12. What optimizes genetic algorithms?
Computational methods often employ genetic algorithms (GAs). The appeal of GAs is that they are modeled after biological evolution. The latter is the main motivation for tolerating such an inefficient awkward process. The GA search technique begins with a large random pool of representations of “potential solutions.” Genetic algorithms are seen as a subset of evolutionary algorithms and as “evolutionary computation.” The methodology is inspired by modeling a random beginning phase space, various kinds of mutations, inheritance and selection. The experimenter chooses the fittest solutions from each generation out of the “evolving” phase space of potential solutions. The goal of the process is optimization of a certain function.
All too many evolutionary computationists fail to realize the purely formal nature of GA procedures. GAs are not dealing with physicodynamic cause-and-effect chains. First, what is being optimized is a formal representation of meaning and function. A representation of any kind cannot be reduced to inanimate physicality. Second, “potential solutions” are formal, not merely physical entities. Third, at each iteration (generation) a certain portion of the population of potential solutions is deliberately selected by the agent experimenter (artificial selection) to “breed” a new generation. The optimized solution was purposefully pursued at each iteration. The overall process was entirely goaldirected (formal). Real evolution has no goal [refs.]. Fourth, a formal fitness function is used to define and measure the fittest solutions thus far to a certain formal problem. The act of defining and measuring, along with just about everything else in the GA procedure, is altogether formal, not physical [refs.].
Despite the appealing similarities of terms like “chromosomes”, GAs have no relevance whatsoever to molecular evolution or gene emergence. Inanimate nature cannot define a fitness function over measures of the quality of representations of solutions. GAs are no model at all of natural process. GAs are nothing more than multiple layers of abstract conceptual engineering. Like language, we may start with a random phase space of alphabetical symbols. But no meaning or function results without deliberate and purposeful selection of letters out of that random phase space.
No abiotic primordial physicodynamic environment could have exercised such programming prowess. Neither physics nor chemistry can dictate formal optimization, any more than physicality itself generates the formal study of physicality. Human epistemological pursuits are formal enterprises of agent minds. Natural process GAs have not been observed to exist. The GAs of living organisms are just metaphysically presupposed to have originated through natural process. We can liberally employ GAs and so-called evolutionary algorithms for all sorts of productive tasks. But GAs cannot be used to model spontaneous life origin through natural process because GAs are formal.
As I understand GAs, this is misleading. "The optimized solution" implies that there is a single optimal solution that is known to the searcher beforehand. As PGS describes them, though, and from other things I've read, that's not true. The searcher establishes parameters for what will constitute a successful solution, but doesn't know exactly what that solution will be until the GA arrives at it.
He also says "The appeal of GAs is that they are modeled after biological evolution. The latter is the main motivation for tolerating such an inefficient awkward process." From PGS's description, they're not inefficient or awkward at all--they're actually more efficient than trying to arrive at a solution through a completely directed process, because they can explore the entire solution space at the same time.
The rest of that excerpt seems to be mostly bald assertions. For one, I don't see why we can't learn about physical processes from formal representations. He puts it in italics, but it's still just his assertion.