Re: Irreducible complexity analogsIsn't trial and error what evolution is doing

Bill Hamilton (whamilto@mich.com)
Sat, 30 Nov 1996 23:49:14 -0500

At 11:52 AM 11/30/96, Jim Bell wrote:
>Glenn asks:
>
><<Isn't trial and error what evolution is doing with mutations? Natural
>selection is what distinguishes the error.>>
>
>That's a good question. And you're right. But the difference between this k=
ind
>of trial and error, and the human kind Bill described, is that the human mo=
del
>has a goal in mind, a telos, intelligently selected. And the steps chosen a=
re
>not (mostly) "mutations" (sometimes they are, as in 3-M's Post-it Notes).
>Usually, they are chosen because someone thinks they'll work. Then they are
>fine tuned, etc.
>
Perhaps I've unduly muddied the waters by proposing an example in which
there are human agents. The point I have attemped to make is that the
overall development process itself, and its evolution through time were not
planned. In fact, some of the intelligent agents acting within the system
actually were working against one another -- not deliberately, but because
they simply didn't communicate. Jim's use of the Post-it as an example of
mutation suggests that perhaps he still has not grasped what I am trying to
say. He is focussing on product, rather than process. Perhaps a large
compnay example will help. GM has a development process called "Four
Phase" which it endeavors to apply to all new product developments. Four
Phase details the development stages any new product must pass through, and
specifies "gates" -- points at which certain approvals are necessary to
proceed to the next stage. Four Phase is not a static plan for developing
new products. It is constantly being updated to take advantage of new
production planning and control methodologies, to reduce development time,
and to meet a host of other objectives that are moving targets. This
development of the development process itself is not haphazard. Committees
oversee it and make conscoius decisons about what changes to recommend.
Among the criteria for making changes are those aimed at minimizing
conflicts as changes are introduced. Moog did not have a process for the
orderly evolution of its development process. And it suffered for that
lack.

Jim may be saying that even though the development of the development
process was not overseen by humans, that somehow it was implementing a
teleology imposed from outside. That may be, but the difficulty would be
in showing that such was the case -- which is exactly the difficulty we
have with teleology in biology.

>Your view of evolutionary trial and error is certainly not the sort of TE
>Terry Gray has explained. Are you saying that God, rather than directing
>change, is actually watching the mutations make errors to be selected?

Perhaps the term "errors" implies value judgments we oughtn't to be
including in our discussions, because they introduce an unnecessary and
unproductive emotional content. As a control theorist I could as easily
say that God had deliberately made the genetic code capable of producing a
sufficiently rich variety of mutations as to deal with any disturbance
imposed on the system by outside agencies (aka secondary causes). I would
also suggest that God no doubt introduces variations that suite his
purposes when He chooses.

One final point: We frequently talk about mutation, but mutation is not
the only mechanism of evolution. Many genes have large numbers of alleles,
and the number of possible combinations of alleles in a given genome can be
astronomical. Just the normal process of sexual reproduction can produce a
great deal of variation, which can contribute to evolutionary development.
If I'm allowed to use genetic algorithms as an example (okay with Glenn,
not okay with Jim, I presume) I might add that high mutation rates are
usually not productive. I've used genetic algorithms for optimization of
piston designs for automotive shock absorbers with vario=A8s kinds of smart
fluids, and I have found that mutation rates must be kept quite small or
the solution just jumps around and never converges to a solution.
Haldane's dilemma operates in the genetic algorithm world as it does in the
real world. But the point is we can get considerable progress toward
global optimal solutions with very small or even zero mutation rates.

Bill Hamilton
----------------------------------------------------------------------------=

--William E. Hamilton, Jr., Ph.D.1346 W. Fairview LaneRochester, MI 48306(810) 652 4148