I may have misunderstood you, but anyways: Developing AI through evolution is pretty common in AI nowadays. Using genetic algorithms (GA) to evolve parameters for other AI solutions, for instance the weights in an artificial neural network (ANN).

Just finished a project where we did exactly that. A simple ANN with feedback (memory) where the different weights between nodes, gain, bias etc. were trained through GA. After a few generations intelligent behavior started to emerge.