Note: I've written a posting saying something similar a few days ago,
but it seems it has not made it to the Usenet. Sorry if this is a
Steve McGrew <stevem at comtch.iea.com> wrote:
> I've been looking for help in designing a software model of evolution that
> might be useful for testing some of the ideas in evolutionary theory,
> and for teaching the principles of evolution to students in grades
> 8 through 12.
It seems to me that it would be quite difficult to meet both these
demands (useful for research and for teaching) at the same time. Models
designed for research typically use abstract representations for
components that are relevant for evolution. Understanding these abstractions
usually requires that one has developed quite a bit of understanding of
evolution already. It is perfectly possible to use computer models for
teaching evolution, but modelling the basic principles of evolution may
not be too likely to reveal new insights.
> I thought a genetic algorithm all by itself was a pretty good model, but got
> pretty unanimous contrary opinions from folks in the biology field. I don't
> think the objection is that software *can't* model evolution, but that so far
> it hasn't.
Certainly, no model that has been developed to date has ever modelled
evolution as a whole. Nonetheless, computer modelling approaches have
often been successfully used to investigate specific features of evolution,
such as evolutionary optimization, evolution of diversity, evolution of
developmental processes, evolution of ecosystems and food webs etc.
The fact that it is not possible to capture all aspects of evolution
and to represent them in an integrated computer model does not mean that
modelling cannot be useful to investigate specific issues about evolution,
or to test specific ideas in evolutionary theory.
> So, I'd like to hear opinions on the *essential* features a software model
> will need to have, in order to give experts in the field some confidence that
> its behavior will be usefully similar to natural evolution.
As it is not possible to model the entire process of evolution, which
features of evolution are to be considered "essential" depends on the
specific issue you want to investigate.
> [and please don't
> say it has to accurately model biochemistry! This needs to be a *simplified*
> model that retains the important features of the system it is modeling.]
I think it is quite destructive to criticize models for "not being accurate".
All modelling approaches basically consist of the identification and
abstraction of the fundamental properties of the system one is investigating
in order to implement these key features within a test environment which
is easy to handle and well suited for experimentation. This test instance
of the system under investigation is what is called the model.
From this perspective, it is quite evident that no model ever represents
all aspects of some sufficiently complex system with arbitrary accuracy.
The test environment is typically the simplest conceivable environment.
Nonetheless, modelling often is the basis for pivotal insights, and therefore,
it is nonsense to demand that a model has to include all features of a
system perfectly accurately.
Computer simulations are a good test environment for a broad array
of complex systems, and the computer modelling of living systems has been
a central field of activity in Artificial Life from the beginning (see e.g.
the preface of Artificial Life II, edited by Chris Langton et al, published
by Addison-Wesley, 1992). Therefore, my suggestion for all who are interested
in using computer models of evolution for teaching and research is to look
at the Artificial Life literature. Proceedings of Artificial Life confeferences
have been published by Addison-Wesley, MIT Press and Springer. Also, the
Artificial Life Journal is published by MIT Press. Good electronic sources
are the newsgroup comp.ai.alife and the Santa Fe Institute's ALife page,
+- Jan Kim -- X.400: S=kim;OU=violet;O=mpiz-koeln;P=mpg;C=de -----+
| Internet: kim at mpiz-koeln.mpg.de |
| WWW: http://www.mpiz-koeln.mpg.de/~kim/kim.html |
*-----=< hierarchical systems are for files, not for humans >=-----*