Information Technology Reference
In-Depth Information
Note again that, like RNA, proteins can function simultaneously as geno-
type and phenotype. Note, however, that despite the richer functional diver-
sity, such systems are also equally constrained: any modification in the
replicator is immediately reflected in its performance.
In theoretical terms, the differences between DNA, RNA, and proteins are
most useful to help understand the fundamental differences between GAs,
GP and GEP. Both GAs and GP are simple replicator systems, using only one
kind of entity: linear strings of 0's and 1's in the case of GAs, and complex
ramified structures composed of several different elements in the case of GP.
Many believe that a simple “RNA world” existed in the early history of life,
perhaps contemporary to a simple “protein world”. RNA and proteins some-
how started working together, recruiting also DNA. The complex DNA/pro-
tein system of life on Earth is the descendant of this evolutionary process.
It is surprising that computer scientists some 4 billion years after these
events took pretty much the same steps of life on Earth, first inventing sim-
ple replicator systems and only later inventing sophisticated replicator/phe-
notype systems. The genetic algorithm invented by Holland in the 60's (Hol-
land 1975) is analogous to a simple RNA replicator with its linear chromo-
somes and limited functionality, whereas the algorithm popularized by Koza
(1992) is analogous to a simple protein replicator with its richer functional-
ity. Curiously enough, the conscious attempts to create a genotype/pheno-
type system, despite trying very hard to emulate the DNA/protein system,
are far from being the desired leap forward (Banzhaf 1994, Ryan et al. 1998).
On the other hand, the full-fledged genotype/phenotype system of gene
expression programming was invented in 1999 by myself (Ferreira 2001),
totally unaware of all the hard work done by other researchers to create a
genotype/phenotype system. In fact, I first heard of GP in Mitchell's topic
(Mitchell 1996) and was so impressed that I tried to make a GP on my own.
I suppose I just applied what I knew from biochemistry and evolution and,
therefore, it never crossed my mind to make a system without an autono-
mous genome. Obviously, the complicated things of information metabo-
lism were discarded as they are irrelevant to a computer system where the
rules are not dictated by chemistry. Consequently, double-stranded chromo-
somes, RNA-like intermediates, and complicated genetic codes with com-
plicated translation mechanisms did not make their way into gene expres-
sion programming. Furthermore, I also knew that for a genotype/phenotype
machine to run smoothly, the genetic operators could not be constrained and
they should always produce valid structures. And the result was the first
