Defining the black-box system allows three key considerations which allows one to differentiate between evolutionary and conventional risk modeling designs.
Design consideration # 1: A tractable inverse model. If there is a tractable “inverse model” of the black-box system, then there is a way of deducing in advance a sequence of variations which bring about a desired set of objective values and risk acceptance modality. “Conventional” methods may be applied. Evolution’s blind generate-and-test nature is not needed, although evolutionary algorithms are quite competitive.
Design consideration #2: The inverse model is not tractable, but the forward model is: While one may predict the influence of variations upon the objective values / risk acceptance but the black-box system is not tractably invertible, one cannot derive and deduce in advance a sequence of variations to bring about a desired set of objective values. This implies an iterative approach, where variations are carefully selected according to the forward model and are applied in sequence. Interative design-and-test is a common component of traditional approaches. Search techniques, including evolutionary algorithms, can be competitive or the only choice.
Design consideration #3: Neither inverse or forward models are tractable. There is neither a way of discerning which variations will give improvements in the objective values, nor a way of predicting what will be the effects of variations on the objective values. By a hesitantly settled definition, evolutionary risk processes are those which proceed by incrementally applying variations which are essentially blind. Selection can lead to an improvement in objective values with neither a forward nor an inverse model. Whether evolutionary methods actually succeed in finding a satisfactory design is another issue, but these are the only way to go in general. Thus, there is an entire class of risks which can only be tackled by evolutionary methods.
