Good evening, ladies and gentlemen...
Bang Goes the Theory had a segment using the repeated freehand tracing of a single line to illustrate some of the basic mechanisms of evolution. It lacked any selective pressure, so I whipped up a program to add some. I started with ten perfectly straight lines, each of which had ten offspring with a small chance of containing replication errors (1% to shift each pixel one place right or left). From that pool of one hundred, I then chose the ten that most resembled my target shape and repeated the process. To keep it simple, I didn't require the line be continuous (in fact, it's not technically a line as there's only one pixel for each vertical position on the image). I used the sum of the horizontal distance squared for each pixel from it's target x value as a measure of fitness. You may recognize the final shape, which took 2,200 generations to settle upon. Its efficiency could be increased dramatically as the algorithm employed was crude.
This is a lot less involved (asexual reproduction for a start) than many genetic algorithms I've written, but it's much less abstract.
