Genetic algorithms and robotic folly
by Don Batten
A computer programmer and an engineer programmed a computer to produce ‘virtual
robots’ (that is, that operated inside the computer) that moved along a surface.1 They then constructed some of the
virtual robots from plastics and pistons, etc. with electronic controls to operate
the pistons.
Image by Brendon O’Loughlin, after
Nature
A stylised sketch of the robot nicknamed ‘the arrow’. Double headed
arrows denote pistons which extend and retract alternately, providing motion.
Some of the popular media have waxed lyrical about this development, probably because
funny little robots make for an interesting story, but also supposedly because it
shows that ‘evolution works’.
Are these computer exercises relevant to biological evolution? Scientists and engineers
have used computers to optimize structures and equations for many years, by getting
the computer to change the values of some coefficients slightly and then test to
see if the result is closer to the desired outcome. If it is, then the coefficients
are changed again and the outcome is tested again. If not, then go back and try
varying the coefficients in a different direction and test again. Many thousands
of such cycles can produce the desired outcome that would be too time-consuming
and tedious to find by manual techniques. These are known as ‘iterative’
methods, where the best values of the coefficients are determined.
Nothing new?
In recent times it has become fashionable to invoke ‘evolution’ everywhere
(even to justify infidelity, rape and the like). In keeping with this fashion, the
iterative procedures used for many years in engineering have been recast as ‘evolutionary
computation’. The variation in the coefficients has been likened to mutations,
and the testing of the outcome as ‘survival of the fittest’. The only
variation is basically that, with genetic algorithms, a number of models are generated
in parallel and tested, with a proportion of the best being selected (likened to
natural selection) for further iterations. A traditional optimization technique
works in serial mode, looking for one best solution, testing one model at a time,
whereas a genetic algorithm works in parallel, possibly generating a number of different
‘solutions’. The success of genetic algorithms in solving some problems
has then been used as evidence that ‘evolution works’.
Atheist Richard Dawkins popularized the idea of computer simulations of ‘evolution’,
using highly unrealistic programs to indoctrinate naive readers with his materialistic
views—for example, his ‘methinks its like a weasel’ exercise in
The Blind Watchmaker.2,3
Problems with computer simulations
There are a number of reasons why these computer exercises are not relevant to biological
evolution:
- Such computer simulations are strictly confined to a limited number of components.
For example, in the current example, the maximum number of components seems to be
about 13. The number of critical components—that is, those necessary for the
robot to function—is only about 4 or 5 parts. Real organisms have many thousands
of different components.
- The components are ‘given’ by the programmer. In this case the program
has available rods (‘bars’) and pistons (‘actuators’)—only
two possible types of components. The rods and pistons are joined or not joined
at their ends by ball-joints. The lengths are varied one at a time in small increments.
The ‘neural network’ that ‘evolved’ is also very simple
in effect: operate the piston, or if there is more than one, the choices are to
operate them synchronously or asynchronously. In other words, there is a very limited
set of options for ‘mutations’ to occur. Dawkins used this trick also
in his ‘methinks it’s like a weasel’ con. In the real world, even
the simplest bacterium has hundreds of thousands of sites where mutations can occur.
Computer programmers have to strictly limit their ‘mutations’, otherwise
they know that error catastrophe will result—where the program gets lost and
cannot arrive at any solution. This is a fundamental problem with the evolutionary
story for living things—mutations cause the destruction of the genetic information
(and consequently they are known by the thousands of diseases they cause), not its
creation.
- The ‘selection’ is only for one trait—movement. In the real world
of living organisms, selection must be for hundreds of different traits at once.
Mutations are not confined to one part of the organism’s program (DNA), and
therefore to one trait. For every mutation that might affect a trait such as movement,
hundreds of mutations will affect other traits, such as reproduction, metabolism
of sugars, etc., so they all have to be selected for. And to complicate things further,
a given trait can be affected by mutations in different parts of an organism’s
DNA, a single mutation can affect more than one trait, and many traits involve the
co-ordinated action of more than one gene. Inclusion of many traits in the computer
program would render the procedure unworkable (it is very difficult to get iterative
processes to work with more than one goal).
The computer exercise did not start with nothing—it started with a program
generated by intelligent scientists that specified the way in which the robots could be constructed
- The programmer has pre-programmed the computer for a specific goal. ‘Evolution’
can have no specific goals, such as locomotion, as it is purposeless, being driven
by chance, not intelligence.
- The computer exercise did not start with nothing—it started with
a program generated by intelligent scientists that specified the way in which the
robots could be constructed.
- Given the components (pistons, rods, etc.) programmed into the computer, it is no
great achievement to have achieved movement in the robots—all that is required
is to lift one end of a piston off the ground and have it expand and contract.
- In spite of the chutzpah (such as calling the robots ‘lifeforms’),
the robots cannot reproduce themselves. They are dependent on their human creators
to manufacture them. They are not ‘lifeforms’ in any meaningful sense
of the word. The simplest of living things can gather the raw materials and and
energy to manufacture the components to reproduce themselves.
- The robots produced by the program contrast with living things in that they look
‘jerry-built’ (the one illustrated above is about the most regular looking
robot produced and it looks more regular in our drawing than in 3D). Even atheists
like Richard Dawkins admit that living things look like they are beautifully designed—they
look like an intelligent creator cleverly designed them (and then he uses evolutionary
story-telling to try to explain how they actually made themselves by mutations and
natural selection). However, the robots ‘evolved’ in the computer do
not look like they were cleverly designed—they look like they were thrown
together. The most complex virtual robot illustrated on the Nature website,
dubbed the ‘arrow’, can be seen at:
http://www.nature.com/nature/journal/v406/n6799/extref/406974ai1.mpeg
The model made of it can be seen at:
http://www.nature.com/nature/journal/v406/n6799/extref/406974ai2.mpeg
It is clear that the parts of it that look like they were the result of intelligent
design are the components specified in the original computer program (pistons, joints
and rods). The arrangement of the parts looks like the
result of a haphazard process. Living things do not look like they came about by
a haphazard (random) process. They look like they were designed. See for example, Design in living organisms (motors)
- ‘Genetic algorithms’ use completely unrealistic ‘genome’
sizes (very small), mutation rates (extremely high) and selection coefficients (very
high).4 They also do not take into
account non-viability—that is, an organism would not be viable at all (and
therefore evolution could not proceed!) until the system that is supposedly evolving
in the computer actually worked. Real-world organisms need to be viable and maintain
viability. ReMine addresses the problems of mutation rates and selection coefficients
for the evolutionary story, showing that the neo-Darwinian mechanism just cannot
explain the amount of information in genomes.5,6
Conclusion
These are some of the reasons that ‘evolution’ simulations in computers—such
as this latest one given a ‘beat up’—have no relevance to the
materialists’ belief in molecules-to-man evolution. In fact the severe limitations
on such procedures, even with fast, powerful modern computers, shows how real-world
(biological) molecules-to-man evolution is impossible, even if there were the eons of time claimed
by evolutionists.
Related articles:
Related products
References
- Lipson, H. and Pollack, J.B., Automatic design and manufacture
of robotic lifeforms, Nature 406:974–978, 2000.
Return to Text.
- Gitt, W., Weasel Words,
Creation 20(4):20–21, 1998. Return to
Text.
- Truman, R., Dawkins’ weasel
revisited, Journal of Creation 12(3):358–361,
1998. Return to Text.
- For example, Schneider, T.D., Evolution of biological information,
Nucleic Acids Research 28(14):2794–2799, 2000. Return to Text.
- ReMine, W., The Biotic Message, Saint Paul Science,
Saint Paul, Minnesota, USA, 1993. Return to Text.
- Batten, D., Review of The Biotic Message: Evolution Versus
Message Theory by Walter J. ReMine, Journal of Creation 11(3):292–298,
1997. Return to Text.
|
