The psuedo-code you have listed above is the basic principle of a genetic algorithm (GA) but to really go down the GA route you need to initially constuct MANY random solutions as follows...

1. Define a way of encoding your solutions into a data set e.g. Binary, hex, tree...
2. Create 20 random candidate solutions using your encoding.
3. Measure how well each of these work (the 'fitness')
4. Mutate the populations at random.
5. Crossover different candidate solutions. (many different methods can be applied here to choose which to cross)
6. Measure the new fitness of the new 20 candidate colutions.
7. Is this fitness satisfactory? If not go back to step 4 (or stop if you are on the Nth itteration)