From: Iain Strachan (firstname.lastname@example.org)
Date: Sat May 10 2003 - 16:48:40 EDT
Thanks for the reply; just a few loose ends that may be worth tying up:
Perhaps I wasn't clear. The authors know very well that very few mutations are beneficial, but once a beneficial one has been found, then Natural Selection acts to preserve the good ones and the bad ones don't survive. That is all standard evolutionary theory & is why their simulation worked.
So far so good.
What I'm saying is that if you have a much bigger set of possibilities to sample, though there may be exponentially more "good" mutations, the number of "bad" ones will also be exponentially more, at a greater rate.
I'm stuck on this point. I hear what you say, but I need to explore a bit more about the validity of the statement about more "bad" than "good" mutations. I am not troubled by the statement if "bad" describes the whole set of mutations that do not move toward or fail to reach the desired objective. But that would not be a fair way to characterize the evolutionary process in nature wherein the "bad" are only those entities which can degrade or destroy a different and otherwise successful mutation. Mutations that in themselves are non-viable "products" simply don't count as either "good" or "bad" - they're neutral in the evolution process.
What does "bad" mean in the context of your statement?
Probably this is best explained by a direct quote from the Nature paper:
"Most mutations in Avida are deleterious or neutral, but a small fraction increases fitness".
However, I think the point is not worth pursuing to death as it's come out of a rather loose argument. I was arguing in this way from your analogy of "parallel computing", where the idea is that exponentially many possibilities can be explored. But this interaction would only be via "crossovers", which are a different thing from "mutations". In fact I'm rather surprised that when simulations like the Nature paper are published, that they hardly ever seem to utilise crossover, which is generally reckoned by practitioners of Genetic Algorithms to do the hard graft of exploring all the possibilities more efficiently. I suppose that doesn't answer the question as to how sexual reproduction itself evolved, which itself is an enormously complex process. I guess you have to explain how that complexity emerged without crossover.
This archive was generated by hypermail 2.1.4 : Sat May 10 2003 - 16:49:02 EDT