Re: On the Genetic Algorithm
cklester wrote:
>
> DB James wrote:
>
> > Some time back, I recalled in a post about a program I wrote maybe ten years
> > ago that
> > calculates formulas based on inputs and solutions. I have since
> > reconstructed it in
> > Euphoria. Calculates formulas? More properly said, it "evolves" an
> > algorithm for
> > the solution to the formula.
>
> Isn't this less "evolution" and more "automated algorithm tweaking."
>
> From what I understand, evolution doesn't work toward a target, does it?
>
> This is why I have a problem with programmers calling their programs
> "evolution."
>
> 1) It requires an intelligent creator to write the evolution program.
> 2) We start out with working entities because an intelligent designer
> creates them. There's no real environmental pressure (entities 'die'
> arbitrarily) and tests for fitness are simply algorithmic targets.
> 3) While one group 'dies' because they don't figure out an equation, another
> group lives, despite not having figured out an equation. Also, the
> group that dies no doubt solves some formula... just not the arbitrarily
> and narrowly defined one.
>
> Thus, GAs simulate design or targeted tweaking more than evolution.
>
Hi ck,
It is certainly a big question: what is evolution? I think it is hard to get
one's mind around because A) it has become obscured as a "political" topic and
more importantly because B) we have difficulty seeing a third possibility beyond
randomness and design. Matt used the term "fitness" which works to suggest
something other than the two creative forces just mentioned. My sense of it is
that if one accepts the following ideas, then evolution seems inevitable: all
things have characteristics which react with the characteristics of other things;
some of these things become coherent and have some level of integrity and
persistence; all such things react to the world around them by internal and
external modification; the longer such things persist, the more integrity they
tend to develop. All this seems to me perfecly natural, not metaphysical.
So, the changeable environment of any such system (or integrity or whatever one
wants to call it) tends to challenge its tendency to persist. A forest fire can
grow to the point it creates its own inward wind that feeds the flames. Yet it
modifies its environment so severely that it must move fast to maintain its "food
supply". Also, it may increase the chance of rain. So it is "born", it grows,
it matures, it spreads by propagation, it shrinks, and eventually dies (though it
can take a long time to go). I've heard of fires "living" underground for years
slowly eating away at coal seams.
Okay, so if evolution is a natural outcome of the characteristics of the
components of the universe interacting in time, then it is to the point to say
that life did not create evolution, more the other way around. There is no
evidence as far as I know that a "purpose" is implied in the way evolution
proceeds. However, even if we lack the intellectual terms for it, it does seem
that the tendency of integrities to persist is real. It is as if evolution tends
to filter out the dimmer persistences, and favors those that have more integrity,
more coherence, more (possibly) memory, encoding of lessons learned, as it were.
In a sense this would be the opposite of intelligent design, that is, not the
imposition of being from above -- it would be the emergence of persisting
integrities from below. It is like an assertion by the universe to the second
law of thermodynamics: "get knotted!"
Ha, you probably didn't want all that, so I'll just say I am satisfied that the
imposition of purpose by the programmer has nothing to do with the potency of
evolution. The math "beings" have no knowledge of the terms used to filter them
according to their behavior. But they respond anyway by continuing to change
according to their genetic code. The code records in a direction that the
environment dictates, and eventually it comes up with an answer to a question
that was never directly asked. Seems a little cruel and unfair, but them's the
breaks...
> > So much for the way it works. I was never much interested in reproducing
> > theories
> > of how real life uses information in complex ways to effect both continuity
> > and change.
>
> Oh. Well, nevermind then. :D
>
As I said, I wanted straight high-octane evolution without the sophisticated
tweaks of natural chromosomes.
> > If this sort of thing were optimized and tweaked, it could potentially
> > generate new
> > previously unknown formulas from correct inputs and results, but that is far
> > off now...
>
> How do you determine "correct inputs and results" from "unknown formulas?"
>
> -=ck
As Matt said, wherever there are known outputs and a range of inputs, some sort
of GA treatment should be able to find correllations between inputs and outputs.
In fact, I can easily imagine such a program first working out the likelihood of
which inputs are the important ones, then finding the correllations. Neural nets
are a similar kind of program in that learning does happen. They are good at
cutting through noise and finding patterns.
I would not be surprised to see GAs being valued just because they respond
without the preconceptions that hamper us.
--Quark
|
Not Categorized, Please Help
|
|
