[Doctor Stochastic]

The SA article was interesting but not very informative. Injection of random choices into computation is good way of exploring regions of interest (in the space covered by the computation) that would not be obvious. (This gives me an idea about how to do this much better.)

Random proposes, selection disposes.

[js1138]

I am not claiming this is the be-all, end-all of computing. It does demonstrate that selection can build things in unanticipated ways. You could, of course, argue that the "things" were already there, like the statue in the stone.

But isn't that a given? Or how do you determine that something is really new is everything is "in the space covered by the computation"?

It seems to me that in traditional computational algorithms, you follow a predictable, repeatable set of rules, and assume that the rules guarantee the correctness of the result.

In genetic programming, the transformational rules are irrelevant (more on this). Only the fitness function matters. (The rules for making attempts do matter, but only to the extent that they speed up the process. Can anyone say for certain that the rules for biochemistry aren't also rigged? I'm pretty sure that evolutionists are concerned with process, not with how the rules originated.)