D'Abrera on Science and Philosophy

[Kenelm Philip]

	John Grehan said:

> ...but as for the initial resemblance there is the problem of not being
> able to select what is not already present.

Many people tend to underestimate the sheer _power_ of natural selection.
Some results from the field of a-life (so-called 'artificial life' on com-
puters) may thus be of interest. Some years ago, a-life workers implemented
natural selection on computers as a way to construct programs. For instance,
random strings of code were subjected to a test of their ability to sort
numbers (a standard task for computer programs) and then 'mated', using
crossover and a bit of mutation, with their reproduction keyed to their
success at sorting. In 15 minutes of computer time, a program for sorting
numbers evolved that nearly tied the best sorting algorithm any programmer
had ever written, requiring 61 exchanges to sort 10 numbers, compared to
60 for the best man-made program. Something out of nothing indeed! Further-
more, programs produced by this 'genetic algorithm' (GA) are much more
robust than programs written by programmers.

	It is also of interest to note that further work with GAs has
indicated that the importance of mutation may have been overestimated. A
great deal can be accomplished with the recombinations of existing genetic
material. Evolution by punctuated equilbrium has also arisen in these
experiments--with the added feature that the entire genome is available for
analysis throughout. Finally, this work has shown the crucial role played
by predators/parasites in allowing populations to escape being trapped
on local maxima of fitness, and thus access still higher maxima lying
across 'valleys' that could not be crossed without the 'aid' of the effect
of predators on the genome.

	Computer programs have long been regarded as something along the
lines of Paley's watch (a watch implies a watchmaker--a program implies
a programmer). Not so--powerful, complex, and functional (and robust)
programs can be _evolved_ from strings of random code by no more than
a process modelled after natural selection. And in far less time than would
be needed for a programmer to write the code, owing to the fact that a
fast computer can run through many thousands of 'generations' of the 
process in a few minutes.

	So the statement that "Natural selection could _never_ do _that_!"
is a bit suspect these days...

							Ken Philip
fnkwp at aurora.alaska.edu