Bounded Rationality: The Adaptive Toolbox

2012-11-30 13.16.23-1Gerd Gigerenzer and Reinhard Selten provide a good summary of this topic in “Rethinking Rationality.” This is the introductory chapter to Bounded Rationality, which they edited. To connect the rational and the psychological, bounded rationality includes emotions, social norms, imitation, and other cultural tools. The initial modern definition of rationality was created in about 1654 by Blaise Pascal.  That definition was simply to maximize expected value.  Blaise Pascal saw it as creating a new form of morality as well–replace faith with moral expectation.  That definition ran into trouble with the St. Petersburg paradox, and Allais, and Ellsberg paradoxes. So David Bernouilli redefined rationality as maximizing expected utility.  Bounded rationality was created by Herb Simon in the 1950s.  Optimization was not part of this new rationality.  The metaphor for bounded rationality was a pair of scissors, one blade is “cognitive limitations” and the other is the “structure of the environment.” In Herb Simon’s words written in 1956: “a great deal can be learned about rational decision making…by taking account of the fact that the environments to which it must adapt possess properties that permit further simplification of its choice mechanisms”.  Gigerenzer argues that simple and robust heuristics can match a specific optimizing strategy.  Different researchers see “bounded rationality” differently.  Gigerenzer says it is not optimization nor irrationality nor is it optimization under constraints which includes limited search, but then requires cost/benefit of further search.  Gigerenzer’s model of bounded rationality uses fast and frugal stopping rules (when to stop search) that do not involve optimization.  Kahneman and others have pointed out humans falling short of rationality in such things as the base rate fallacy and conjunction fallacy.  Gigerenzer & Selten contend that these fallacies are “based on norms that have been put forth without analzying the structure of the environments… Moreover, when information is presented in natural frequencies rather than probabilities, base rate neglect is perfectly rational.”

Gigerenzer’s favorite illustration of a heuristic is the gaze heuristic.  He proposes building a robot that can catch fly balls.  One team of engineers starts by figuring out the instrumentation they will need to measure ball velocity and spin, projection angle, direction, wind speed, distance from the robot, etc. The other team of engineers watches some human outfielders and sees that they wait a half second or so and then take off in the general direction of the ball while adjusting running speed so that the angle between the eye and the ball remains constant. This works.

baseballimagesThe adaptive tool box contains a number of tools that can each a certain class of situations.  There is not a single hammer for all purposes.  Bounded rationality includes simple search rules, simple stopping rules, and simple decision rules. There are two kinds of search”  those that search for alternatives such as satisficing strategy and those that search for cues such as fast and frugal heuristics.

In the book, four big questions are addressed:

1.  Is there evidence for an adaptive toolbox?  There answer is yes.  There is not a universal calculus, but there rules or heuristics.  These heuristics are fast, frugal, and computationally cheap. The heuristics are matched with the environment and there are mechanisms for selecting a heuristic.

2. Why and when do simple heuristics work?  They are not multiple regression or Bayesian networks, but they can exploit structures of information in the environment.  Simple strategies are robust compared to models with large numbers of parameters, which risk overfitting.  Finally, some situations are beyond optimization.

3.What role do emotions and other noncognitive factors play?  Emotions provide effective stopping rules.  Imitation and social learning enable fast learning.

4.What is the role of culture in bounded rationality?  Cultural systems do not need to be correct to work, e.g. navigation by the stars.

Gigerenzer and Selten conclude the chapter with a discussion of interdisciplinarity as vital to bounded rationality.  They suggest a shift from discipline oriented research to more problem focused research.  They point out the sunk cost fallacy as evidence of the insularity of disciplines.  An example of the sunk cost fallacy is where you have a choice to invest in A or B, where B is the more promising option; however, because you have previously invested in A, you choose A.  Hundreds of papers were written in psychology and economics about this while similarly hundreds were written about the Concorde fallacy in evolutionary biology.  These are the same fallacy, but for many years there was no cross reference nor realization that the fields came to opposite conclusions.

Gigerenzer, G. & Selten, R.(1999) Rethinking Rationality, pp 1-11. Bounded Rationality, Cambridge, MA:  MIT Press.