In the 1960s, researchers in artificial intelligence took three approaches: simulation of evolution and life, simulation of networks of neurons, and algorithms to perform intelligent tasks. The MIT and Stanford AI labs specialized in the last of these, and to maximize the availability of grant money, they managed to kill off the funding of the first two approaches.
In the 1980s, the study of neural nets was reborn in work by the physicist John Hopfield, and the study quickly spread into the computer science community (the MIT AI lab had long since lost the clout to do anything about that). Genetic algorithms also came back into vogue, and in 1993 Karl Sims wrote a delightful paper in SIGGRAPH about using simulated evolution to create art.
Sims’ program generated random formulas(the genome) specifying the pixel values of an image. These could be mutated and cross bred, and the user can specify which image he thinks are beautiful. The program required super computers then, but it was also written in an inefficient language (LISP). Today PCs are fast enough, and I wrote a simple image evolver in C++ which compiles directly into efficient machine code.

It’s surprising how rapidly one finds an interesting image, even though the first generation of random images are usually very boring. It has been noted that cross breeding is extremely important. Evolution by just mutation and selection proceeds very slowly, but when several survivors are able to cross breed, evolution converges very rapidly on complex forms.

Here are some ancestors of that image, showing how its complex forms began to develop and evolve.
The accelerating effect of cross breeding is suggestive about evolution in the biological world. The evolution of sexual reproduction in multicellular life forms may have sparked a sudden leap forward in development, perhaps even the so-called cambrian explosion of diverse animal forms 600 million years ago.