The genetic programming may work better for creating a strong AI than to solve that kind of problem. The intelligence did evolve, heh, heh.

If you are trying to evolve a program that solves something of this kind, you end up with an entirely wrong algorithm, constants in which are going to evolve to make the best outcome there can be. The solution just gets stuck in local maximum. Like human eye that has retina backwards. It is more productive to try to iterate over possible solutions starting from the simplest (to implement occam's razor), using some separate process (hill climbing from several random points?) to tune the constants to their best.