Simulation and the Search for Stability in Design

Design problems, if they can be called problems at all, are complex. Disciplines such as architecture and planning are described as dealing almost exclusively with situa-tions in which the brief is relatively ill -defined relative to the real range of problem considerations, the perception of the problem itself may change radically as design progresses, and the solution is typically arrived at by a unique process that cannot be predicted in advance. Even where the situation can be clearly specified, complexity exists of a more formal nature. Such systems are notoriously sensitive to initial conditions, and any in-accuracy in modelling them can accumulate to make exact prediction of their behaviour impossible. Where the system‟s possible future trajectories diverge in this way, it is called chaotic. The weather is perhaps the para-digmatic example. If the variables that diverge in this way are relevant to the goals of the designer—that is, if the success of the proposal rests on the prediction of the system being even approximately accurate—then design becomes impossible.

Thankfully, many complex systems exhibit stability, sometimes in particular regions of their state space, and sometimes at a higher level of ab-straction. In agent based models of economic behaviour, the individual agents‟ matter very little, but the structure of their interactions constrains the behaviour of the entire system. In models of bounded rationality, behaviour of a single agent will be entirely unpredictable at any given moment, while the behaviour of the group as a whole will in the long term converge to fluctuate around a stable, predictable outcome. At this higher level, and because of the system‟s structure, certain measures that describe the behaviour of the system are invariant with respect to changes in any of the individuals. This is highly relevant to design because designers typically do not need to work toward the precise behaviour of an individual at a precise moment in time, but for a varied group and over a relatively long period. Recalling the example of urban planning makes this obvious.

This position paper proposes that all design is aimed at stable regions within a system‟s state space. These basins of attraction, in which the per-formance of the system is invariant to small changes in conditions or a failure in collecting precise data for the model, are in fact the only possible areas in which design is possible at all. What is suggested here is that the problem with the statistical approach is that it simply assumes the structure of the system and relevant variables in advance. With organised complex systems this must be carefully investigated to find those levels and regions in which prediction might be possible. It is the task of the designer to do this.