If we frame the idea of goodness of design in terms of suitability of design, then our question becomes, "Out of all possible designs, why are suitable ones less abundant?" With this idea, goodness of design is broadly quantifiable using covariances of factors between a design and its context.

We find a normal distribution of more or less suitable designs, with most being moderately suitable and relatively few being either exceptionally suitable or exceptionally unsuitable.

Finding a good design generally requires sifting through many inferior designs first. A design may not be readily apparent as good. The number of inferior designs that must be vetted may be impractically large if the candidates are examined randomly.

The fact that there are so many more ways to be wrong than to be right makes it probable that we will be less than right most of the time, if left to search randomly.

Discovering a strategy for reducing the number of inferior designs that must be considered before finding an adequate design is itself a design problem, as is that strategy in turn, and so on regressively.

Interestingly, the further along this regression of strategies of strategies, the less constrained the choices, and therefore the more expensive it becomes to move further toward the exceptional end of the fitness distribution.

By quickly perceiving the relative fitness of highest-level strategies, the human mind ends the regression. Mechanical/software systems are more tightly bound to evaluating more of the search paths.

The mind seems to do this by weighing the probabilities of the high-level strategies based on loose matching of factors to the richness of prior experience. The extraordinary matching power comes from the immense variation of factors compared in such a brief time.

**Michael Webb, 2002**

[ home ]