|
|
|
|
|
|
|
However, in all but the most constrained situations, enumerative plans are a false lead. |
|
|
|
|
|
|
|
|
The flaw, and it is a fatal one, asserts itself when we begin to ask, "How long is eventually?" To get some feeling for the answer we need only look back at the first example. For that very restricted system there were 10100 structures in  . In most cases of real interest, the number of possible structures vastly exceeds this number, and for natural systems like the genetic systems we have already seen that numbers like 210,000@ 103000 arise. If 1012 structures could be tried every second (the fastest computers proposed to date could not even add at this rate), it would take a year to test about 3·1019 structures, or a time vastly exceeding the estimated age of the universe to test 10100 structures. |
|
|
|
|
|
|
|
|
It is clear that an attempt to adapt by means of an enumerative plan is foredoomed in all but the simplest cases because of the enormous times involved. This extreme inefficiency makes enumerative plans uninteresting either as hypotheses about natural processes or as algorithms for artificial systems. It follows at once that an adaptive plan cannot be considered good simply because it will eventually produce fit structures for the environments confronting it; it must do so in a reasonable time span. What a "reasonable time span" is depends strongly on the environments (problems) under consideration, but in no case will it be a time large with respect to the age of the universe. This question of efficiency or "reasonable time span" is the pivotal point of the most serious contemporary challenge to evolutionary theory: Are the known genetic operators sufficient to account for the changes observed in the alloted geological intervals? There is of course evidence for the existence of adaptive plans much more efficient than enumeration. Arthur Samuel (1959) has written a computer program which learned to play tournament calibre checkers, and humans do manage to adapt to very complex environments in times considerably less than a century. It follows that a major part of any study of the adaptive process must be the discovery of factors which provide efficiency while retaining the "universality" (robustness) of enumeration. It does not take analysis to see that an enumerative plan is inefficient just because it always generates structures in the same order, regardless of the outcome of tests on those structures. The way to improvement lies in avoiding this constraint. |
|
|
|
|
|
|
|
|
The foregoing points up again the critical nature of the adaptive plan's initial uncertainty about its environment, and the central role of the procedures it uses to store and access the history of its interactions with that environment. Since different structures perform differently in different environments, the plan's task is set by the aspects of the environment which are unknown to it initially. It must |
|
|
|
|
|