As for GA, an even more symbiotic possibility exists. Recall in GA we start
with a population of designs. Then we evaluate these designs to determine
their performances. We then choose to reproduce, cross breed, and mutate the
"good" designs to produce successive generations of populations
which converge to the best design in the sense of genetic breeding of a
"master" race or population. Note GA calls for an evaluation of a
population of performances. This can be costly or infeasible. OO, on the
other hand, says that this evaluation can be done with surrogate models if we
are only interested in the separation of "good" from the
"bad". Thus OO should extend the usefulness of GA to more complex
problems. Conversely, we have so far only described the use of OO for
narrowing down one step search. Surely, we can learn from each search step and
sample evaluation. Our search can be gradually refined and iterated. Here GA
can be of great help. Later on we shall describe one such instance of an
iterated search. But only the surface has been scratched.