Darwinian algorithms
Fross
A.I. (artificial intelligence) systems is to set up a program that mimics
Darwinian evolution. They don't sit down and design these complex systems
before hand and then execute their design. They simply set up a few basic
rules and let a darwinian process bring forth the complex system. I find
it fascinating that we can apply the basic principles of evolution to
computer programs and we get unpredictable results that are usually too
complex for the programmers to figure out. Sure their new program works,
but how?
Why don't creationists see this as a major point for evolution? If the
theory works on replicating software programs, why would it not work on
replicating DNA? Computer processors are slow and limited compared to
reality, which runs at light speed! Yet the basics of Darwinian evolution
work on them. What would stop it from working in reality?
I've brought this up before in hopes that a creationist would give some
input. I haven't heard this issue addressed.
Best,
Fross
John Wilkins
Their standard evasion^Hresponse is to say that computers are programmed
by inteligences, and that is why they work.
I think they worry about the implication God might be a software
engineer without a clue how to solve multivariate problems. Or
something.
--
John Wilkins
B'dies, Brutius
Fross
the process of evolution.
John Wilkins
> I just don't see why there's a dichotomy between the existence of God and
> the process of evolution.
There isn't, to 90% of religious believers. But some, who think that all
knowledge must be at least implicit in a literal reading of a sacred
text, think that if evolution is true their sacred text cannot be, and
so they object to science and evidence.
Bobby D. Bryant
> I just don't see why there's a dichotomy between the existence of God and
> the process of evolution.
Most people don't. But many have promoted evolution denial to an article
of faith for their sect, so here we are.
--
Bobby Bryant
Austin, Texas
Dissident
> Fross <jay...@hotmail.com> wrote:
>
>
>>I just don't see why there's a dichotomy between the existence of God and
>>the process of evolution.
>
>
> There isn't, to 90% of religious believers. But some, who think that all
> knowledge must be at least implicit in a literal reading of a sacred
> text, think that if evolution is true their sacred text cannot be, and
> so they object to science and evidence.
It just occurred to me that this is a standard problem in
contract law, and contracts have a way around it, called
a "seperability clause". You know, the one you always
see that says something like "if any of the provisions
are found to be invalid, it shall not affect the validity
of any other parts of this contract".
What the Bible needs is a seperability clause. Now, since
so much Biblical scholarship involves reading meaning
into obscure and contradictory passages, can't some clever
scholar convince the fundies that if creation is proven
false, it doesn't have to mean folks will all go around
killing and raping at will because suddenly the entire
Bible is invalid?
TomS
<OtacnWW0O-P...@comcast.com>, "Fross" stated..."
>
>I just don't see why there's a dichotomy between the existence of God and
>the process of evolution.
IMHO ...
There isn't, and lots of people don't think there is any conflict.
IMHO, there are complex roots for the idea that there is such a
dichotomy.
There is the idea that anything that is upsetting has to be
immoral. And the only place that some people can turn for reasurance
about being related to the rest of creation, is to religion.
There are gross misunderstandings of what evolution is about.
For a lot of people, their idea of evolution is so strange, that it's
no wonder that they don't accept it.
There are gross misunderstandings of what any moderately
sophisticated theology of creation says. (This also works to make
many scientists non-monotheists. It's no wonder that intelligent
people can't accept a strange theology.)
That's some of the reasons, IMHO.
Tom S.
Scott
"Fross" <jay...@hotmail.com> wrote in message
news:r-WdnaHWSpW...@comcast.com...
> Software engineers are quickly discovering that the best way to get
complex
> A.I. (artificial intelligence) systems is to set up a program that mimics
> Darwinian evolution. They don't sit down and design these complex systems
> before hand and then execute their design. They simply set up a few basic
> rules and let a darwinian process bring forth the complex system. I
find
> it fascinating that we can apply the basic principles of evolution to
> computer programs and we get unpredictable results that are usually too
> complex for the programmers to figure out. Sure their new program works,
> but how?
That brings up a question. If these A.I systems are designed to mimic
Darwinian evolution and - for lack of a better word- *progressively* evolve
to greater complexity in intelligence could Darwinian evolution likewise be
expected to accomplish the same feat each time within an ecosystem were it
rerun in the natural world?
Scott
Louann Miller
wrote:
(genetic algorithms)
>Why don't creationists see this as a major point for evolution? If the
>theory works on replicating software programs, why would it not work on
>replicating DNA? Computer processors are slow and limited compared to
>reality, which runs at light speed! Yet the basics of Darwinian evolution
>work on them. What would stop it from working in reality?
>I've brought this up before in hopes that a creationist would give some
>input. I haven't heard this issue addressed.
Good luck on that one. I'd like to hear it too.
Louann
Andres Soolo
> It just occurred to me that this is a standard problem in
> contract law, and contracts have a way around it, called
> a "seperability clause". You know, the one you always
> What the Bible needs is a seperability clause. Now, since
> so much Biblical scholarship involves reading meaning
There once was, kind of. Nowadays fundamentalists aren't usually
told it, but there are writs that their forerunners found to be
"not holy enough" and decanonicized. The Book of Enoch is a classic
example.
And the process works the other way around, too :-)
The Mormon scriptures are a classic example.
--
Andres Soolo <so...@math.ut.ee>
Life is not for everyone.
Laurence A. Moran
Let's do a thought experiment. Assume that I've just written a computer
program that uses the principle of inheritance of acquired characteristics.
Assume that the program works just as well as existing genetic algorithms.
Would you take that as strong evidence that Lamarck must have been correct
and that's how evolution happens in nature? Of course not. There's no
obvious connection between workable computer programs and biological
evolution.
Computer programs based on strong positive selection for pre-determined
characteristics work. This does not mean that strong positive selection
for pre-determined characteristics is the mechanism that predominates in
biological evolution.
If Creationists knew enough about evolution they would use this argument.
And they would be right.
Larry Moran
Matt Silberstein
lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>In article <trkn4v4ace35lm68g...@4ax.com>,
>Louann Miller <loua...@yahoo.net> wrote:
>>On Thu, 13 Feb 2003 04:10:05 +0000 (UTC), "Fross" <jay...@hotmail.com>
>>wrote:
>>
>>(genetic algorithms)
>>
>>>Why don't creationists see this as a major point for evolution? If the
>>>theory works on replicating software programs, why would it not work on
>>>replicating DNA? Computer processors are slow and limited compared to
>>>reality, which runs at light speed! Yet the basics of Darwinian evolution
>>>work on them. What would stop it from working in reality?
>>
>>>I've brought this up before in hopes that a creationist would give some
>>>input. I haven't heard this issue addressed.
>>
>>Good luck on that one. I'd like to hear it too.
>
>Let's do a thought experiment. Assume that I've just written a computer
>program that uses the principle of inheritance of acquired characteristics.
Could you explain a bit more how this would work. How did the
computer program acquire these characteristics and how do they
get passed on? I ask because I strongly suggest that you either
have normal evolution or it is impossible.
>Assume that the program works just as well as existing genetic algorithms.
>Would you take that as strong evidence that Lamarck must have been correct
>and that's how evolution happens in nature? Of course not. There's no
>obvious connection between workable computer programs and biological
>evolution.
If you had a program that did this I would take it as strong
evidence that such a process could work. (Demonstration has a
tendency towards that.) I would then look to see if the process
modeled things that occur in nature. If it did I would wonder if
a similar process then works in nature. If it did not I would
look for whatever it was that stopped the process.
>Computer programs based on strong positive selection for pre-determined
>characteristics work.
Computer programs based on strong positive selection for
non-predetermined characteristics, but for strong selection on
"results" also works.
>This does not mean that strong positive selection
>for pre-determined characteristics is the mechanism that predominates in
>biological evolution.
Of course not. It does suggest that strong positive selection
does do something. And we do find that strong positive selection
does things in nature. The question of whether, in nature (that
is, in biology) strong selection is the only/main mechanism for
evolution is a *different* question. I am pretty sure that, in
nature, where there is strong positive selection there is less
drift. That does not mean that there is always strong positive
selection.
>If Creationists knew enough about evolution they would use this argument.
>And they would be right.
Only because they would have changed the question.
--
Matt Silberstein TBC HRL OMM
Politics is the art of the possible
Bismarck, but not the ship
Bobby D. Bryant
Heck, if they deny all the direct biological evidence why should we expect
them to accept the computational demonstration of principle?
Bobby D. Bryant
> Computer programs based on strong positive selection for pre-determined
> characteristics work. This does not mean that strong positive selection
> for pre-determined characteristics is the mechanism that predominates in
> biological evolution.
Agreed. However, genetic algorithms are very useful as demonstrative
refutations of the frequent creationist claim that random mutations
filtered by natural selection can't "do" anything. (Usually phrased even
more sloppily that that.)
Bobby D. Bryant
> That brings up a question. If these A.I systems are designed to mimic
> Darwinian evolution and - for lack of a better word- *progressively*
> evolve to greater complexity in intelligence could Darwinian evolution
> likewise be expected to accomplish the same feat each time within an
> ecosystem were it rerun in the natural world?
I'm having a bit of difficulty pinning down exactly what you're asking. If
you mean "mimic" quite literally, then the problem is that our computers
are nowhere near big/fast enough to do this, and the models for an
ecosystem would be far too complex for us to implement anywhere nearly
precisely enough.
Also, since evolution (according to the theory) operates by filtering the
results of random mutation, that randomness should allow reality and even
a perfect model to diverge over time. If we could turn the clock back on
earth's bio-history, there's no reason to expect that evolution would
produce the same species on the second run.
If that scattershot answer missed your question, please rephrase it in a
reply.
Bigdakine
>From: lam...@bioinfo.med.utoronto.ca (Laurence A. Moran)
>Date: 2/13/03 9:40 AM Hawaiian Standard Time
>Message-id: <b2grgi$br$1...@bioinfo.med.utoronto.ca>
No, but a negative result would cast considerable doubt on the idea that it can
be the mechanism.
Demonstrations aren't proofs, they show feasibility. Dawkins' weasel program
isn't a proof either, but does demonstrates the efficiency of cumulative
selection over just plain mutation. We all know Dawkin's weasel program is not
a computational analogy for the real world, however, if Weasel didn't work, its
rather hard to see how anything like it in the real world would.
One might come up with a simulation of Lamarckian evolution. OTOH, where is the
evidence that Lamarckian evolution happens?
We have evidence that NS happens. And it has provided the impetus for a class
of mathematical methods called stochastic hill climbing methods.
Low and behold, they work. Its what engineers call *proof of concept*.
>
>If Creationists knew enough about evolution they would use this argument.
>And they would be right.
They'd be right in pointing out that a demonstration isn't proof that nature
works that way. They would have to, however, show why such a mechanism
couldn't, in principle, work in nature. And, lets see them give a *proof of
concept* for any of their mechanisms.
Oops nevermind, they don't have any mechanisms.
Stuart
Dr. Stuart A. Weinstein
Ewa Beach Institute of Tectonics
"To err is human, but to really foul things up
requires a creationist"
Laurence A. Moran
Matt Silberstein <mat...@ix.netcom.com> wrote:
>In talk.origins I read this message from
>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
[snip]
>>Computer programs based on strong positive selection for pre-determined
>>characteristics work.
>
>Computer programs based on strong positive selection for
>non-predetermined characteristics, but for strong selection on
>"results" also works.
Who determines what "results" you're looking for? Your programs can't
undergo selection unless you predetermine what's going to be a good
result. (Yes, I know there are some strange exceptions but in general
the goal of the program has to be defined in advance.)
This is why genetic algorithms aren't a very good model for biological
evolution. Evolution has no purpose or goal. Genetic algorithms do.
Larry Moran
Bobby D. Bryant
Strictly speaking, it's the algorithmist who has the goal rather than the
algorithm. The algorithmist simply has to supply a fitness function that
s/he thinks will cause the algorithm to find the desired goal.
I've been idly thinking somewhat lately about setting up a GA with a
"rich" fitness function that represents some environment rather than some
problem goal, and seeing whether the GA produces any interesting species
in that environment. (I strongly suspect that this has already been done,
though I've never looked in to it.)
At any rate, even if the human supplies a "goal" such as fitness function
that evaluates the quality of solutions to the Travelling Salesman
Problem, that fitness function is just an abstract "environment" for the
GA. It knows nothing of goals or even what problem it is working on; it
just sees that some solutions get scored higher than others.
I see no reason at all not to invoke GAs as demonstrations of principle,
especially in response to certain specific creationist claims. That's a
far cry from offering GAs as proof that biological evolution historically
happened by the theoretical mechanism, but hopefully no one is offering
them in that manner.
Scott
"Bobby D. Bryant" <bdbr...@mail.utexas.edu> wrote in message
news:pan.2003.02.13....@mail.utexas.edu...
Sorry, Bobby, but that's not what I'm asking. Actually, after I asked this
question I noticed that John and dkomo were discussing the similar thing in
John's "Cosmic Evolution" thread.
I'm not asking if it's possible to mimic the earth's evolutionary history in
a computer and have the same species be produced over and over again. It's
not a species question. Although I'd like to think computer programs could
at least *hint* at an answer to what I'm asking.
As Foss said, programmers "simply set up a few basic rules and let a
Darwinian process bring forth the complex system"/AI.
From a biological POV, could it be possible given a few basic rules (time,
energy, diversity, etc.) complex ecosystems will always evolve and have a
likely probability to create intelligence, irregardless of whatever species
evolve? If you could rerun the earth's clock and assume flight or the eye
will evolve again why not intelligence? Are hierarchies of complexity
(including hierarchies of intelligence) more than a fluke of evolution?
I know most biologist will say human intelligence is chance, a fluke of
history and not science. But is it really? (I admit to being intuitively
skeptical to intelligence being a *too good to be true* story) In any case
it's a question I think ought to be asked from time to time.
Scott
Andrew Arensburger
> Software engineers are quickly discovering that the best way to get complex
> A.I. (artificial intelligence) systems is to set up a program that mimics
> Darwinian evolution. They don't sit down and design these complex systems
> before hand and then execute their design. They simply set up a few basic
> rules and let a darwinian process bring forth the complex system. I find
> it fascinating that we can apply the basic principles of evolution to
> computer programs and we get unpredictable results that are usually too
> complex for the programmers to figure out. Sure their new program works,
> but how?
That's sometimes the case, but not always. Yes, you can use EC
to write code, and the resulting code will almos certainly be
incomprehensible. But you can also use EC to solve conceptually simple
problems.
For instance, you might use EC to "breed" a crossword puzzle,
but there won't be anything mysterious about the result.
> Why don't creationists see this as a major point for evolution? If the
> theory works on replicating software programs, why would it not work on
> replicating DNA?
One valid (IMO) answer is that Evolutionary Computation is not
Evolution, even though they have some features in common. At some
level, real-life evolution involves matter, whereas EC is just
software (which in turn is just math).
Or take the fact that in EC, it's generally clear which parts
of the "organism" are the "genome" and which ones aren't. In real
life, things aren't so clear. As I understand it, there are proteins
that attach themselves to certain segments of DNA or RNA and prevent
that segment from being read. Are these proteins part of the genome,
or not? My genes are responsible for building the genetic machinery
that a virus will exploit to make copies of itself. So are my genes
part of the virus's genome? And so forth.
On top of this, software engineers aren't interested in
simulating real-life biology; they just want something that works. So
they can break the rules, e.g., allow a low-scoring "organism" to
continue into the next round. And that's okay.
So the success of evolutionary computation means that there's
no a priori mathematical reason to reject the theory of evolution.
> Computer processors are slow and limited compared to
> reality, which runs at light speed! Yet the basics of Darwinian evolution
> work on them. What would stop it from working in reality?
The fact that something works on a computer does not mean that
it works in real life. It's encouraging, certes, but it's not proof.
Obviously, the more models you have that exhibit the same behavior as
the real world, the better, but there's always the chance that you've
overlooked something, or that there's some complicating factor in an
infinitely messy world.
--
Andrew Arensburger, Systems guy University of Maryland
arensb.no-...@glue.umd.edu Office of Information Technology
Okay, that's it! Arm the electric bagpipes!
Bobby D. Bryant
One important thing to notice is that GAs don't _always_ bring forth good
solutions to the problem. I've seen a couple of old papers trying to
address what makes a problem hard for a GA, but the results were tentative
and pretty abstract mathematically. I don't think we're anywhere near the
point of being able to look at a problem we want to solve and answer
reliably whether or not a GA will solve it well.
> From a biological POV, could it be possible given a few basic rules (time,
> energy, diversity, etc.) complex ecosystems will always evolve and have a
> likely probability to create intelligence, irregardless of whatever species
> evolve? If you could rerun the earth's clock and assume flight or the eye
> will evolve again why not intelligence? Are hierarchies of complexity
> (including hierarchies of intelligence) more than a fluke of evolution?
>
> I know most biologist will say human intelligence is chance, a fluke of
> history and not science. But is it really? (I admit to being intuitively
> skeptical to intelligence being a *too good to be true* story) In any case
> it's a question I think ought to be asked from time to time.
My intuition is that there's no guarantee that you would get intellegence,
eyes, or even multi-cellular organisms. Even estimating their
probabilities seems impossible so long as we only have a single sample.
If we could find several additional solar systems infested with life it
would give us a better idea about how such questions should be answered.
Notice that I'm not a biologist either; I will be interested in seeing any
answers our resident biologists offer.
John Thomas Grisham
Back in the late 1970's, before DOS ... before CPM, I worked along
side early programmers that were making up their own language for 8088
processors. They didn't have monitors or keyboards. They used toggles
to enter lines of code into 4k of RAM. And, that was a great leap in
technology over the previous 30 years of developement. If, you can get
that A.I. program to run on 4k of RAM without an environment or
operating system, then, that would almost be comparable to DNA,
evolving on its own.
If, there was no intellegent design, there would be no A.I. program or
environment to run it on.
JTG
Nathan Urban
> On Thu, 13 Feb 2003 21:42:46 +0000, Laurence A. Moran wrote:
> > Who determines what "results" you're looking for? Your programs can't
> > undergo selection unless you predetermine what's going to be a good
> > result. (Yes, I know there are some strange exceptions but in general
> > the goal of the program has to be defined in advance.)
> > This is why genetic algorithms aren't a very good model for biological
> > evolution. Evolution has no purpose or goal. Genetic algorithms do.
> Strictly speaking, it's the algorithmist who has the goal rather than the
> algorithm. The algorithmist simply has to supply a fitness function that
> s/he thinks will cause the algorithm to find the desired goal.
Right. Genetic algorithms don't have any more "purpose" or "goal"
than evolution does. They just work on whatever fitness function
they're given. If we knew the fitness function for Earth's biosphere,
we could in principle apply a genetic algorithm to mimic biological
evolution. (Complicated by the fact that the fitness function itself
changes over time as it is a function of the environment...)
> I've been idly thinking somewhat lately about setting up a GA with a
> "rich" fitness function that represents some environment rather than some
> problem goal, and seeing whether the GA produces any interesting species
> in that environment. (I strongly suspect that this has already been done,
> though I've never looked in to it.)
Sure. This is what the "Artificial Life" (A-Life) people do. A famous
example is Tierra:
Bobby D. Bryant
> Back in the late 1970's, before DOS ... before CPM, I worked along
> side early programmers that were making up their own language for 8088
> processors. They didn't have monitors or keyboards. They used toggles
> to enter lines of code into 4k of RAM. And, that was a great leap in
> technology over the previous 30 years of developement. If, you can get
> that A.I. program to run on 4k of RAM without an environment or
> operating system, then, that would almost be comparable to DNA,
> evolving on its own.
Huh?
> If, there was no intellegent design, there would be no A.I. program or
> environment to run it on.
Nobody with a clue says otherwise. But that entirely misses the point.
Noelie S. Alito
A basic question about GAs: Do they typically include a scheme to
occasionally let a "lucky bastard" get through in spite of its deficiencies?
<chop>
Noelie
--
Avoiding studying for her igmetpet test for over 42 minutes.
Laurence A. Moran
Bobby D. Bryant <bdbr...@mail.utexas.edu> wrote:
[snip]
>I see no reason at all not to invoke GAs as demonstrations of
>principle, especially in response to certain specific creationist
>claims.
Why bother? Don't we have enough really good biological examples
of natural selection in action? GA's just muddy the waters and
they aren't very good models of what happens with living
organisms.
Larry Moran
John Wilkins
Well, that sounds like a Darwinian process alright. Hull calls it
"survival of the most adequate"... a good solution is nice, but all it
really has to do is do a little better than its nearest competitor
strategies.
Also, Darwinian processes that *fail* to do better are called "extinct".
>
> A basic question about GAs: Do they typically include a scheme to
> occasionally let a "lucky bastard" get through in spite of its deficiencies?
Only if they are written in JScript.
>
> <chop>
>
> Noelie
> --
> Avoiding studying for her igmetpet test for over 42 minutes.
John
Avoiding finishing the thesis for 6 years :-P
Matt Silberstein
lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>In article <inun4votfvvhc6dpr...@4ax.com>,
>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>In talk.origins I read this message from
>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>
>[snip]
I take it that we both agree that we probably can't model
Lamarkian evolution because it does not actually work.
>>>Computer programs based on strong positive selection for pre-determined
>>>characteristics work.
>>
>>Computer programs based on strong positive selection for
>>non-predetermined characteristics, but for strong selection on
>>"results" also works.
>
>Who determines what "results" you're looking for?
Who determines the "results" in the world"? This is a fallacy:
human set human experiments, that is all we can get in the lab.
>Your programs can't
>undergo selection unless you predetermine what's going to be a good
>result. (Yes, I know there are some strange exceptions but in general
>the goal of the program has to be defined in advance.)
You are confusing the solution and the mechanism that produces
the solution. Sure, we human define the "problem", traveling
salesman or whatever. But that does not predetermine the
solution. Just like the world did not determine that parrots fly,
flying was a "solution" "found" by aves.
>This is why genetic algorithms aren't a very good model for biological
>evolution. Evolution has no purpose or goal. Genetic algorithms do.
Evolution has not purpose, evolution is a general term. Specific
evolving lines have "problems" and, on occasion, find
"solutions". Just like lines in software. There is no intrinsic
difference between computer software/hardware and biological
organisms. Biology is more complex, has more parts and more
interactions and has been doing this far longer. Computers help
us see bits of biology, like other models they abstract from what
we see. And, unlike most other models, they are also examples of
the thing, not merely models.
John Wilkins
> In talk.origins I read this message from
> lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>
> >In article <inun4votfvvhc6dpr...@4ax.com>,
> >Matt Silberstein <mat...@ix.netcom.com> wrote:
> >>In talk.origins I read this message from
> >>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
> >
> >[snip]
>
> I take it that we both agree that we probably can't model
> Lamarkian evolution because it does not actually work.
It depends. I have argued In Another Forum (in press, that is; see
below) that Lamarckian accounts are always reducible to Darwinian
accounts at another level.
However, there are two problems with Lamarckian evolution - one is
definitional; do we mean inheritance of acquired characters? Or do we
mean an internal drive towards complexity? Or do we mean (as Fisher did)
blending, i.e., non-particulate inheritance? All these can be modelled,
and as Fisher noted, the latter can be a Darwinian process in its own
right, so long as the rate of the arrival of novelty of alleles exceeds
the rate of the averaging of traits in the blend. The soft inheritance
process is not hard to simulate - just program in an "evaluator"; of
course getting an evaluator that is prescient might be harder to
achieve. And innate tendencies to complexity can perhaps be set up with
a ratchett mechanism.
For all that, I cannot see how soft inheritance can work in the absence
of prior knowledge of good solutions, and I cannot see how good
solutions can be discovered from a starting point of ignorance without a
selection process.
The problem with the ratchett is, of course, that the choice of property
to ratchett up is subject to the same problems of predicting the future
as anything else. End result? You *still* need either foreknowledge or
selection. In the absence of foreknowledge, only selection will do, and
in *every* case in which foreknowledge is posited, it is itself the
outcome of selection.
>
> >>>Computer programs based on strong positive selection for pre-determined
> >>>characteristics work.
> >>
> >>Computer programs based on strong positive selection for
> >>non-predetermined characteristics, but for strong selection on
> >>"results" also works.
> >
> >Who determines what "results" you're looking for?
>
> Who determines the "results" in the world"? This is a fallacy:
> human set human experiments, that is all we can get in the lab.
Again, it depends on the conditions - if humans are the conditions, then
they are what gets adapted to, but I don't see why this is an objection
to GAs as a model of some aspects of Darwinian evolution. Humans no more
have absolute foreknowledge than anything else.
>
> >Your programs can't
> >undergo selection unless you predetermine what's going to be a good
> >result. (Yes, I know there are some strange exceptions but in general
> >the goal of the program has to be defined in advance.)
>
> You are confusing the solution and the mechanism that produces
> the solution. Sure, we human define the "problem", traveling
> salesman or whatever. But that does not predetermine the
> solution. Just like the world did not determine that parrots fly,
> flying was a "solution" "found" by aves.
>
> >This is why genetic algorithms aren't a very good model for biological
> >evolution. Evolution has no purpose or goal. Genetic algorithms do.
>
> Evolution has not purpose, evolution is a general term. Specific
> evolving lines have "problems" and, on occasion, find
> "solutions". Just like lines in software. There is no intrinsic
> difference between computer software/hardware and biological
> organisms. Biology is more complex, has more parts and more
> interactions and has been doing this far longer. Computers help
> us see bits of biology, like other models they abstract from what
> we see. And, unlike most other models, they are also examples of
> the thing, not merely models.
And we may observe, as Elliot Sober once said, lions and gazelles have
intentions. Intentions are themselves no reason to insulate a situation
from considerations of evolutionary fitness. They are, in the end,
strategies that work, or not.
===
Wilkins, John S. 2001. The appearance of Lamarckism in the evolution of
culture. In Darwinism and evolutionary economics, edited by J. Laurent
and J. Nightingale. Cheltenham UK: Edward Elgar.
Paul J Gans
>Software engineers are quickly discovering that the best way to get complex
>A.I. (artificial intelligence) systems is to set up a program that mimics
>Darwinian evolution. They don't sit down and design these complex systems
>before hand and then execute their design. They simply set up a few basic
>rules and let a darwinian process bring forth the complex system. I find
>it fascinating that we can apply the basic principles of evolution to
>computer programs and we get unpredictable results that are usually too
>complex for the programmers to figure out. Sure their new program works,
>but how?
>Why don't creationists see this as a major point for evolution? If the
>theory works on replicating software programs, why would it not work on
>replicating DNA? Computer processors are slow and limited compared to
>reality, which runs at light speed! Yet the basics of Darwinian evolution
>work on them. What would stop it from working in reality?
>I've brought this up before in hopes that a creationist would give some
>input. I haven't heard this issue addressed.
>Best,
>Fross
Well, it is and it isn't like evolution. Whenever you
have a system with imperfect replicators and a means
of selecting those that are more "fit" for their
environment (or a means of "killing off" those that
are less fit for their environment) you are going
to move toward a system that fits your requirements.
An artificial system can do this quite well. It is,
in fact, what is behind the "simulated annealing"
method of finding a reasonable candidate for an
absolute minimum in a function of many variables
with many local minima.
Darwinian evolution is a tad different in that there
is no fixed environment. The surroundings consist
of a whole haggle of other Darwinian evolvers plus
the ever changing "natural" landscape -- floods,
ice ages, tornados, volcanos, etc.
In the first case it is clear that what comes out
will be nearer the selection criteria than what
went in.
In the Darwinian case the result isn't clear at all.
Some organisms become more "complex" in some sense.
Others lose function and become less "complex". There's
no direction and no guaranteed result.
What *is* clear (and should be to creationists) is that
imperfect replicators will of necessity change. The
only argument is over the limits to such change. Do
any exist? Creationists would say yes and that the
change is limited within one "kind". Evolutionists
say no, the change is not limited.
------ Paul J. Gans
Paul J Gans
>I just don't see why there's a dichotomy between the existence of God and
>the process of evolution.
I don't think there is. The only reasonable question on that
score seems to me to be how much day to day activity does God
put in. Too much and science doesn't work. But that doesn't
mean "none".
---- Paul J. Gans
>> Their standard evasion^Hresponse is to say that computers are programmed
>> by inteligences, and that is why they work.
>>
>> I think they worry about the implication God might be a software
>> engineer without a clue how to solve multivariate problems. Or
>> something.
Paul J Gans
>> Software engineers are quickly discovering that the best way to get complex
>> A.I. (artificial intelligence) systems is to set up a program that mimics
>> Darwinian evolution. They don't sit down and design these complex systems
>> before hand and then execute their design. They simply set up a few basic
>> rules and let a darwinian process bring forth the complex system. I find
>> it fascinating that we can apply the basic principles of evolution to
>> computer programs and we get unpredictable results that are usually too
>> complex for the programmers to figure out. Sure their new program works,
>> but how?
>> Why don't creationists see this as a major point for evolution? If the
>> theory works on replicating software programs, why would it not work on
>> replicating DNA? Computer processors are slow and limited compared to
>> reality, which runs at light speed! Yet the basics of Darwinian evolution
>> work on them. What would stop it from working in reality?
>>
>>
>> I've brought this up before in hopes that a creationist would give some
>> input. I haven't heard this issue addressed.
>>
>Their standard evasion^Hresponse is to say that computers are programmed
>by inteligences, and that is why they work.
>I think they worry about the implication God might be a software
>engineer without a clue how to solve multivariate problems. Or
>something.
Say not so! Please, don't let God be a software engineer.
Please?
I don't want to worship Microsoft.
---- Paul J. Gans
Paul J Gans
>John Wilkins wrote:
>> Fross <jay...@hotmail.com> wrote:
>>
>>
>>>I just don't see why there's a dichotomy between the existence of God and
>>>the process of evolution.
>>
>>
>> knowledge must be at least implicit in a literal reading of a sacred
>> text, think that if evolution is true their sacred text cannot be, and
>> so they object to science and evidence.
>It just occurred to me that this is a standard problem in
>contract law, and contracts have a way around it, called
>a "seperability clause". You know, the one you always
>see that says something like "if any of the provisions
>are found to be invalid, it shall not affect the validity
>of any other parts of this contract".
>What the Bible needs is a seperability clause. Now, since
>so much Biblical scholarship involves reading meaning
>into obscure and contradictory passages, can't some clever
>scholar convince the fundies that if creation is proven
>false, it doesn't have to mean folks will all go around
>killing and raping at will because suddenly the entire
>Bible is invalid?
You'd think so, wouldn't you? In fact there already
is such an example. Most folks in the world are
not Christians. And yet they don't go round raping
and killing willy-nilly.
I know that the standard response to that is "But
they still believe in A god!", but that won't work.
Because what that means is that you don't have to
be Christian to be moral. And most fundies won't
accept that because it means that one doesn't have
to be a fundy to be moral.
----- Paul J. Gans
Paul J Gans
>"Fross" <jay...@hotmail.com> wrote in message
>news:r-WdnaHWSpW...@comcast.com...
>> Software engineers are quickly discovering that the best way to get
>complex
>> A.I. (artificial intelligence) systems is to set up a program that mimics
>> Darwinian evolution. They don't sit down and design these complex systems
>> before hand and then execute their design. They simply set up a few basic
>> rules and let a darwinian process bring forth the complex system. I
>find
>> it fascinating that we can apply the basic principles of evolution to
>> computer programs and we get unpredictable results that are usually too
>> complex for the programmers to figure out. Sure their new program works,
>> but how?
>That brings up a question. If these A.I systems are designed to mimic
>Darwinian evolution and - for lack of a better word- *progressively* evolve
>to greater complexity in intelligence could Darwinian evolution likewise be
>expected to accomplish the same feat each time within an ecosystem were it
>rerun in the natural world?
>Scott
No. See my recent post on this. AI is not Darwinian.
The overall movement toward more complex systems that
we observe in the world around is is very likely due
to starting with unicellular entities. They had no
where to go but toward multicellularity.
Even then, it took a long long time.
But I do not think that there is anything in evolution
that means that intelligence is inevitable.
---- Paul J. Gans
Paul J Gans
I believe that most main-stream Christian religions
of a non-fundy nature have no problem with this.
I've heard it argued that where the Bible disagrees
with experience, we have to trust experience and treat
the Bible as metaphoric. Where the Bible does not
disagree with experience, we should treat it as true.
---- Paul J. Gans
Paul J Gans
Sure. Imperfect replicators *must* show change as long as
any selection method at all is operating.
But computer experiments usually have fixed selection
rules and no competition. Thus clearly imperfect replicators
will move toward satisfying those rules.
As I pointed out previously, it is very much like "simulated
annealing" in that respect and simulated annealing works
(or can work) very well.
In fact, I'd bet that a clever person could prove this
mathematically.
But I don't think that those experiences generalize in
any easy way to a world with many interacting imperfect
replicators living in a situation where selection depends
on the other replicators as well as "natural" factors.
I'm not sure there's a guarantee of anything in such
a system.
---- Paul J. Gans
Paul J Gans
>[snip]
And here I always thought that those two things, each
in its own field, has the purpose of providing research
topics to grad students?
Was I wrong?
----- Paul J. Gans
Richard Uhrich
> Dissident <qq...@7600.net> wrote:
>
>>It just occurred to me that this is a standard problem in
>>contract law, and contracts have a way around it, called
>>a "seperability clause". You know, the one you always
>>
> [...]
>
>>What the Bible needs is a seperability clause. Now, since
>>so much Biblical scholarship involves reading meaning
>>
>
> There once was, kind of. Nowadays fundamentalists aren't usually
> told it, but there are writs that their forerunners found to be
> "not holy enough" and decanonicized. The Book of Enoch is a classic
> example.
And Luther almost dropped Revelations as just being too weird.
>
> And the process works the other way around, too :-)
> The Mormon scriptures are a classic example.
>
>
--
Richard Uhrich
---
Ignorance more frequently begets confidence than does knowledge. --
Charles Darwin
Paul J Gans
>Scott
It *is* a good question. I just don't think it can be
answered at this time.
If we had more than one intelligent species the answer
might be more clear. On the other hand, we might actually
have more than one intelligent species but deny it. That
makes the question more complicated because now we have
to define "intelligence". And I'm not going there... ;-)
---- Paul J. Gans
Paul J Gans
>In talk.origins I read this message from
>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>>In article <inun4votfvvhc6dpr...@4ax.com>,
>>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>>In talk.origins I read this message from
>>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>>
>>[snip]
>I take it that we both agree that we probably can't model
>Lamarkian evolution because it does not actually work.
[snip]
Of course we can model Lamarkian evolution. Inside
the computer we can do anything. We can create
an immortal beast with mutable "genes" that responds
to outside stimuli by changing its "genes" in the
direction the stimuli push it. It need not reproduce
at all.
There are many other more realistic models as well.
But it doesn't mean that in the real world that it
would work.
----- Paul J. Gans
Bobby D. Bryant
> And Luther almost dropped Revelations as just being too weird.
I have read a theory that it was actually written by someone in a Jewish
Gnostic sect.
Bobby D. Bryant
> A basic question about GAs: Do they typically include a scheme to
> occasionally let a "lucky bastard" get through in spite of its deficiencies?
There are various schemes commonly used to pick the reproducers for the
next generation. They all favor the more fit (for obvious reasons), but
for lots of the schemes that favoring is "soft".
In my experiments I've been letting the top 10% breed with others in the
top 10% in random pairings, but high on my to-do list is to change that to
a scheme that lets lots of the losers get lucky. (Geeks can always dream,
can't they?) I'll probably use something like this:
Sort the current generation into an array ranging DNA[best] .. DNA[worst].
for current in worst .. best loop ### i.e., backwards through the array
DNA[current] = breed( DNA[ current ], DNA[ random( best, current ) ]);
end loop;
This steps through the array backwards to that it can replace the DNA _in
situ_. Every DNA in the population gets to contribute to the new
population, though the better DNA will contribute more frequently. (The
random() function returns an index of some DNA in the range [best, current],
i.e. better than or equal to the current DNA. Some DNA will breed with
itself, resulting in new DNA that differs only by a mutation. If this is
undesirable it can be programmed around, but I left the algorithm in its
simplest form for clarity.)
If that algorithm leaves too much cruft in the breeding population (as
determined by observing its performance) then I'll change the breeding to
= breed( DNA[ random( best, current ) ], DNA[ random( best, current ) ])
to remove the guaranteed participation and increase the preference for
re-using the "good" DNA, but still let some of the cruft through without
having to get *too* lucky.
My reasoning for using such an algorithm is that letting a certain amount
of cruft through will help maintain diversity in the population, and may
actually produce good results that would not be obtainable in a single
step from two "good" DNA strings. The drawback is that I'll spend more of
my time evaluating crummy solutions. That drawback may not be as bad as
it sounds, since if I run on a population size of 1000 for 1000
generations I'll eventually throw out all but one of the million specimens
evaluated anyway. So I can afford to invest some time pursuing
sub-optimal solutions, so long as I don't pursue so many that I don't
spend enough time with the better solutions.
Bobby D. Bryant
> In article <pan.2003.02.13....@mail.utexas.edu>, Bobby D.
> Bryant <bdbr...@mail.utexas.edu> wrote:
>
> [snip]
>
>>I see no reason at all not to invoke GAs as demonstrations of principle,
>>especially in response to certain specific creationist claims.
>
> Why bother? Don't we have enough really good biological examples of
> natural selection in action?
Enough for everyone but creationists.
> GA's just muddy the waters and they aren't very good models of what
> happens with living organisms.
I disagree about muddying the waters. Lots of creationists muddy the
waters by appealing to non-existent principles, and GAs are useful for
rubbing their collective nose in the fact that certain of those principles
*are* non-existent. (Witness the lengths IDers and others will go through
to convince themselves that GAs don't really work.)
But I agree with what I think is your major point: if we want to
understand how the biological world works, there's no better source than
looking at the biological world.
Bobby D. Bryant
> I take it that we both agree that we probably can't model Lamarkian
> evolution because it does not actually work.
Taking Lamarckian evolution = "evolution by inheritance of acquired
characteristics", I think the "does not actually work" applies to the
biology rather than to the algorithm.
Though this is a good illustration of what Lawrence was getting at: just
because we can make a model work, that's no proof of what really happened
biologically.
For instance, suppose we wanted to use Lamarckian evolution for artificial
neural networks. We start by ordinary neuroevolution (as we call it),
i.e. we encode the network's architecture and weight values into some kind
of pseudo-DNA string and start evolving by fitness evaluations and
selection. To make it Lamarckian all you would have to do is _train_ each
network a bit after constructing it from the pseudo-DNA and before
evaluating it, and then put the _trained_ values of the weights back into
the DNA before breeding. That would make the acquired characteristics
heritable, and the whole thing would probably work pretty well.
The above scheme isn't used because it requires, in addition to an
ordinary fitness function, some kind of training mechanism for the
phenotype. You could easily use backpropagation or the like, but if you
have the training patterns needed for backpropagation you'd probably just
train the network that way from square one rather than adding evolution to
the mix. But the Lamarckian mechanism would be fun to try, and it might
actually prove useful for problems where you have training patterns but
backpropagation tends to get stuck on sub-optimal solutions.
[And there's a straightforward research project for some undergraduate's
senior research class.]
As it is, the current model for evolution is based on the biological
evidence rather than computational evidence; GAs are only useful for
refuting claims that the theoretical mechanism doesn't work even in
principle. Not a lot of people are refuting the Lamarckian mechanism (as
defined above) because not a lot of people are touting it as the actual
biological mechanism. The computational model described above is simply
irrelevant to discussions of biological evolution.
Bobby D. Bryant
> Sure. Imperfect replicators *must* show change as long as any selection
> method at all is operating.
>
> But computer experiments usually have fixed selection rules and no
> competition. Thus clearly imperfect replicators will move toward
> satisfying those rules.
>
> As I pointed out previously, it is very much like "simulated annealing"
> in that respect and simulated annealing works (or can work) very well.
>
> In fact, I'd bet that a clever person could prove this mathematically.
IIRC there is a proof that simulated annealing will actually discover the
optimal solution to a problem, but it requires an annealing rate too slow
to finish before the sun burns out. (Do notice that "IIRC", though!)
But there is no such proof for genetic algorithms. What we need
mathematically is a characterization of "problems" that lets us identify
which ones GAs will work well on and which ones GAs will work poorly on.
The much-discussed NFLT uses a definition of "problem" that is wholly
useless for such work. (AFAICT the NFLT is a dead-end result, in the
sense that it proves a curiosity but doesn't provide a platform for
guiding future investigations.)
I think the Schema Theorem and the "replicator equations" do provide
proofs that "better" DNA will come to dominate the population in a GA, but
that's a far cry from saying that the GA will actually produce a useful
solution to the problem at hand. (Because "better" can still be very,
very bad.)
Bigdakine
>From: Paul J Gans ga...@panix.com
>Date: 2/13/03 7:16 PM Hawaiian Standard Time
>Message-id: <b2huml$cht$3...@reader1.panix.com>
Microsoft is Satan, not God.
Stuart
Dr. Stuart A. Weinstein
Ewa Beach Institute of Tectonics
"To err is human, but to really foul things up
requires a creationist"
dandelion
> A basic question about GAs: Do they typically include a scheme to
> occasionally let a "lucky bastard" get through in spite of its
> deficiencies?
Some do, some don't. The most interesting models (to my taste) develop when
the system is open-ended: i.e. the fate of diginisms is determined purely
by other diginisms (and possibly age). Like in the real world, fitness is
thn determined socially.
If you are employing a GA in order to optimize, I see no real advantage in
getting some poor sod with virtually no fitness to reproduce, though it is
(emotionally) tempting.
As to GA's developing 'intelligence' I would not like to be seen to make any
predictions. First off, due to computational limitations (time and memory
space) models we do have are rather simplistic, with critters usually
taking up no more than (say) a few kilobytes (and even that's pretty big),
populations of no more than somewehere in the thousands (millions for
really huge computers like the Connection Machine) and environments that
are typically *very* limited.
However, some of the more impressive demonstrations (I remember one where
the diginisms consist of rectangular shapes and the object was grabbing a
cube) come up with some pretty 'intelligent' behavior.
But, given the limits of time, computing power and memory space, I would not
hold my breath. And even if they would, I'm not sure we would recognize it.
The intelligence (if it were to evolve) would be very specific to the
simulated universe the diginisms evolved in, and to these diginisms, the
programmer/operator would be God herself, in ever sensible definition of
the world.
If intelligence *did* evolve, it would be murder to switch off the computer.
--
This is an unauthorized cybernetic announcement.
dandelion
Of course not, silly. They can tech your students *both* good programming
(They feel the need for speed if programs run for several weeks) *and*
elementary biology.
--
Screw up your courage! You've screwed up everything else.
dandelion
> Who determines what "results" you're looking for?
In purely ersearch cases, no-one (see the traditional reference to 'Avida'
I'm going to make now (http://www.caltech.edu/avida). If you are amploying
GA's to provide a specific solution, the engineer will.
They are a piece of engeneering. We understand how it works andeither *with*
or *wthout* preset goals, they will work. You do not *need* an explicit
goal, your have the option of defining one.
If you do *not* specify any explicit fitness function (i.e. you do not
specify what is fit and what is not), fitness will be determined
'socially', those critters that succeed (for whatever reason) in
reproducing faster than others are fit, those tha do no, are not. In such
cases, the criteria for 'fitness' will shift dramatically during the run of
the GA.
I can still vividly remember getting 'Tierra' (Avida's forerunner) to work
on a dark evening in 1993. Tierra is an openended GA, there is not explicit
fitness function, just a limited amount of memory and processor cycles.
Just 'limited resources' like we have in the real universe.
I started with a creature of 80 instructions and after 2 days, had an
ecosystem containing creatures ranging from 14 to 45 instructions,
displaying symbiosis, parasitism and all kinds of other interesting
behavior I had never counted on (There were two critters that could *only*
reproduce in pairs, for instance). The original 80-instruction creatures
had all vanished.
In many cases the Theory of Gravitation is used to produce quite specific
results. Now you would not want to argue that the law of gravity *only*
works because someone specified a 'desired result' would yo?
> Your programs can't
> undergo selection unless you predetermine what's going to be a good
> result.
Bzzzzt. Wrong. See above.
> (Yes, I know there are some strange exceptions but in general
> the goal of the program has to be defined in advance.)
For 'programs in general' you are right, but GA's are the notable exception.
They are explicitly designed to come up with answers we do *not* specify in
advance.
If we would have to do that, GA's would be worthless.
> This is why genetic algorithms aren't a very good model for biological
> evolution. Evolution has no purpose or goal. Genetic algorithms do.
They have a 'goal' only if we would *like* them to have a goal. That demand
is usually driven by organisms called 'investors' who metabolize on 'cash
return' and 'shareholder value'
That's the reason most (mind you *most*) GA's have explicit goals. There is
however no absolute *need* for them to have one. Au contaire, most of the
really *interesting* ones do *not* have one.
If you are *really* interested, get avida and see for yourself.
--
Real Time, adj.:
Here and now, as opposed to fake time, which only occurs there
and then.
Felix Sheldon
> John Wilkins <wil...@wehi.edu.au> wrote:
<snip>
> >Their standard evasion^Hresponse is to say that computers are programmed
> >by inteligences, and that is why they work.
>
> >I think they worry about the implication God might be a software
> >engineer without a clue how to solve multivariate problems. Or
> >something.
>
> Say not so! Please, don't let God be a software engineer.
> Please?
>
> I don't want to worship Microsoft.
>
It's generally accepted that MS is 'playing for the other team'.
Their leader is of course, William Gates III. If we apply some
numerology (in ASCII, of course) the TRUE NATURE of this corporation
is REVEALED!!
python
def add(x,y): return x+y
reduce(add,map(ord, "BILLGATES")) + 3
== 666
Bobby D. Bryant
> In many cases the Theory of Gravitation is used to produce quite specific
> results. Now you would not want to argue that the law of gravity *only*
> works because someone specified a 'desired result' would yo?
I suspect creationists will be reduced to claiming that sort of thing
before it's over.
Scott
"Laurence A. Moran" <lam...@bioinfo.med.utoronto.ca> wrote in message
news:b2hj52$e3j$1...@bioinfo.med.utoronto.ca...
> In article <pan.2003.02.13....@mail.utexas.edu>,
> Bobby D. Bryant <bdbr...@mail.utexas.edu> wrote:
>
> [snip]
>
> >I see no reason at all not to invoke GAs as demonstrations of
> >principle, especially in response to certain specific creationist
> >claims.
>
> Why bother? Don't we have enough really good biological examples
> they aren't very good models of what happens with living
> organisms.
The advantage is observing a system evolve in *real* time. For example,
millions of years can be compressed into a week.
Scott
dandelion
I suppose. And any child sifting sand from stones creates 'order' by
applying some kinetic energy to create 'brownian motion'..
It's not the evidence is not there, they would refuse to see it if it
happened under their noses. And it will, once they let avida run for a day
or two.
Those lurkers, however, ho *are* curious will see it, too. Now that i a
predictio, ain't it? I mean, falsify-able and all? Repeatable?
You can *every* bit of the source (it's GPL'ed IIRC) to see if those raskal,
liberal, pinko-commu^h^h^h^h^hevolutionist rascals are pullin' yer leg.
Have fun!
http://dllab.caltech.edu/avida/
(In contrast to what have been posting. Ooops....)
--
SHIFT TO THE LEFT! SHIFT TO THE RIGHT!
POP UP, PUSH DOWN, BYTE, BYTE, BYTE!
dandelion
Bobby D. Bryant wrote:
> On Fri, 14 Feb 2003 11:28:09 +0000, dandelion wrote:
>
>> In many cases the Theory of Gravitation is used to produce quite specific
>> results. Now you would not want to argue that the law of gravity only
>> works because someone specified a 'desired result' would yo?
>
> I suspect creationists will be reduced to claiming that sort of thing
> before it's over.
>
I suppose. And any child sifting sand from stones creates 'order' by
applying some kinetic energy to create 'brownian motion'..
It's not the evidence is not there, they would refuse to see it if it
happened under their noses. And it will, once they let avida run for a day
or two.
Those lurkers, however, ho are curious will see it, too. Now that is a
prediction, ain't it? I mean, falsifyable and all? Repeatable?
You can check every bit of the source (it's GPL'ed IIRC) to see if those
steenking, liberal, pinko-commu^h^h^h^h^hevolutionist rascals aren't
pullin' yer leg.[1]
Have fun!
http://dllab.caltech.edu/avida/
(In contrast to what have been posting. Ooops....)
[1] Apologies to those defunct irony meters.
--
If you give Congress a chance to vote on both sides of an issue, it
will always do it.
-- Les Aspin, D., Wisconsin
Louann Miller
<bdbr...@mail.utexas.edu> wrote:
>On Fri, 14 Feb 2003 02:24:10 +0000, Laurence A. Moran wrote:
>
>> In article <pan.2003.02.13....@mail.utexas.edu>, Bobby D.
>> Bryant <bdbr...@mail.utexas.edu> wrote:
>>
>> [snip]
>>
>>>I see no reason at all not to invoke GAs as demonstrations of principle,
>>>especially in response to certain specific creationist claims.
>>
>> Why bother? Don't we have enough really good biological examples of
>> natural selection in action?
>
>Enough for everyone but creationists.
A statement which was true in 1859, when there were only a tiny number
of such examples, and remains true today when there are tons more.
Where n equals "number of pieces of evidence in favor of evolution by
natural selection,"
the number of pieces of evidence necessary for a creationist to accept
that evidence as conclusive will always equal n + x.
(No, not n + 1. X equals 'some very large number but I'm not going to
tell you what it is.')
Louann, who finished re-reading "Cryptonomicon" last night and may
well talk like this for a while until she gets back to normal.
Louann Miller
<bdbr...@mail.utexas.edu> wrote:
>On Fri, 14 Feb 2003 05:44:58 +0000, Richard Uhrich wrote:
>
>> And Luther almost dropped Revelations as just being too weird.
>
>I have read a theory that it was actually written by someone in a Jewish
>Gnostic sect.
At that point, Christianity _was_ a Jewish sect. Not sure about the
Gnostic part.
Laurence A. Moran
Matt Silberstein <mat...@ix.netcom.com> wrote:
>In talk.origins I read this message from
>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
[snip]
>>Your programs can't
>>undergo selection unless you predetermine what's going to be a good
>>result. (Yes, I know there are some strange exceptions but in general
>>the goal of the program has to be defined in advance.)
>
>You are confusing the solution and the mechanism that produces
>the solution. Sure, we human define the "problem", traveling
>salesman or whatever. But that does not predetermine the
>solution.
Really? I thought that the ideal solution was to find the shortest
possible route for the salesman and this goal is decided in advance.
You don't know when you start what the ideal shortest path actually
is but you do know that the shortest path is want you want and nothing
else will be acceptable. Each of the steps in the GA attempts to find a
shorter and shorter pathway for the salesman. The GA ignores other
possible solutions such as skipping one of the cities, using email to
sell the product, trading cities with a salesman in an adjacent
territory, travelling on Interstate Highways, buying a faster car,
or using a teleportation device.
Evolution isn't restricted to one kind of pre-determined solution to
a problem and it isn't, by any means, guaranteed to find the best
possible solution.
I'm not denying that GA's resemble the mechanism of biological natural
selection but I do object to using them to demonstrate that natural
selection is real. We have plenty of real-life examples of biological
natural selection so we don't need a computer simulation. Furthermore,
the computer model introduces a number of flawed concepts that make
it more difficult to explain real biological evolution. (Note that
when I use the word "evolution" I am not just referring to "natural
selection.")
In my opinion, the problems creating by bringing GA's into the argument
far outweigh any advantages.
Larry Moran
Richard Uhrich
(Neat book, huh? I read it a few years ago. A bit of something for
everyone.)
--
Richard Uhrich
---
Laurence A. Moran
[snip]
>>This is why genetic algorithms aren't a very good model for biological
>>evolution. Evolution has no purpose or goal. Genetic algorithms do.
>
>And here I always thought that those two things, each
>in its own field, has the purpose of providing research
>topics to grad students?
>
>Was I wrong?
Hmmmm ... I stand corrected.
Larry Moran
Laurence A. Moran
[snip]
>>I know most biologist will say human intelligence is chance, a fluke of
>>history and not science. But is it really? (I admit to being intuitively
>>skeptical to intelligence being a *too good to be true* story) In any case
>>it's a question I think ought to be asked from time to time.
>
>It *is* a good question. I just don't think it can be
>answered at this time.
>
>If we had more than one intelligent species the answer
>might be more clear. On the other hand, we might actually
>have more than one intelligent species but deny it. That
>makes the question more complicated because now we have
>to define "intelligence". And I'm not going there... ;-)
I've been watching CNN.
I'm not sure that there are *any* intelligent species on this planet.
Larry Moran
Nathan Urban
> I'm not denying that GA's resemble the mechanism of biological natural
> selection but I do object to using them to demonstrate that natural
> selection is real.
Does anyone try to use them to show that natural selection is real?
Most people use them to show that natural selection _works_. They are
simplified toy models, yes. You think that's a reason why they
shouldn't be brought into the discussion, but I think that's a reason
why they should: toy models are useful to illustrate fundamental
principles without having to worry about the myriad complicating
factors.
Matt Silberstein
<ga...@panix.com>:
>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>In talk.origins I read this message from
>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>
>>>In article <inun4votfvvhc6dpr...@4ax.com>,
>>>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>>>In talk.origins I read this message from
>>>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>>>
>>>[snip]
>
>>I take it that we both agree that we probably can't model
>>Lamarkian evolution because it does not actually work.
>
>[snip]
>
>Of course we can model Lamarkian evolution. Inside
>the computer we can do anything. We can create
>an immortal beast with mutable "genes" that responds
>to outside stimuli by changing its "genes" in the
>direction the stimuli push it. It need not reproduce
>at all.
Can we do this? That is not at all clear. How do we know how to
change the genes in order to get the affect we want? The only way
we can to that is with another, non-Lamarking process. In that
situation either we already know what we need, so we are not
really modeling anything. Or we don't already know the answer, so
we need a mechanism to figure out which genes to change. This is
John's point in response. What appears to be a Lamarkian process
(inheritance of acquired characteristics) on one level is really
Darwinian on a deeper level.
Let us look at a human example. Suppose those who work out and
get big muscles have more offspring. How can we figure out which
genes lead to bigger muscles? We need some other process to do
this. Either we already know what genetic changes lead to what
results or we don't. If we do, it is an uninteresting problem. If
we don't, then a darwinian process is needed.
>There are many other more realistic models as well.
>
>But it doesn't mean that in the real world that it
>would work.
--
Matt Silberstein TBC HRL OMM
Politics is the art of the possible
Bismarck, but not the ship
dandelion
> "Fross" <jay...@hotmail.com> wrote in message
> news:<r-WdnaHWSpW...@comcast.com>...
>> Software engineers are quickly discovering that the best way to get
>> complex A.I. (artificial intelligence) systems is to set up a program
>> that mimics
>> Darwinian evolution. They don't sit down and design these complex
>> systems
>> before hand and then execute their design. They simply set up a few
>> basic
>> rules and let a darwinian process bring forth the complex system. I
>> find it fascinating that we can apply the basic principles of evolution
>> to computer programs and we get unpredictable results that are usually
>> too
>> complex for the programmers to figure out. Sure their new program works,
>> but how?
>> Why don't creationists see this as a major point for evolution? If the
>> theory works on replicating software programs, why would it not work on
>> replicating DNA? Computer processors are slow and limited compared to
>> reality, which runs at light speed! Yet the basics of Darwinian
>> evolution
>> work on them. What would stop it from working in reality?
>>
>>
>> I've brought this up before in hopes that a creationist would give some
>> input. I haven't heard this issue addressed.
>>
>>
>> Best,
>>
>>
>> Fross
>
> Back in the late 1970's, before DOS ... before CPM, I worked along
> side early programmers that were making up their own language for 8088
> processors.
Late in the seventies, I was doing pretty much the same.
> They didn't have monitors or keyboards. They used toggles
> to enter lines of code into 4k of RAM.
1K of RAM, and you had a hex-keyboard. Required on-paper compiling and
linking, though. Been there, done that, T-shirt wore out ages ago. Your
pals weren't all tha smart of they used 'toggles'.
> And, that was a great leap in
> technology over the previous 30 years of developement.
You are missing it by a decade. The big advances were made in the 60's. with
languages like Algol and many simular, the notion of 'subroutine' and/or
procedure and/or function was developed, Operatings systems were invented
and developed, many of the basic problems were solved.
Late 70's UNIX was widely available and VAX's still ruled the world. Intels
8088 was a very well supportd machine. Zilog's Z80 was much better. The
late 70's and early 80's only saw the popularisation of these technieques
due to the development of ic's (and Intels famed 4040, which was a bit
before the 8080 which was way ahead of the 8088. Late seventies, we had the
Intel 8080 IIRC, not the 8088 which came early 80's)
Up until that time we had 'bit-slicers', and 43 bit words, puchcards, tapes
and null-modems. IBM 'desktops' with keyboards that weighed 5 kg, 8"
floppies that sounded like jet-engines when started and daisywheel
printers that sounded like a secretary on steroids.
Remember the good ol' days? I do. Let's chat. Remember the Commodore PET?
The early Acorn and BBC computers? the 6809? Z80?
> If, you can get
> that A.I. program to run on 4k of RAM without an environment or
> operating system, then, that would almost be comparable to DNA,
> evolving on its own
Pretty primitive for that day and age. Toggles are *definately*
pre-Dijkstra. That's early 60's crap. Late 70's you would have burned an
EPROM. They *were* available at the time. I know. I burned them (in both
the creative and destructive sense of the word).
I *still* have a scar on my hand from a nasty run in with a soldering iron.
> If, there was no intellegent design, there would be no A.I. program or
> environment to run it on.
If you had any knowledge on the subject, you would notice the thing that is
required is 'closure', the basic prerequisite of an evolving system. No set
of circumstances may arise such that the system crashes. Well, there is no
way in which two atoms or molecules can possibly unite (form a covalent
bond) that would crash the universe.
If you think differently, i would invite you to name such a reaction. If you
disagree that chemistry forms the basic hardware of life, please indicate
on what grounds you would make such an assumption. I'm sue science would be
interested.
Molecules exist in a world where many things happen simultaneously, in fact,
everything does (on a large scale!). That's what we perceive as the 'flow
of time'. That's many, many, many, many zillions of interactions all
happening at once, continuously. And none of it cn crash the universe.
Next we need something to replicate. In the biological/chemical sense of the
world, we need an autocatalityc set. Some seachring did indeed turn up
examples of autocatalytic sets, that is, sets of mulecules which calalize
the production of other members of the same set of molecules.
A quick google on "Autocataytic Peptide" turns up 1,630.hits. Please select
one.
See for instance:
http://www.santafe.edu/sfi/People/kauffman/sak-peptides.html
Peptides are chains of amino-acids. We *know* aminoacids form spontaneuously
in many circumstances. Exposure to hard-UV and the right substrate. We
*know* many organic compound were presents (C, N, O, S, H2O, etc), so it's
not *that* hard to conceive of some peptide (amino-acid polymer) being
formed that *does* in fact selfreplicate.
Practically the same goes for RNA, which is also known to be autocatalytic.
See for instance
http://www.nap.edu/html/creationism/origin.html
or
http://www.biologie.uni-freiburg.de/data/wagner/abstracts/d2_biebricher.html
<quote>
RNA Replication
Biebricher, C.K.
Max-Planck-Institute for Biophysical Chemistry, Am Fassberg 11, D-37070
Göttingen, Germany
Replication is a crucial step in the origin of life. It is now generally
accepted that the primary genetic material was RNA rather than DNA. In a
series of experiments, Orgel and coworkers have shown that partially primed
single-stranded RNA can be completed to double strand RNA under reasonably
prebiotic conditions. This reaction is not providing autocatalytic
replication, however, because the double strand can not be reactivated to
serve as a template for further replication rounds. Reactivation of the
double strand by thermic melting has been proposed, but is unlikely to
occur under plausible conditions. On the other hand, autocatalytic RNA
replication can be catalysed by a single enzyme, the RNA replicase, which
is coded for by many RNA viruses. This replication has two remarkable
features: (1) the replicase is highly selective in accepting templates for
replication, suggesting a participation of the RNA in the replication
process, and (2) it synthesizes a complementary replica strand and recycles
the template strand, both as single strands. Sequence comparison of
replicable RNA species suggests that the secondary structure of the RNA
participates in the post-replicative melting. Replicating RNA has been show
to undergo efficient Darwinian evolution, the ability of an RNA to direct
its own synthesis being a particularly simple phenotypic expression of a
genotype. While DNA-dependent RNA polymerases also catalyse RNA replication
when instructed by a specific RNA, no non-enzymic or ribozymic replication
system has yet been found. I suggest that single-stranded RNA replication
has been the first prebiotic system capable to Darwinian evolution.
LOCATION Lecture Hall II
DATE Sunday, April 5
TIME 04:30 pm
</quote>
If the is a replicator and sufficient raw material, Von Neumann tells us,
this critter will reproduce exponentially, until it runs out of raw
materials, and Von Neumann was a mathematician, not a biologist.
Hence we would (conceivably end up with two competing sets of replicators.
One based on aminoacids, one on U,C,G & T (IIRC, U instand of A, but feel
free to correct me). And we have the forces of evolution working.
We still have RNA. Prions (like the ones that cause the dreaded BSE) are
selfreplicating proteins (IIRC, again, feel free to correct me, but
remember the cites above for selfreplicating poly-peptides).
I surmise at a certain point a symbiosis between a peptide based replicator
system and an RNA based one prooved most efficient and DNA provided the
common ground. Since cell-membranes are in fact low-energy structures (they
form spontaneously, fatty acids in watery emulsion will usually do it), the
formation of the first cells does not seem to be quite that implausible.
AFAIK, the above sums up the scientific picture at the moment. All possible
thanks to chemistry (natures hardware platform), time and *known*
processes. Please feel fee to correct me if I have mised some point or
other.
In comparison, our flimsy hardware does not show closure, one bad
instruction wil throw the system into a fit. It will dump core (as a good
OS should) or crash completely (as most one chip CPU's would way back when)
Certainly not the (by 70's standards) outdated piece of crap you describe.
They had punchcards, tape, even CRT's. PDP's ruled the world. Unix was
available.
You're a decade off, maybe two. In techie terms, that's missing it by a aeon
or two.
May your credibiliy rest in peace.
--
Living in LA is like not having a date on Saturday night.
-- Candice Bergen
Scott
"Paul J Gans" <ga...@panix.com> wrote in message
news:b2hv6k$cht$6...@reader1.panix.com...
> Scott <sc...@nospam.net> wrote:
>
> >"Fross" <jay...@hotmail.com> wrote in message
> >news:r-WdnaHWSpW...@comcast.com...
> >> Software engineers are quickly discovering that the best way to get
> >complex
> >> A.I. (artificial intelligence) systems is to set up a program that
mimics
> >> Darwinian evolution. They don't sit down and design these complex
systems
> >> before hand and then execute their design. They simply set up a few
basic
> >> rules and let a darwinian process bring forth the complex system. I
> >find
> >> it fascinating that we can apply the basic principles of evolution to
> >> computer programs and we get unpredictable results that are usually too
> >> complex for the programmers to figure out. Sure their new program
works,
> >> but how?
>
>
> >Darwinian evolution and - for lack of a better word- *progressively*
evolve
> >to greater complexity in intelligence could Darwinian evolution likewise
be
> >expected to accomplish the same feat each time within an ecosystem were
it
> >rerun in the natural world?
>
>
> No. See my recent post on this. AI is not Darwinian.
>
> The overall movement toward more complex systems that
> we observe in the world around is is very likely due
> to starting with unicellular entities. They had no
> where to go but toward multicellularity.
>
> Even then, it took a long long time.
>
> But I do not think that there is anything in evolution
> that means that intelligence is inevitable.
>
Somewhere in the back of my mind I recall reading about evolutionary
complexity being driven by chaos theory so I did a search and found some
links related to this thread. I putting it here, Paul, only because the
first link has some links to Evolutionary computation and artificial life.
Looked interesting.
http://www.cea.uba.ar/aschu/links.html
http://www.santafe.edu/sfi/publications/Bulletins/bulletin-fall96/mcshea.htm
l
http://discuss.santafe.edu/robustness/stories/StoryReader$24
TGIF
Scott
Ken Cope
"Louann Miller" <loua...@yahoo.net> wrote in message
news:1f6q4vkor96m4k4jc...@4ax.com...
I hope you're not seeing posters in terms of whether they're wizards or
dwarfs, or ...
Be sure not to read _In the Beginning was the Command Line_ or you'll
be too busy hacking linux to post here.
dandelion
Dolphins. I agree with Douglas Adams on dolphins. Maybe Bonobo's.
--
Don't worry over what other people are thinking about you. They're too
busy worrying over what you are thinking about them.
Bobby D. Bryant
And the nice thing is, with free downloads and perhaps even on-line demos
available, lurkers and questioning creationists can *see* random
mutations, filtered by selection, as making "improvements" to a
population, right on their very own screen. What could be more
devastating to the tripe we constantly hear about how random mutations
always make things worse, etc., (and to the mendacious authorities
peddling that tripe) than for someone to *see* an animated demonstration of
what can really happen?
I think that rather than removing GAs from the discussion we should be
thinking about how to set up a demo site that explains some basic
principles and provides animated demos and/or screenshots that the curious
can peruse.
(p.s. -- I've linked this repeatedly in the past, but any recent lurkers
may want to visit my static screenshots and explanatory text at
http://www.cs.utexas.edu/users/bdbryant/talk-origins/. What I suggest
above should be something much nicer than this, but this is all I have for
now.)
Bobby D. Bryant
> Laurence A. Moran wrote:
>
>> I've been watching CNN.
>>
>> I'm not sure that there are *any* intelligent species on this planet.
>
> Dolphins. I agree with Douglas Adams on dolphins. Maybe Bonobo's.
Bonobo's dolphins?
dandelion
Yeah, his, too. Banonos.
(screw that ')
--
One can't proceed from the informal to the formal by formal means.
Louann Miller
<uhr...@san.rr.com> wrote:
>> Louann, who finished re-reading "Cryptonomicon" last night and may
>> well talk like this for a while until she gets back to normal.
>(Neat book, huh? I read it a few years ago. A bit of something for
>everyone.)
I found that I was spotting different funny parts than I had before.
Like when Shaftoe is in North Africa, on an airplane, thinking to
himself that the SAS pilots/crew have _way_ too many personal weapons
hung all over themselves and that that kind of attitude reminds him
way too much of Guadalcanal.
So he goes back to sleep on the deck in the cargo hold, and has to
remove three revolvers and several knives so he can lie down
comfortably. Although he keeps the grenades strapped to his chest,
because he never sleeps on his stomach.
BTW the mass-market paperback version (yes, mass market -- 1152 pages,
nearly cubical, spine should read 'contents under pressure' has some
spoiler pages for Quicksilver. The year is about 1710-1720. The major
characters in the snippet are one Enoch Root and a smartass Boston kid
named Ben. Enoch has been sent by someone called Sir Isaac, in
England, to visit a little spot called the Massachusets Bay Colony
Institute of Technologickal Arts.
I know what was in that cigar box now...
Louann
Louann Miller
<fil...@ozcot.com> wrote:
>> Louann, who finished re-reading "Cryptonomicon" last night and may
>> well talk like this for a while until she gets back to normal.
>
>I hope you're not seeing posters in terms of whether they're wizards or
>dwarfs, or ...
Well, I can definitely state that the level at which I think of
posters as dwarfs, hobbits, elves, etc. has not changed after this
recent read.
Guess my real name
<< Say not so! Please, don't let God be a software engineer.
Please?
I don't want to worship Microsoft.
>>
Fear not. God's not an EVIL software engineer.
"In the middle of the journey of my life,
I found myself in a dark wood
Where the straight way was lost.
Oh, it is hard to speak of what I saw there,
Which even in recall renews my fear."
Dante Alighieri
John Wilkins
There's your mistake right there... there are no intelligent species on
*CNN*. Stop watching it immediately.
A friend of mine, a lecturer in philosophy but otherwise a nice fellow,
noted that if the news is upsetting you, try not watching it and see how
your life proceeds. He is ignorant of daily news and a well adjusted
fellow (apart from the philosophy thing); and still manages to involve
himself in Issues of Concern to Society.
We live in an age of extremely efficient communications... run by
journalists. It's like giving responsibility for food distribution to
bulemics.
--
John Wilkins
"Listen to your heart, not the voices in your head" - Marge Simpson
John Wilkins
> J. Paul Gans took his tongue out of his cheek long enough to say:
>
> << Say not so! Please, don't let God be a software engineer.
> Please?
>
> I don't want to worship Microsoft.
> >>
>
> Fear not. God's not an EVIL software engineer.
>
That is to assert a logical contradiction, I believe. God is the
original BOFH.
Richard Uhrich
Damn! With that teaser, I guess i'll have to read Quicksilver. (BTW,
I'm embarassed to say I don't even recall a cigar box in Crypto.)
Bobby D. Bryant
> Be sure not to read _In the Beginning was the Command Line_ or you'll be
> too busy hacking linux to post here.
And by all means avoid the big white algorithms book, or you'll go so
insane that even Cthulhu won't want to possess you.
Matt Silberstein
lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>In article <pdoo4v4r2mrtecm50...@4ax.com>,
>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>In talk.origins I read this message from
>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>
>[snip]
>
>>>Your programs can't
>>>undergo selection unless you predetermine what's going to be a good
>>>result. (Yes, I know there are some strange exceptions but in general
>>>the goal of the program has to be defined in advance.)
>>
>>You are confusing the solution and the mechanism that produces
>>the solution. Sure, we human define the "problem", traveling
>>salesman or whatever. But that does not predetermine the
>>solution.
>
>Really? I thought that the ideal solution was to find the shortest
>possible route for the salesman and this goal is decided in advance.
That is the *goal*, not the *solution*. And that is the goal of
the "designer". The "goal" of the things in the system are, if
they have goals at all, to survive the next round. Just like
organisms have the "goal" of reproductive success. Perhaps some
"designer" wants lots of single celled organisms (and we biguns
are just noise), perhaps it wants feathers or bones or something
entirely different. And the mechanism of this solution is,
itself, quite distinct from the goal. (And the mechanism for
finding the solution is something else again.) Larry, I assume
you are right here, but you have to stop mixing up these levels.
To put it another way, artificial selection is still natural
selection. Human breed, say, apple for sweetness. That does not
mean the tree has the goal of producing sweet fruit.
>You don't know when you start what the ideal shortest path actually
>is but you do know that the shortest path is want you want and nothing
>else will be acceptable.
How are my desires relevant?
>Each of the steps in the GA attempts to find a
>shorter and shorter pathway for the salesman.
No more than each proto-bird tries to fly. The mechanism is
reproduction (which implies, to me at least, drift) and
selection, the test is shortest path. The mechanism in biology is
reproduction and selection. The test differers for various
organisms. And the "goal" is unknown, if there is such a goal.
>The GA ignores other
>possible solutions such as skipping one of the cities, using email to
>sell the product, trading cities with a salesman in an adjacent
>territory, travelling on Interstate Highways, buying a faster car,
>or using a teleportation device.
Yes, they are simplified models. Funny how models do that.
>Evolution isn't restricted to one kind of pre-determined solution to
>a problem and it isn't, by any means, guaranteed to find the best
>possible solution.
Evolution, in biology and not, does have the ability to get close
to the top. And it can find rather surprising solutions. GA
don't, so far, find ways to send email, but it took carbon a long
time to figure that out as well.
>I'm not denying that GA's resemble the mechanism of biological natural
>selection but I do object to using them to demonstrate that natural
>selection is real.
I agree with that objection. They demonstrate that NS can do
certain things. We can see other demonstrations of that in
biological organisms. To learn about life we need, first and
foremost, to look at life. If someone argues that NS has no power
then we can look at NS in a different medium. It answer that
question, whether that convinces anyone is a separate issue.
>We have plenty of real-life examples of biological
>natural selection so we don't need a computer simulation.
Need? No. We don't "need" any more examples to show that
evolution occurs and that creationism is wrong. None the less the
argument continues.
>Furthermore,
>the computer model introduces a number of flawed concepts that make
>it more difficult to explain real biological evolution. (Note that
>when I use the word "evolution" I am not just referring to "natural
>selection.")
Nor do I. Selection is an essential aspect of evolution, as are
mutation and reproduction. Drift, to my mind, is an inherent by
product of mutation and reproduction. Drift, like adaption, are
results of evolution.
>In my opinion, the problems creating by bringing GA's into the argument
>far outweigh any advantages.
That may be. It may be that different people are persuaded by
different examples. But I suspect that few people not otherwise
persuaded will find the GA examples compelling. Me, I see it
reverse: GAs are an example of how biology has things to teach us
about the rest of the world. Evolution's power in biology tells
us that it may also be a powerful tool elsewhere. I find that a
compelling argument, but I am not a creationist.
Laurence A. Moran
Matt Silberstein <mat...@ix.netcom.com> wrote:
>In talk.origins I read this message from
>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
[snip]
>>In my opinion, the problems creating by bringing GA's into the
>>argument far outweigh any advantages.
>
>That may be. It may be that different people are persuaded by
>different examples. But I suspect that few people not otherwise
>persuaded will find the GA examples compelling. Me, I see it
>reverse: GAs are an example of how biology has things to teach us
>about the rest of the world. Evolution's power in biology tells
>us that it may also be a powerful tool elsewhere. I find that a
>compelling argument, but I am not a creationist.
What do you mean when you refer to "evolution's power?" When you
say that "evolution's power" may also be a powerful tool
elsewhere I have a tendency to think that you don't understand
evolution. Perhaps I'm wrong. What is it about the "power" of
evolution that so impresses you when it led to one million
species of beetles? How are you going to use the "powerful tool"
that produced humans with wisdom teeth, appendixes, and sore
backs? I can't wait to see the GA's that build better airplanes
by modelling real evolution, which produces many badly desinged
species that teeter on the verge of extinction.
Larry Moran
Noelie S. Alito
news:pan.2003.02.14....@mail.utexas.edu...
> On Fri, 14 Feb 2003 01:58:35 +0000, Noelie S. Alito wrote:
>
> > A basic question about GAs: Do they typically include a scheme to
> > occasionally let a "lucky bastard" get through in spite of its deficiencies?
>
> There are various schemes commonly used to pick the reproducers for the
> next generation. They all favor the more fit (for obvious reasons), but
> for lots of the schemes that favoring is "soft".
>
> In my experiments I've been letting the top 10% breed with others in the
> top 10% in random pairings, but high on my to-do list is to change that to
> a scheme that lets lots of the losers get lucky. (Geeks can always dream,
> can't they?) I'll probably use something like this:
>
> Sort the current generation into an array ranging DNA[best] .. DNA[worst].
> for current in worst .. best loop ### i.e., backwards through the array
> DNA[current] = breed( DNA[ current ], DNA[ random( best, current ) ]);
> end loop;
>
> This steps through the array backwards to that it can replace the DNA _in
> situ_. Every DNA in the population gets to contribute to the new
> population, though the better DNA will contribute more frequently. (The
> random() function returns an index of some DNA in the range [best, current],
> i.e. better than or equal to the current DNA. Some DNA will breed with
> itself, resulting in new DNA that differs only by a mutation. If this is
> undesirable it can be programmed around, but I left the algorithm in its
> simplest form for clarity.)
Hokay. I should have mentioned that the background for my question
is an issue called "The Horizon Effect" that came up when we were
learning game theory: For a given lookahead [of a chess-like game],
an algorithm might cull paths due to emphasis on avoiding useful
sacrifices, because the benefits of the sacrifice are beyond the horizon
of the lookahead. This is essentially a matter of getting trapped in
local maxima of optimization, which is a consideration for GAs, of
course. It occurred to me that the GA equivalent of permitting
near-negative vectors would be "letting losers breed".
> If that algorithm leaves too much cruft in the breeding population (as
> determined by observing its performance) then I'll change the breeding to
>
> = breed( DNA[ random( best, current ) ], DNA[ random( best, current ) ])
>
> to remove the guaranteed participation and increase the preference for
> re-using the "good" DNA, but still let some of the cruft through without
> having to get *too* lucky.
Ah, like the occasional result of droit_du_seigneur (sensu Pratchett).
> My reasoning for using such an algorithm is that letting a certain amount
> of cruft through will help maintain diversity in the population, and may
> actually produce good results that would not be obtainable in a single
> step from two "good" DNA strings. The drawback is that I'll spend more of
> my time evaluating crummy solutions. That drawback may not be as bad as
> it sounds, since if I run on a population size of 1000 for 1000
> generations I'll eventually throw out all but one of the million specimens
> evaluated anyway. So I can afford to invest some time pursuing
> sub-optimal solutions, so long as I don't pursue so many that I don't
> spend enough time with the better solutions.
Gotta confess I'm a bit leery of the results of GAs, in expecting
that the resulting solution has all the qualities of the <ahem> human
retina.
Noelie
--
The perfect is the enemy of the good.
Noelie S. Alito
news:3e4cd025$0$23913$8fcf...@news.wanadoo.nl...
> Noelie S. Alito wrote:
>
> > A basic question about GAs: Do they typically include a scheme to
> > occasionally let a "lucky bastard" get through in spite of its
> > deficiencies?
>
> the system is open-ended: i.e. the fate of diginisms is determined purely
> by other diginisms (and possibly age). Like in the real world, fitness is
> thn determined socially.
>
> If you are employing a GA in order to optimize, I see no real advantage in
> getting some poor sod with virtually no fitness to reproduce, though it is
> (emotionally) tempting.
As I described in my response to Monsieur Bryant, I was checking
to see if that is ever used to avoid local maxima of fitness.
> As to GA's developing 'intelligence' I would not like to be seen to make any
> predictions. First off, due to computational limitations (time and memory
> space) models we do have are rather simplistic, with critters usually
> taking up no more than (say) a few kilobytes (and even that's pretty big),
> populations of no more than somewehere in the thousands (millions for
> really huge computers like the Connection Machine) and environments that
> are typically *very* limited.
I say we design a planet-computer which--uh, no, it's been done....
> However, some of the more impressive demonstrations (I remember one
> where the diginisms consist of rectangular shapes and the object was
> grabbing a cube) come up with some pretty 'intelligent' behavior.
Is that "intelligent" in the sense that "humans weren't likely to come
up with that unorthodox but functional solution"? (This calls to mind
the tale of Edison's assistant attempting to measure the volume of
a light bulb using sophisticated calculation, when Edison took the
bulb from him, filled it with water, and measured the water.)
> But, given the limits of time, computing power and memory space, I would not
> hold my breath. And even if they would, I'm not sure we would recognize it.
> The intelligence (if it were to evolve) would be very specific to the
> simulated universe the diginisms evolved in, and to these diginisms, the
> programmer/operator would be God herself, in ever sensible definition of
> the world.
Aye, and in that world, an entity with human-like genius might be the
environmental equivalent of an idiot.
> If intelligence *did* evolve, it would be murder to switch off the computer.
I find it interesting that we associate the [moral?] crime of murder with
the *intelligence* of the killee. (Similarly, in some states like Texas,
the likelihood of capital punishment is greater for those with lower
apparent intelligence.)
Noelie
--
MiBlack: You're that smart?
Vizzini: Have you ever heard of Plato, Aristotle, Socrates?
MiBlack: Yes.
Vizzini: Morons.
Bobby D. Bryant
> "dandelion" <dand...@meadow.org> wrote in message
> news:3e4cd025$0$23913$8fcf...@news.wanadoo.nl...
>> If you are employing a GA in order to optimize, I see no real advantage
>> in getting some poor sod with virtually no fitness to reproduce, though
>> it is (emotionally) tempting.
>
> As I described in my response to Monsieur Bryant, I was checking to see
> if that is ever used to avoid local maxima of fitness.
What people usually do when a GA gets stuck is to either (a) turn up the
mutation rate, or (b) start over from a different initial state. I
personally find your idea more appealing, though I've never had time to
try it out yet.
It's a classic example of the "exploration-exploitation" tradeoff. Do I
spend my precious CPU time exploiting the good stuff I've discovered so
far (e.g., breeding the best ones), or do I spend it exploring areas that
haven't started paying off yet (e.g., breeding the worst ones). Different
problems reward different strategies. Unfortunately, AFAIK no one knows a
rule of thumb that will tell you which is best for your current problem.
[BTW, lurkers should notice that "start over" is a perfectly legitimate
activity in a stochastic algorithm.]
>> If intelligence *did* evolve, it would be murder to switch off the
>> computer.
>
> I find it interesting that we associate the [moral?] crime of murder
> with the *intelligence* of the killee. (Similarly, in some states like
> Texas, the likelihood of capital punishment is greater for those with
> lower apparent intelligence.)
Actually it's a multi-modal thing. Low intelligence increases your
chances of being executed by the state, but it also increases your chances
of getting elected and supervising the execution of your own kind.
dandelion
> He is ignorant of daily news and a well adjusted
> fellow (apart from the philosophy thing); and still manages to involve
> himself in Issues of Concern to Society.
I threw out my television after 9/11. I don't want to have buildings
collapsing in my sitting-room. Ever since, I keep informed over the net,
rdio and by newspapers, whic have the big advantage of bein there whe I
want it, instead of the oher way around.
Standing there (sitting was not an option), watching it and *not* being able
to do *anything* ws too much. Turning it off never came to my mind. Exit
teevee.
> We live in an age of extremely efficient communications... run by
> journalists. It's like giving responsibility for food distribution to
> bulemics.
So make sure to get you news from more than one side. Speaking german sure
helps these days, or French. And then there's the old trio of reliabiliy :
"Who says so? Who is he? How does he know?". And an added "what's the
spin?" for any news-broadcast, interview and editorial you see.
They all have spin, but an accurate estimate of direction and
angular-velocity may giv you a good idea of what the true story is. I
always wonder what they are not telling me aswell as listening to what they
are telling me.
These days, you have to be your own investigative-reporter.
--
"This is a country where people are free to practice their religion,
regardless of race, creed, color, obesity, or number of dangling
keys ..."
Walter Bushell
like Homo the Sap?
--
Sartre was an optimist. He thought Hell was _other_ people.
Walter
Ian H Spedding
john.w...@bigpond.com says...
[...]
> We live in an age of extremely efficient communications... run by
> journalists. It's like giving responsibility for food distribution to
> bulemics.
You know, having worked in press libraries, I rather like that.
Would you have any objection to my incorporating it, properly
attributed, into a .sig file?
Ian
--
Ian H Spedding
Ken Cope
"Bobby D. Bryant" <bdbr...@mail.utexas.edu> wrote in message
news:pan.2003.02.15....@mail.utexas.edu...
MIT may get my money yet. Old dog, new tricks time... Perl, or Python?
John Wilkins
I worked as a copy boy in a metropolitan newspaper. I saw enough folk
who thought they were Superman to be impressed by journalists. By all
means use it, but I want my name attached. One day I am going to have to
tackle the subject of the media and information "flow", and I want to be
known as a curmudgeon (Larry being my model here).
John Wilkins
Until around 1700 or so, humans were extremely sparsely distributed. 300
years out of 3 million (OK, 150,000 for H.s.) is not impressive, and it
won't take a lot for us to dramatically crash our population, I think.
One decent epidemic, and all infrastructure will collapse.
Ian H Spedding
wil...@wehi.edu.au says...
I'd always intended to credit you as the author but I was
wondering how you prefer to be styled. As we know, no less an
authority than Pagano has called you a "philosopher-comedian" so
I thought of combining that with your ambition to be a curmudgeon
and refer to you as either a 'philosopher-curmudian' or a
'philosopher-comedgeon'. Alternatively, you could just have the
letters CPC - for Curmudgeonly Philosopher Comedian - after your
name.
Dave
> John Wilkins wrote:
>
>> He is ignorant of daily news and a well adjusted
>> fellow (apart from the philosophy thing); and still manages to involve
>> himself in Issues of Concern to Society.
>
> I threw out my television after 9/11. I don't want to have buildings
> collapsing in my sitting-room. Ever since, I keep informed over the net,
> rdio and by newspapers, whic have the big advantage of bein there whe I
> want it, instead of the oher way around.
You were late... I've never had one, it has always been radio, newspapers
and net. I saw CNN for the first time in my life last night (literally),
and was absolutely shocked at the "sincerity" of Tom Ridge, and the
absolute inanity of half of the comments. I'm convinced now that I will
never want one.
Likewise, I have *never* seen TV footage of the WTC collapse. I don't
particularly want to --- I'm sure it was awful and some very bad people did
it, but I'm not inclined to trust CNN to tell me how bad it was and or
about the people who did it.
I cannot imagine what it is like to be glued to CNN, and to believe what one
hears and sees there. OTOH, I can understand how the US Govt can get away
with such irresponsible policy if people are glued to it.
> Standing there (sitting was not an option), watching it and not being able
> to do anything ws too much. Turning it off never came to my mind. Exit
> teevee.
Oh, but don't you miss the good documentaries? I hate that line... there
are good books.
> So make sure to get you news from more than one side. Speaking german sure
> helps these days, or French. And then there's the old trio of reliabiliy :
> "Who says so? Who is he? How does he know?". And an added "what's the
> spin?" for any news-broadcast, interview and editorial you see.
>
> They all have spin, but an accurate estimate of direction and
> angular-velocity may giv you a good idea of what the true story is. I
> always wonder what they are not telling me aswell as listening to what
> they are telling me.
Nice. I like the ABC = Australian Broadcasting Corporation. Fairly nicely
independent of the Govt. There are some good newspapers.
> These days, you have to be your own investigative-reporter.
Let me know if anything interesting happens in the world.
--
Attack is not the best form of defense.
Gavin Tabor
Bobby D. Bryant wrote:
> On Fri, 14 Feb 2003 01:58:35 +0000, Noelie S. Alito wrote:
>
>
>>A basic question about GAs: Do they typically include a scheme to
>>occasionally let a "lucky bastard" get through in spite of its deficiencies?
>>
>
> There are various schemes commonly used to pick the reproducers for the
> next generation. They all favor the more fit (for obvious reasons), but
> for lots of the schemes that favoring is "soft".
>
> In my experiments I've been letting the top 10% breed with others in the
> top 10% in random pairings, but high on my to-do list is to change that to
> a scheme that lets lots of the losers get lucky. (Geeks can always dream,
> can't they?) I'll probably use something like this:
>
> Sort the current generation into an array ranging DNA[best] .. DNA[worst].
> for current in worst .. best loop ### i.e., backwards through the array
> DNA[current] = breed( DNA[ current ], DNA[ random( best, current ) ]);
> end loop;
>
> This steps through the array backwards to that it can replace the DNA _in
> situ_. Every DNA in the population gets to contribute to the new
> population, though the better DNA will contribute more frequently. (The
> random() function returns an index of some DNA in the range [best, current],
> i.e. better than or equal to the current DNA. Some DNA will breed with
> itself, resulting in new DNA that differs only by a mutation. If this is
> undesirable it can be programmed around, but I left the algorithm in its
> simplest form for clarity.)
>
> If that algorithm leaves too much cruft in the breeding population (as
> determined by observing its performance) then I'll change the breeding to
>
> = breed( DNA[ random( best, current ) ], DNA[ random( best, current ) ])
>
> to remove the guaranteed participation and increase the preference for
> re-using the "good" DNA, but still let some of the cruft through without
> having to get *too* lucky.
>
> My reasoning for using such an algorithm is that letting a certain amount
> of cruft through will help maintain diversity in the population, and may
> actually produce good results that would not be obtainable in a single
> step from two "good" DNA strings. The drawback is that I'll spend more of
> my time evaluating crummy solutions. That drawback may not be as bad as
> it sounds, since if I run on a population size of 1000 for 1000
> generations I'll eventually throw out all but one of the million specimens
> evaluated anyway. So I can afford to invest some time pursuing
> sub-optimal solutions, so long as I don't pursue so many that I don't
> spend enough time with the better solutions.
>
>
Presumably by doing so you are leaving a route out to avoid getting
trapped in a local rather than a global optimum?
Gavin
--
Dr. Gavin Tabor
School of Engineering and Computer Science
Department of Engineering
University of Exeter
catshark
<ian_sp...@lineone.net> wrote:
Shouldn't that be (not working out all the permutations):
CPC - Curmudgeonly Philosophical Comedian, or
CCP - Comically Curmudgeonly Philosopher, or
PCC - Philosophically Comic Curmudgeon.
My personal favorite is the last.
---------------
J. Pieret
---------------
Nunc Id Vides, Nunc Ne Vides
- Unseen University Motto -
Bobby D. Bryant
> Bobby D. Bryant wrote:
>> My reasoning for using such an algorithm is that letting a certain
>> amount of cruft through will help maintain diversity in the population,
>> and may actually produce good results that would not be obtainable in a
>> single step from two "good" DNA strings. The drawback is that I'll
>> spend more of my time evaluating crummy solutions. That drawback may
>> not be as bad as it sounds, since if I run on a population size of 1000
>> for 1000 generations I'll eventually throw out all but one of the
>> million specimens evaluated anyway. So I can afford to invest some
>> time pursuing sub-optimal solutions, so long as I don't pursue so many
>> that I don't spend enough time with the better solutions.
> Presumably by doing so you are leaving a route out to avoid getting
> trapped in a local rather than a global optimum?
Yes. And as I said (or maybe only hinted) elsewhere in this thread, it's
probably a good idea to do that for some problems and a bad idea for
others.
Bobby D. Bryant
Sorry; don't understand the question.
dandelion
The tools formerly known as 'awk/sed/sh', new and revised, now on
Object-Oriented steroids.
The dichitomy i, however, false. Scripting languages are a dime a dozen,
these days. In german the phrase is 'Die qual der wahl', the agony of
choice. It's an agony worth having,. though.
--
Don't worry about the world coming to an end today. It's already
tomorrow in Australia.
-- Charles Schultz
Ken Cope
As publishers of the big white algorithms book, MIT Press would profit as
I'm finally considering learning some programming (as opposed to the very
occasional scripting). I'm leaning toward Perl, though I hear good things
about Python...
I've been playing with computers since putting a CPM card in my Apple IIe
for WordStar, and converting some lines of Basic I'd been learning on a Vax
to draw randomly sized and colored circles on my Sony (Macs? they're b&w!)
I don't know how I skipped the Amiga. I had Jim Blinn's advice to be an artist
and let programmers build the tools, but I couldn't escape doing 3D in the 80s
without using a lot of utilities and scripting and file format conversions (did
anybody ever tell you VI is a 3D modeling tool?) I used some emacs on a
refrigerator sized Symbolics machine, and the descendant of S-Geometry
is running on my Windows2K box, with its Lisp Listener still there. (A few
summers ago I converted some scavenged parts into a linux box; didn't like
SuSe so went with Red Hat, still just a toy for me.) These days, 3D apps
require you to build the workarounds to functionality they've left out, so
perhaps I'll knuckle down...that or just go back to pencils and dead trees...
Bobby D. Bryant
> Bobby D. Bryant wrote:
>
>> On Sun, 16 Feb 2003 18:56:01 +0000, Ken Cope wrote:
>>
>>> "Bobby D. Bryant" <bdbr...@mail.utexas.edu> wrote in message
>>> news:pan.2003.02.15....@mail.utexas.edu...
>>>> On Fri, 14 Feb 2003 17:40:37 +0000, Ken Cope wrote:
>>>>
>>>> > Be sure not to read _In the Beginning was the Command Line_ or
>>>> > you'll be too busy hacking linux to post here.
>>>>
>>>> And by all means avoid the big white algorithms book, or you'll go so
>>>> insane that even Cthulhu won't want to possess you.
>>>
>>> MIT may get my money yet. Old dog, new tricks time... Perl, or Python?
>>
>> Sorry; don't understand the question.
>
> The tools formerly known as 'awk/sed/sh', new and revised, now on
> Object-Oriented steroids.
Yes, I understand that. I just don't understand why they were mentioned
with a question mark after their names in the indicated context.
Bobby D. Bryant
> "Bobby D. Bryant" <bdbr...@mail.utexas.edu> wrote in message
> news:pan.2003.02.17....@mail.utexas.edu...
>> On Sun, 16 Feb 2003 18:56:01 +0000, Ken Cope wrote:
>> > MIT may get my money yet. Old dog, new tricks time... Perl, or
>> > Python?
>>
>> Sorry; don't understand the question.
>
> As publishers of the big white algorithms book, MIT Press would profit
> as I'm finally considering learning some programming (as opposed to the
> very occasional scripting). I'm leaning toward Perl, though I hear good
> things about Python...
For programming of the type in the BWAB, I would guess that Python would
be a better choice than Perl. But I don't actually use either (and I also
avoid using algorithms in my programs whenever I can help it, nyuk, nyuk),
so hold out for a second opinion.
Matt Silberstein
lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>In article <4nfs4v0l6btpdqeql...@4ax.com>,
>Matt Silberstein <mat...@ix.netcom.com> wrote:
>>In talk.origins I read this message from
>>lam...@bioinfo.med.utoronto.ca (Laurence A. Moran):
>
>[snip]
>
>>>In my opinion, the problems creating by bringing GA's into the
>>>argument far outweigh any advantages.
>>
>>That may be. It may be that different people are persuaded by
>>different examples. But I suspect that few people not otherwise
>>persuaded will find the GA examples compelling. Me, I see it
>>reverse: GAs are an example of how biology has things to teach us
>>about the rest of the world. Evolution's power in biology tells
>>us that it may also be a powerful tool elsewhere. I find that a
>>compelling argument, but I am not a creationist.
>
>What do you mean when you refer to "evolution's power?" When you
>say that "evolution's power" may also be a powerful tool
>elsewhere I have a tendency to think that you don't understand
>evolution.
Well please stop. I certainly don't understand it as well as you
do, I understand it better than many.
> Perhaps I'm wrong. What is it about the "power" of
>evolution that so impresses you
Evolution is one of a class of general problem solving methods.
The NFLT tells us that there is no magic bullet, no generally
best problem solving method. Different methods work for different
problem sets. I strongly suspect that evolution has two
interesting properties. First, and certainly important, we can
implement it. Solutions which you can't implement are not worth
much. Second, I strongly suspect that evolution is a good
solution finder when a better solution is similar to the current
best solution. That is, if you have a connected fitness landscape
evolution will help get you near peaks. And mutations will also
have the chance of getting you to another hill entirely in case
you are stuck at a local maximum. To put it another way, I think
evolution has the ability to find solutions in the kind of
problem space we tend to find in the world. As such it is very
powerful.
>when it led to one million
>species of beetles? How are you going to use the "powerful tool"
>that produced humans with wisdom teeth, appendixes, and sore
>backs? I can't wait to see the GA's that build better airplanes
>by modelling real evolution, which produces many badly desinged
>species that teeter on the verge of extinction.
First off, I care about the ability to find good solution. That
not as good solutions are also found it not all that relevant.
Second, the problem of biology are quite a bit more difficult
than the ones humans try to solve. Biological systems deal with
problems way more complex than any single human or human team.
People, engineers even, deal with *simply* problem compare to the
problem facing those beatles.
Finally, and here we might disagree, I think that biological
evolution is already astoundingly high up those fitness peaks.
You can point out, and have pointed out*, that we can see that
plenty of biological sub-systems are far from "optimum". Yes, but
it has been 3+ billion years of climbing those hills. Evolution
probably did a great job early one (once it got started). But the
interesting things about real world fitness landscapes is that
when you get to the top of the hill you don't find a pointy spot
or a nice flat meadow. You find a new set of bumpy hills. So, no,
extinction does not make me suspect that evolution does a bad
job.
*Off topic: what is the right construction there? Should it be:
1) You can, and have, pointed out
2) You can, and have, point out
3) You can point out, and have pointed out
I kind of like 1. 2 seems entirely wrong. I went with 3 because
it was not wrong even if it was awkward.
Don Cates
<mat...@ix.netcom.com> wrote:
[snip]
>*Off topic: what is the right construction there? Should it be:
>
>1) You can, and have, pointed out
>2) You can, and have, point out
>3) You can point out, and have pointed out
>
>I kind of like 1. 2 seems entirely wrong. I went with 3 because
>it was not wrong even if it was awkward.
How about:
3) You can, and (did/do), point out ...
>--
>
>Matt Silberstein TBC HRL OMM
[Hmmm, someone concerned about corect grammatical construction]
Who is this really?
--
Don Cates
ca...@cc.umanitoba.ca
Louann Miller
<d.philp.no....@mmb.usyd.edu.au> wrote:
>dandelion wrote:
>> I threw out my television after 9/11. I don't want to have buildings
>> collapsing in my sitting-room. Ever since, I keep informed over the net,
>> rdio and by newspapers, whic have the big advantage of bein there whe I
>> want it, instead of the oher way around.
>
>You were late... I've never had one, it has always been radio, newspapers
>and net.
>Oh, but don't you miss the good documentaries? I hate that line... there
>are good books.
Very few documentaries are as good as a book on the same subject --
simple bandwidth problem. The exceptions are the ones on something
like WWII where first-hand interviews of primary sources give much
more emotional depth and nuances to a subject.
I don't watch TV for the documentaries. I watch it to see what kind of
trouble Buffy's getting into this week. Sometimes I want a meaty text
full of solid facts, sometimes I want slapstick. I have no guilt
about indulging either taste.