BoundedRationality_TheAdaptiveToolbox.pdf

170 Laura Martignon
construction mechanisms that guarantee high accuracy. The Bayesian paradigm
discussed in this chapter provides efficient search methods for good classification
trees.
It is amazing to note that both the Bayesian toolbox and adaptive toolbox
meet in the end, adopting, at least in the final steps (see Figure 9.3), very similar
inference machines, namely trees. As I have pointed out, being or not being a
Bayesian is a matter not only of information formats but of memory and retrieval
capacity. Nothing is more psychologically plausible than the Bayesian approach,
when natural frequencies (of the correct cue configurations) are either
provided or retrieved from memory. All other models are surrogates for the profile
memorization method, not just when training and test set differ, but also
when fitting known data. Once being a natural Bayesian becomes impossible,
models make their entrance. They can be fast and frugal models (e.g., the Take
The Best tree) or amount to simple trees searched by means of laborious implementations
of Occam's Razor.
The normative and the adaptive toolboxes came to exist because of our limited
memory, limited retrieval capacity, and the need to make predictions on unknown
data. Both toolboxes have a bias toward simplicity. The normative
toolbox uses complex computations to become simple while the adaptive toolbox
became simple by adaptively exploiting the structures of environments
through the millennia.
ACKNOWLEDGMENT
I am very grateful to Robin Hogarth for his helpful comments and advice.
REFERENCES
Akaike, H. 1973. Information theory and an extension of the maximum likelihood
principle. In: Second International Symposium on Information Theory, ed. B.N.
Petrov and F. Csaki, pp. 267-281. Budapest: Akademiai Kiado.
Albers, W. 1997. Foundations of a theory of prominence in the decimal system. Working
Papers Nos. 265—271. Institute of Mathematical Economics. Bielefeld, Germany:
Univ. of Bielefeld.
Breiman, L., J.H. Friedman, R.A. Olshen, and C.J. Stone. 1993. Classification and
Regression Trees. New York: Chapman and Hall.
Cooper, G. 1990. The computational complexity of probabilistic inferences. Artif. Intel.
42:393-405.
Cover, T.M., and J. A. Thomas. 1991. Elements of Information Theory. New York: Wiley.
Czerlinski, J., G. Gigerenzer, and D.G. Goldstein. 1999. How good are simple heuristics?
In: Simple Heuristics That Make Us Smart, G. Gigerenzer, P.M. Todd, and the ABC
Research Group, pp. 97—118. New York: Oxford Univ. Press.
Dawes, R.M. 1979. The robust beauty of improper linear models in decision making.^m.
Psych. 34:571-582.