What does the genetic algorithm do AFTER it sees the randomness is broken?
I have no problem to see mutation produces some anomalies. But how do you program on what would take place after that?
I don't understand your question. And I'm a professional software engineer who actually implemented genetic algorithms.
What you do when using this algorithm is work towards a pre-determined goal. But that goal is not a specific system. That goal is a certain outcome, an achievement. And it is based on parameters.
Take a very simplistic textbook example: the traveling salesman.
The salesman needs to visit 10 location. He could visit all off them in a wide variety of orders.
We can impose parameters on his journey. For example, we want to shortest route. Or we want the fastest route. Or, for some reason, we might even want to most innefficient route.
Using a genetic algorithm for this would test against those parameters. You'ld have a population of "possible" routes and calculate the fitness of each individual route - with respect to the parameters.
If we are trying to get to the fastest route, we'ld calculate completion time and continue with the fastest individuals.
When we are trying to get to the most fuel efficient route, we'ld calculate that, etc.
And based on that "fitness", we'll choose the best performing (again:
best performing as relative to the parameters) to reproduce, mutate and restart in a new generation.
And we'll inevitably end up with systems that perform well and
seem designed to do exactly what they do. But they aren't. Instead, they are
filtered based on their performance. And because of the filter, they get better at it with every passing generation.
Here's a fun example of such an implementation...
http://boxcar2d.com/
Just click it open and leave your browser open for a few hours (it might be required to leave it as the active window as well - not sure if it continues to run in the background, some browser might block it). After that, you'll see cars driving quite happily over the track, while at first you only had shapes with perhaps more then 1 wheel that breaks down 1 second after it appears on screen.
The goal of that GA is to evolve "cars" that perform well on the track.
The first generation is fully random. It's just a bunch of polygons with random wheels in random places, with a random torque and spring tension, etc.
Generation after generation, small random and unguided mutations are introduced. A fitness function then selects best performing cars.
