From: Walter Hicks (wallyshoes@mindspring.com)
Date: Fri Oct 31 2003 - 20:51:21 EST
bivalve wrote:
> >I have been working with university research groups that use Genetic Algorithms (GA) to develop various physical systems. The GA (in a true DARPA like fashion) tries to imitate nature. The algorithms include chromosomes, genes, swapping, mutations, generations, and on and on. It is supposed to work just like evolution. The nice thing is that it does a fairly good job. However, the significant phrase is "fairly good". After the GA, it is necessary to use it as starting point for more conventional engineering optimizations. Sorry, Mr. Dawkins, but there is no "Climbing Mount Improbable" with this theory.
> >So what does that say? I guess that it says that the theory does not yet work.
> >When the theory is correct, it will be possible to demonstrate a theory of evolution that really works.<
>
> It might instead indicate limits to the modeling as a representation of evolution. Without knowing the details of the project, I suspect that the main discrepancy may be in the goals.
I can say as others have said to me: If you want to understand GA, you need to get a book and read it.
> Engineering is looking for optimization, and thus resorts to a strategy similar to that promoted by Intelligent Design advocates. Evolution only requires good enough; rigorous selective pressure would be necessary to approach optimization, and even then it is contingent on historical possibilities and other constraints. E..g., humans might find a third grasping appendage handy, but tails were lost in the ape/human lineage (and were not prehensile in that group anyway). This historical constraint limits human evolution, but an engineer could simply graft something on. Raising the mutation rate in the hopes of growing another appendage would probably be more detrimental due to harmful mutations than effective towards the goal, and so other constraints come into play.
Yes, but as I said, the basic point is to imitate nature because nature has been so effective in getting :the "best": solutions to a given parameter (not just a useful one ----- and this is indeed the case. That is why DARPA clings to the approach. If the evolutionary model were correct, then GA should indeed get the best answer for parameter X, just as nature does. When radar was invented, we found out that bats were way ahead of us. I believe that they still are. We cannot make devices that smell as well as a dog can, etc. If the biological model were correct, then GA would produce far more than the rough answer that it currently produces.
Generally speaking, a low mutation rate is used to get the best results from GA. I know it must smart, but GA will be better only when the theoretical biological models become more correct. (IMO) Along those lines I believe that there is no counterpart to RNA in GA and that it has been recently noted that RNA is more important than originally thought. Is that valid? Is RNA part of the biological evolution models at present?
Anyhow, when baseline theory gets better, then GA will get better with it.
Meanwhile, live with it.;)
Walt
-- =================================== Walt Hicks <wallyshoes@mindspring.com>In any consistent theory, there must exist true but not provable statements. (Godel's Theorem)
You can only find the truth with logic If you have already found the truth without it. (G.K. Chesterton) ===================================
This archive was generated by hypermail 2.1.4 : Fri Oct 31 2003 - 20:54:57 EST