I find it amusing that work on genetic algorithms is critiqued
as not providing a complete simulation of all the nuances of
biological reality when the same critics apparently wish us to
accept Schutzenberger's even less nuanced simulations as
relevant. The irony there is precious.
Of course, if it hadn't been for Bradbury, I might have missed
out on that dose of irony completely. I've put Jones in an
"archive-unread" filter, as well as "DNAunion".
The quote with Waddington saying that he wasn't interested in
Schutzenberger's computers is often quoted out of context.
In isolation, it makes it look like some distressed non-techie
biologist is simply frustrated with an elegant argument given
by Schutzenberger. In fact, though, Waddington had another and
better reason for being short with Schutzenberger, which was that
Schutzenberger was digressing away from a concrete biological
counter-example given by Lewontin. Waddington's interjection
comes at Schutzenberger's fifth "reply" to Lewontin, which
nonetheless failed to answer Lewontin's direct question.
Dr. Alex Fraser contrasted Schutzenberger's simulation scenario,
which did not work, with a version that did work.
[Quote]
Dr. Fraser: Can I contrast one computer with another? You
have a computer programmed to examine the statement, "All I am
allowed to do is change letters and I hope I produce a
program. Any kind of program will do." This doesn't work.
We now turn around and set up another computer, and we tell it
a basic genetic system of plus-minus alleles in which we are
saying, "Can it produce information?" The decision on whether
the information is useful will be a selective one of "survive
or not survive." This is the same kind of decision-making;
the programs look very similar to those which are being
constructed to try to produce information-containing programs.
The principles are very similar.
However, in the genetic one, the system is that there are
multiplicities of pathways to suitable answers. The machine
can gradually, step by step, get there; each step takes it
toward the answers, and it produces them when all we have fed
into the machince is a genetic system of essentially complete
simplicity. What is surprising is how fast rational information
is produced by the machine within the meaning of the original
context.
So, if you are going to take a program space and say, "We cannot
transform it," but leave out of it the means of combination
and recombination in between and of evolution by selection,
I am certain that your program will not produce sense; but if
you put it in there the machine gets there so fast it is
surprising.
[End Quote - Dr. A Fraser, discussion of Schutzenberger's paper,
"Mathematical challenges to the Neo-Darwinian interpretation of
evolution", p.80]
Wesley
This archive was generated by hypermail 2b29 : Tue Oct 17 2000 - 04:35:45 EDT