W. B. Langdon and R. Poli
You ask "is natural evolution a search technique? If so, what is it
searching for?" I won't attempt to answer these questions directly.
However I think it is instructive to compare natural evolution with
an application of evolutionary algorithms such as my own field
- GA optimization of aircraft wings.
In GA aircraft wing optimization individuals compete for the
use of finite resources e.g. cpu time, memory. This competition is
based on the relative merits of their phenotypes, e.g. drag,
weight. The winners propagate their genes into the next generation.
In natural evolution individuals compete for the use of finite
resources e.g. food, shelter. This competition is based on the
relative merits of their phenotypes, e.g. running speed, antler
size. The winners propagate their genes into the next generation.
It is not clear to me that there is any conceptual difference between
the two processes, and that the second is any less of a search than the
first. To answer the "what is the search for?" question requires
analysis of the laws and rules of the environment in which the
individuals evolve. In the case of wing optimization it would be
possible for a third party to figure out that the environment was
configured to search for commercially successful wing designs.
Deduction of a search objective by analysis of the laws of the natural
environment is a much more complex problem whose solution is the
rightful domain of philosophers and theologians.
I am also interested in your comments on the evolutionary pressures on
the amount of genetic material. Almost 100% of the DNA of bacteria and
viruses is expressed. Presumably this indicates the presence of strong
evolutionary pressure on genome length. However, in eukaryotes the
picture is different. The majority of DNA is not expressed and there
are huge differences in genome length which are not reflected in the
apparent complexity of the organism. For example:
Genome size %coding for
(x10^9 base pairs) protein
Homo Sapiens 3.5 9-27
Tristurus cristatus 19 1.5-4.5
(a newt)
Protopterus aethiopicus 142 0.4-1.2
(a lungfish)
This might suggest that the evolutionary burden of the 'excess' DNA
of plants and animals is small. Most eukaryotic species are therefore
probably not at a particular neutral point of length v cost. It is
interesting to speculate as to how much of the differences in the
selection pressures on genome size are due to differences in rates
of reproduction between prokaryotes and eukaryotes.
Presumably most evolutionary algorithms are likely to be more like
the eukaryotic case in that the cost in cpu time and memory of
maintaining the additional genetic material will usually be small
relative to the cost of evaluating the objective function. If so,
should bloat be supressed or encouraged?
Dr Gordon Robinson
Evolutionary Optimization Group
University of Southampton
United Kingdom