Re: A mistake, A defense and a change of address

Bill Hamilton (whamilto@mich.com)
Sat, 2 Mar 1996 11:08:02 -0500

At 8:46 AM 3/2/96, Glenn Morton wrote:
>Now, you say you do not want to go back to the dark days of chance. You use
>chance in a very funny way. Self-organizational systems are operated by
>stochastic means e.g. Sierpinski's Gasket and I believe evolution. A
>mutation is effected by chance. Then the brutal requirement that the system
>be functional is applied and only certain directions in sequence space are
>allowed. Similarly, in the sequence space of the DNA of living systems, only
>certain pathways are allowable due to the brutal requirement that the system
>live. Thus in my view, God was the one who designed the pathways in the
>sequence space. Like a rat in a maze, living systems can only travel the
>sequence-space paths which God designed into the fabric of the universe.
>Thus while chance drives the mutation, design drives the direction allowed by
>living systems.

Mutations are not the only way in which chance enters into evolution.
Furthermore, some recent experience convinces me (empirically from a
simulated world of course) that it is not the most fruitful way in which
chance enters. I have been using genetic algorithms to study some piston
design problems for controllable dampers. In a genetic algorithm chance
enters in two ways: crossover and mutation. In crossover, a random point
is chosen on a genome selected for reproduction and two genomes are "mated"
by taking the segment of one below the randomly selected crossover point
and combining it with the segment of the other gene above the crossover
point. Two offspring can be created in this way. I don't knw whether
anything exactly like this takes place in the real world, but it
corresponds (very roughly) to things like conjugation in some simple
critters and even more roughly to the role of sex in more complex animals.
The idea is to identify strings of genetic material that contribute to
success in meeting the survival criteria. My experience to date seems to
indicate that crossover probabilities are something you can experiment
with, and that sometimes increasing them to 80% or so can be very helpful.
But increasing mutation rates above a couple percent almost never helps,
and usually hurts.

The point is that randomness can be useful in evolution, but it must be a
sort of randomness that puts together successful substructures.

------------------------------------------------------------------------------
Bill and Linda Hamilton
1346 W. Fairview Lane
Rochester, MI 48306
(810) 652 4148