Well that does sound like a goal driven algorithm to me. Except that you can't actually say if a move made by the algorithm will actually produce a acceptable outcome to the algorithm itself. Due to the sheer number of variables, the only option the algorithm has is to do a kind of A/B testing. To make changes to a smaller subset of subjects and see if they survive, if they do they do. If they don't, well then you can kill that test and move on more modifications.

Evolution is an optimization algorithm, which makes changes, tests and makes more changes/corrects based on feed back.