I am wondering whether anything in Glenn's post shows "new
information" in the sense of innovative design. I will explain -
although I don't wish to "defend" the Wilder-Smith quote Glenn
provides.
This month's _Scientific American_ has, according to Glenn,
relevant examples.
"Brian Howley of Lockheed Martin Missiles and Space guided
the evolution of a program that can figure out how to manoeuvre
a spacecraft from one orientation to another within 2 percent of
the theoretical minimum time--10 percent faster than a solution
hand-crafted by an expert. And researchers at University College
in Cork, Ireland, grew a system that can convert regular
programs, which execute instructions one at a time, into parallel
programs that carry out some instructions simultaneously."
I want to suggest that these examples are of optimisation, not
the generation of new information. Both relate to procedures,
pathways to achieve a goal, and in both cases the program can be
written to explore a large number of possibilities and find the
best solution.
The quote continues:
"To create their software, Fernandez and Howley did not have
to divine insights into neurophysiology or rocket science. The
task of the genetic programmer is simpler. First, build an
environment that rewards programs that are faster, more accurate
or better by some other measure."
Building an environment is an intelligent activity. The
"measures" used to assign rewards are critical for success - and
choosing appropriate measures is an intelligent activity.
"Second, create a population of seed programs by randomly
combining elements from a "gene pool" of appropriate functions
and program statements. Then sit back and let evolution take its
course. Artificial selection works just like the natural variety:
each program is fed data and then run until it halts or produces
a result. The worst performers in each generation are deleted,
whereas the best reproduce and breed--that is, swap chunks of
code with other attractive programs. Occasionally, a random
mutation changes a variable here or adds a command there."
The phrase "just like" claims too much. The programmer has
determined the nature of the variations that can be introduced -
and this is an intelligent activity. Generally, the more
"intelligence" built in here, the faster the program will run.
The reference to "worst performers" is a second opportunity for
me to point out that the performance criteria have to be selected
intelligently. Sometimes, the criteria are not known except at
the broader, general level - in which case the search space is
vastly enlarged and either the problem exceeds the processing
power of the computer or the time to reach a solution becomes
extremely long.
"The technique can generate solutions even when the
programmers know little about the problem. But there is a price:
the evolved code can be as messy and inscrutable as a squashed
bug. Fernandez's gesture-predicting program consists of a single
line so long that it fills an entire page and contains hundreds
of nested parenthetical expressions. It reveals nothing about why
the thumb moves a certain way--only that it does.
"Just as in the real world, evolution is not necessarily the
fastest process either. Howley's speedy workstation churned for
83 hours to produce a satellite-control program that beat human
ingenuity in eight test cases. And when it was presented with
situations it had never encountered, the program failed, a common
problem with evolved software. (Of course, the human expert's
program failed on the new cases as well.)"~W. Wayt Gibbs,
"Programming with Primordial Ooze", Scientific American cot
1996, pp 48-50
OK. Here is the evidence that an enormous search space is being
explored. The more options programmed into the software, the
longer will be the search. It seems to me that there is a
problem here for neo-Darwinians. The number of replications
required to develop "perfect adaptations" using these programs
is extraordinarily high - 83 hours is a long time for a computer
to search, and it represents a correspondingly high number of
replications! Neo-Darwinism has time - but when it is translated
into the number of replications, it does not have as much time
as it would like. How long do you have to evolve a whale from
a terrestrial mammal? Or a trilobite from a soft-bodied
ancestor? A few million years? Perhaps you have a few million
replications? The computer will work through that in a few
minutes - so maybe the 83 hours is telling us something about the
efficiency of Darwinian processes.
Glenn writes:
"Notice that the evolved programs were better than the
intentionally designed programs. The interesting thing to me
is that in a real sense both types of programs are designed.
The traditional algorithm is well thought out by an intelligent
agent with each part intricately designed. The other is designed
by designing an environment in which solutions to various
problems can be found via random mutation. Design can take
several forms. It does not have to be the traditional form of
design."
Design CAN take several forms. Darwinian design seems to me to
be viable where optimisation is the goal, where the performance
of the finished product can be specified but where there are
numerous variables affecting the final performance, and where
there is relatively little knowledge about the variables.
However, as soon as there is knowledge about these parameters,
the programmer is unlikely not to build this into the program -
it will prune the search space and yield solutions more quickly.
One could argue that if the designer had perfect knowledge, the
need to design using darwinian techniques would be gone. This
point may help to explain why some continue to find difficulty
with the idea that God used a Darwinian mechanism to "create".
Glenn writes:
"Christians should be aware that design via evolution is a coming
field."
I am sure you are right. But it seems to me to be a design tool
which fits into the category of adaptation. There is still a big
gap here: the problem of innovation. Varying parameters and
exploring a search space is one matter (and computationally
feasible); providing for innovation, new information, novelty,
etc is another - and I do not think that human designers need
fear that computers will take over this role!
To a large extent, we do not know how to handle creativity in
design. Last April, the "Second International Symposium on
Creativity and Cognition" was held in the University of
Loughborough. Two of the points that emerged were:
- creativity is a complex process and ill defined in cognitive
science terms,
- computer systems used to support designers are unhelpful as
they are too restrictive.
I find a real tension between what I know of human design and
creativity (which I believe we should see as an aspect of us
imaging God), and the claim that the means God has used to create
is "Darwinian". The latter seems to cheapen design to an
optimisation process (which I fully accept as being necessary,
but certainly not the essence of design).
The more we know about these procedures, the more it seems to
emphasise the importance of intelligence and creativity in the
design process - something seriously lacking in computer programs
utilising Darwinian techniques.
Best wishes,
*** From David J. Tyler, CDT Department, Hollings Faculty,
Manchester Metropolitan University, UK.
Telephone: 0161-247-2636 ***