hehe. That's fine. It all depends on the nature of the experiment. Typically you want to optimize a solution in a given deterministic domain. If you wish to, instead, create an open-ended evolutionary system for novelty in a non-deterministic solution space, go for it!
Although, to be fair, in nature you're also not dealing with a static fitness function, a high mutation rate, and there's a wide myriad of other variables that affect the overall dynamics of a population (like sexual vs asexual reproduction, migration, recessive genes, sexual selection, etc)
You're absolutely right. I've been wondering what results i would get if I seeded all of the population with the reference image data, slightly mutated but let the GA look for variations with a 10 or 20% variation tolerance.
Although, to be fair, in nature you're also not dealing with a static fitness function, a high mutation rate, and there's a wide myriad of other variables that affect the overall dynamics of a population (like sexual vs asexual reproduction, migration, recessive genes, sexual selection, etc)