Welcome to Gaia! ::

Oooh, I've seen some other stuff like this, as well.

On the discovery channel a long time ago, I think it was...
A program was designed for a 'shepherd' robot that had the ability to control the direction of a 'sheep' robot in a few limited ways. But to design the program for the shepherd robot, they made a simulation which mapped out the success of each generation, and from each simulation, a 1 out of a few mutated versions of the previous version (the most successful one) was then used as the basis for new mutated forms in the next generation...
Anyway, after leaving this evolution program on overnight, they ended up with an extremely effective program, with varying responses to the different directions of the 'sheep', starting with one that could barely do the job.

I've also heard of a situation where some people used a similar program to design the body of an entirely new robot. The selective factor was that robots that could move well would 'survive' and go onto the next round...
They ended up with sortof a moving tripod thing.

Anyway, something that I think is very interesting, (and can demonstrate the principle quite well), is biomorphs.
I think that it might be a useful tool for exhibiting the difference between evolution and random mutation, because if you just click on a random "child" each time, then you end up with an organism that changes randomly. One generation it might be long and thin, 20 gnerations it might be short and slanted, 20 generations later it might be large and straight.
However, if you're a little selective in which generation 'survives', you can get some very rapid evolution- but of course the speed of the evolution usually depends on how selective you are; because if you always choose the largest one to 'survive', you will soon end up with a huge organism. If you tend to pick larger ones most of the time, you still end up with evolution in that direction, but at a slower rate.