Genetic algorithms only work in an environment designed for supporting the process, including coddling less-fit members of the population. They're barely comparable to real natural selection.
Your objection is barely coherent. A genetic algorithm is an abstraction of the natural process of evolution, and they are very much comparable. In fact, they are rather analogous. To claim that studying genetic algorithms tells us little to nothing about natural selection in nature is preposterous.
I think he's simply saying that genetic algorithms may be based on nature, but come nowhere close in complexity, as to serve as some sort of rigorous "proof" that our theories are true, as the parent suggested.
Put another way, genetic algorithms are derived from our theories about nature, not the other way around. That they can effectively model solution sets to (much simpler) problems that we identify or create, neither proves nor disproves our theories about nature.
What genetic algorithms do is validate the process of evolution and deriving complexity through unguided processes. While this doesn't "prove" biological evolution, it shows that there is nothing inherently impossible about greater complexity being derived from lesser complexity. I'd say this is a very important result.
>What genetic algorithms do is validate the process of evolution and deriving complexity through unguided processes.
No. They absolutely do nothing of the sort.
Do you truly understand how GAs work? There is nothing "unguided" about them. They are pre-programmed, with crossover functions, fitness measures, etc. As the OP mentioned, they operate within relatively simple, well-defined "environments". They are effectively used as a heuristic tool for searching solution sets to defined problems.
Also, no one claimed that it's inherently impossible to derive greater complexity from lesser complexity. That's a straw man. Still, just because it's possible doesn't mean it has actually happened or has any bearing whatsoever on the facts (actual evolution). But, beyond that, GAs are not purposed with "deriving complexity", nor is that a desirable result in their application. The goal is not to produce more complex offspring, but more suitable solutions.
>They are pre-programmed, with crossover functions, fitness measures, etc.
And these are all abstractions of the process of evolution in general, each one being analogous to genetic mutations, sex or gene sharing, fitness within an environment, etc.
>Also, no one claimed that it's inherently impossible to derive greater complexity from lesser complexity. That's a straw man
Plenty of people claim this; I never attributed that statement to you.
>GAs are not purposed with "deriving complexity", nor is that a desirable result in their application.
The "goal" is irrelevant to the question I'm addressing. Many people claim that evolution is impossible because complexity cannot be derived from "nothing". Genetic algorithms disprove that claim. The environment of a GA being artificial is irrelevant. Some GA's do in fact create complexity that even we cannot comprehend, I am reminded of a GA that developed a circuit design (for a DAC perhaps it was) that is completely incomprehensible, yet it works perfectly. This process validates the principle of evolution.
