions.

The first field can loosely be called unbounded rationality, and it has a long history, dating back to classical thinkers like Daniel Bernoulli, but made its greatest achievements in the mid-20th century, with such ideas as expected utility theory and Bayesian decision theory. In this view, individuals have stable and well-ordered preferences. When choosing between different options, either in situations of certainty, risk (known probabilities), or uncertainty (unknown probabilities), people could specify the parameters of a decision and then grind through the necessary computations to reach their rational and preferred option. These models were not always presented as literal models of how people actually make decisions, but "as-if" models of what goes on in the black box inside the brain.

Unbounded rationality, when taken as a literal model of how the world works, lead to frequently absurd predictions. In macroeconomics, the field of Real Business Cycle theory asserts that the long-term unemployed in a deep recession are just rationally choosing to consume more leisure than accept a lower market-clearing wage. Other anomolies included the Allais and Ellsberg paradoxes.

The heuristics and biases programme came as a response to the excesses of unbounded rationality. The idea is that instead of fully solving many of the difficult decisions we face, we take mental shortcuts -- heuristics -- which can lead to biases in decision-making. It lead to many well-documented effects; the "Linda the feminist bank teller" problem is one of the most famous. A description of Linda is read out, including many features that accord with ideas and morals that a feminist might be predicted to have. Many people then rate that Linda is more probably a "bank teller and active in the feminist movement" than a "bank teller". This is, of course, in violation of the rules of probability, as the subset of bank tellers active in the feminist movement cannot be larger than the superset of all bank tellers.

Gigerenzer's book is a direct response to the heuristics and biases programme. He agrees that people use heuristics, only that these shortcuts often lead to more-informed decision-making than an unbounded optimisation. Computers, as an example, are unable to catch a ball in real time, even though they can in theory perform all the necessary computations. Humans, on the other hand, can learn to catch balls very successfully by employing some incredibly simple heuristics. As another example, Gigerenzer argues that simpler techniques such as tallying can outperform the sophisticated statistical machinery of multiple regression in predicting out of sample data.

He also has a solution to the Linda the feminist bank teller problem. By asking people to state how many Lindas out of a reference class of 100 are bank tellers or feminist bank tellers, he's able to get answers from the great majority of people that obey the laws of probability. In his view, the problem lies in the formulation of the problem: people are not good at manipulating probabilities or dealing with abstract laws of logic, as technical terms like "probability" have many possible interpretations that don't match exactly with the academic definition. People are instead very good at manipulating naturally occurring frequencies, as we've had to deal with similar problems throughout human history.

Some of his thoughts are incredibly original and interesting. My one problem is that I feel he's fallen into the trap of both earlier paradigms: that of trying to explain "everything". Economists were once hopeful that the study of unbounded rationality could explain and predict all human behaviour. Similarly, one of the leaders of heuristics and biases said "Give me an axiom and I'll design the experiment that refutes it". Gigerenzer seems to be on a mission to rationalise every heuristics and to refute every bias. For instance, he says the well-documented "overconfidence bias" could be due to skewed distributions of talent. If some drivers cause nearly all the accidents, then the average (median) driver can in-fact be above-average (the mean).

But overconfidence is everywhere, even in games played for incredibly high stakes. In the stock market, for instance, there is far more trading than can rationally be explained, despite the costs of trading. Less informed traders should in theory mimic the smart money traders; in reality they bet against them.

These comments aside, this book is still really valuable for its unique perspective.