A small subset of all programs are sensitive to their inputs (i.e., their output changes as their input changes). Most programs are not sensitive to their inputs. Clearly, these static programs will have low fitness and therefore are simply a drain on the evolutionary system. Any operator which can reduce the percentage of its output programs that have this undesirable aspect has a potentially (see below) desirable trait. The static program production difference shown in figure 3.5 is an artifact, a side-effect of the operator fitness measure. The notion of a static program is absent from the actual operator fitness calculation.
Figure 3.5: Percent of the Population that gives a static response
Like any measure of SMART operator performance, this measure is not perfect. It is possible that an operator exists such that nine times out of ten it produces a static individual and the tenth time it produces an extremely high fitness program. Another possible situation is a SMART recombination operator that always produces one static child and one very highly fit child. Such hypothetical operators aside, reducing the number of static programs created seems safely to be an advantage for a genetic programming system.