A couple recent comments here at TSZ by Patrick caught my eye.
First was the claim that arguments for the existence of God had been refuted.
I don’t agree that heaping a bunch of poor and refuted arguments together results in a strong argument.
Second was the claim that IDCists do not understand cumulative selection and Dawkins’ Weasel program.
The first time I think I was expecting more confusion on Ewert’s part about the power of cumulative selection (most IDCists still don’t understand Dawkin’s Weasel).
In this OP I’d like to concentrate on the second of these claims.
I’ve never doubted the power of cumulative selection. What I ask is, where does that power come from and how does the Dawkins Weasel program answer that question? I’m capable of coding a version of the Weasel program. But I’m also interested in the mathematics involved. I’d like to see if we can agree on what is involved in the coding and how that affects the mathematics involved. What is the power of cumulative selection?
METHINKS IT IS LIKE A WEASEL
A string with a length of 28 characters. Each position in the string can assume one of 27 different values. 27^28 would be the size of the sequence space. Call it the search space.
Many critics of ID claim that evolution is not a search. But “the power of cumulative selection” is dependent upon this view of evolution as a search. After all, the Weasel program is a search algorithm. My view is that the Weasel program incorporates and relies on information that is coded into the program to make the search tractable. The “power of cumulative selection” is an artifact of the design of the program. We lack justification to extend this “power” to biology.
So are there people here who can explain the mathematics and the probabilities and make the necessary connection to biological evolution? Is it a mistake to think that Dawkins’ Weasel program tells us anything about biological evolution.
Here’s what he Weasel program teaches us:
1.) In order to demonstrate the power of cumulative selection one must first define a target.
2.) In order to demonstrate the power of cumulative selection one must define a fitness function that increases the likelihood of the search algorithm to find the target relative to the likelihood of a blind search finding the target.
I have. And I still do. That’s why I say that I am not a Darwinist.
By itself, selection is a statistical filter. And I expect statistical filters to converge very slowly.
What drives evolution, is the introduction of innovation that comes from mutations. So we are not applying cumulative selection to something fixed. Rather, it is applied to something that is dynamically changing.
Count me as one of those who claim it is not a search.
Maybe the Weasel program is a search. I don’t know, because I haven’t studied it. But evolution doesn’t fit the description of “search”. There isn’t a specific that is being searched for, and what might be found is dynamically changing over time.
Sort of. I doesn’t have to be an explicit end state like “me thinks…”, for example Steiner trees.
In computer science, a search algorithm is an algorithm for finding an item with specified properties among a collection of items which are coded into a computer program, that look for clues to return what is wanted.
https://en.wikipedia.org/wiki/Search_algorithm
Maybe the Weasel program isn’t a search. Maybe it doesn’t have “an item with specified properties” (METHINKS IT IS LIKE A WEASEL) and maybe it doesn’t have a “fitness function” that allows the program to ” look for clues to return what is wanted.”
What do you think, Neil?
Mung,
I explained this to Eric Anderson at UD:
As I said previously, I have not studied the Weasel program. Real life evolution is not a search. I don’t know whether Weasel is a search.
Mung:
I already wrote a version in C for console I/O under Linux. Do you have access to a Linux machine?
Here’s the code.
Compile it using ‘gcc -std=c99 weasel.c -o weasel’.
I agree, this is one of the heart of the problems with writing simulations of the evolutionary process. You have to set up a “world”, an “environment” and then make your “organism” evolve in it. This means, necessarily, you can’t get around the fact that you will have to “put in” the parameters of this world you make.
But here’s where I think the discussion with ID proponents unhinge, because the very act of setting up this world/environment/fitnesslandscape to begin with comes with the accusation that the programmer has somehow “smuggled in” solutions to problems.
In a way, this seems to be simultaneously a consession by ID proponents that, of course if you have a world/environment, then evolution will automatically work in that world, while at the same time claiming that in order to have a world in the first place, that world must be designed.
This brings up another problem, which is that it seems to me ID proponents are trying to have their cake and eat it too. We are discussing whether evolution can work or not, not whether there has to be a world in the first place for something to evolve in. Of course there has to be a world for evolution to even be possible, but we were discussing whether evolution was possible, not how the world itself originated. In so far as we are using a computer to try to simulate evolution, we HAVE to first define a world to “test” our evolutionary process in. We simply don’t have a choice. Somebody necessarily has to sit down in front of their computer and write a fucking program, and for that program to be realistic, that programmer necessarily must take cues from how the real world works and design the same features into the program.
With respect to your latter point, the Weasel program merely tells us how selection can gradually adapt something “unfit” to very fit over successive generations. That is it.
But in the context of the former statement aobut sequence space, we need to start actually simulating a kind of physical world of sorts. A sequence has to be a polymer of sorts, and that polymer has to have properties that defines how it interacts with it’s surroundings, and those properties need to be the result of the sequence, such that the individual monomers in the sequence have specific properties.
The Weasel program doesn’t capture this feature. And we currently don’t have the technology to simulate entire populations of billions of such organisms made up of trillions of such polymers and then let them grow, divide and mutate. We just don’t have the hardware for it yet. That means, when we make simulations of evolution, out of a constraint of technology we are forced to make simplifications. Instead of simulating at the level of individual “proteins”, we are forced to simulate at the level of entire organisms with only a few “components”. Say, an overall simplified body-shape, an associated metabolic rate, a preferred food-sources and so on. This is unavoidably and regrettably extremely simplified and as such, cannot capture the entirety of the evolutionary process however much we want it to.
That means the only alternative we have, if we want to study evolution at the molecular level, is to do a combination of phylogenetics (comparative genetics, ancestral sequence reconstruction) and directed evolution (artifical selection) experiments with trying to evolve single polymers.
These then give us hints about how much “function” there is in the real “sequence space”. These experiments support evolution, they don’t contradict or refute it as you have been lead to believe by the flawed experiments done by Axe and Gauger.
No, One must first define an environment that the population has to adapt to (exploit its resources/avoid its hazards).
There is no target solution. There may be a vast number of solutions to thriving in that environment.
False. See the thread “Creating CSI with NS”
Fixed that for you. In this universe, fitness functions are smoother than random surfaces. Hence the inapplicability of NFL theorems.
ETA: Ninja’d by EL!
Try this variation of weasel. Push the Start button and see if you can figure out what is being searched for. What is the target?
http://itatsi.com
Dawkins’s Weasel was a teaching example, one intended to show that the changes in evolution are not “random”. Creationist debaters frequently use the talking point that evolutionary biology assumes that impressive adaptations can be achieved by a process of “random” change. Of course everyone knows that random changes, uninformed by natural selection, can’t produce such adaptations, so the creationist talking point is intended to get the audience to reject natural selection, by misrepresenting it. Dawkins was trying to show that cumulative natural selection is anything but random, that the talking point is a blatant falsehood.
His Weasel example was not intended as a realistic model of evolution. That did not stop creationist debaters from rejecting the Weasel example on the grounds that it is not a realistic model of evolution. Those creationist debaters really have very little shame.
More recently advocates of ID and creationists have taken to asserting that Dawkins’s example was intended to show where the information for an adaptation comes from, and announcing a “gotcha” — that the detailed information is coded into the program in advance. Again, this misrepresents the intention of this teaching example.
Mung may not make either of the first two misrepresentations, but he certainly is putting forth a principle that the detailed goal must be coded into the program in any evolutionary algorithm where natural selection is being modeled.
Not true at all. Look at Karl Sims’s “evolved virtual creatures” or the implementation of them in a screensaver, the
breveprogram, or a similar 2-dimensional environment calledboxcar2d.In all of these, motion in one direction is favored by natural selection, but the detailed structure arrived at is not coded into the software. Nevertheless the simulation achieves coordinated parts that effectively work together.
So Mung is wrong about that.
Elizabeth,
I don’t how you can say that. What computer program says, there are a lot of bombs in this area. And lighting storms. Plus not a lot of water. Find the best solution, we won’t tell you how, but you can randomly break things?
The answer is none.
I think claiming that because you don’t give a program the “detailed” solution, but rather a target means there is no target, is speaking without shame.
The question is, is Mung raising the point that in all evolutionary algorithms, the detailed target structure is coded into the algorithm. That is not true.
But if Mung’s “target” is just a fitness function that, say, gives higher fitness to organisms that move to the right, then (1) I was wrong to call him out, and (2) he has not established that the information that ends up encoded in the genome was present in the program. In the simulations that I cited, it is certainly true that no detailed information about the structure of the desired solution is present in the algorithm.
Minor clarification – you’ve quoted me quoted Mung.
It all depends what you mean by “target” I guess. If you mean a very specific configuration, no but if you mean a more abstract “incremental advantage” then yes. the single target is somewhat falsified by the diversity of life anyways.
Also, over at UD Mung says: “Nick seems to think you can take one book and make a copy of it and turn it into a different book by gradually altering the letters and sequences.”
Well I think that early life was more sentences than books. And that can be done. I don’t think anyone expects a crocodile to evolve into an eagle – that branching happened a long time ago and wouldn’t be accessible without some incredible regression first I suspect. Evolution can “paint itself into a corner”.. which can and more often that not does lead to extinction.
Editz4Spells
phoodoo,
My goodness, you’re confused. But lets *start* with one little change – Randomly *build* things.
Can you try to rewrite that post? I don’t understand what you’re trying to say.
The problem with the word target is that there is a vast network of targets, connectable by single point variations. The target landscape is akin to a foam, mostly empty, but all surfaces connectable.
If a solution in a typical EA is ‘smuggled in’, where does it reside – in the population strings, in the fitness function, in the operation of the algorithm, in a combination thereof?
It is unlikely to be in the strings, since you actually randomise them when you start. It is unlikely to be in the operation of the program, which simply evaluates these random strings against a fitness function at runtime. So the information is ‘smuggled in’ to the fitness function itself? Well, the fitness function can be randomly chosen at runtime. So it’s hard to see how this smuggling actually takes place, nor what the biological analogue would be – the Designer ‘smuggles in’ differential mortality due to cold, say, enriching the population in genetic constructs that deal better with cold?
In the Weasel example, differential mortality is due to being more or less like ‘METHINKS […]’. Is that smuggled in? Or is it not merely the reason for differential mortality in that run? Get the computer to pick n random ASCII characters as its target. What’s smuggled in then? Random strings locate a random target through a mathematically random mechanism.
The information is embodied in chemistry. Some configurations of atoms have properties compatible with life, and some don’t. The fact that viable configurations are connectable is no more remarkable that the fact that the surfaces of soap bubbles are connected.
Or that water follows the contours bottom of the pond.
I do not think it was intended to be a realistic model of evolution.
Does anyone else here think that it was?
It models the primary ID probability argument.
It doesn’t need to model biological evolution. It just needs to model the “isolated islands of function” argument that was being used by IDists.
What it demonstrates is not the biological evolution works, but that probability arguments against evolution don’t work.
Do you want to know, or just not interested?
One clue as the the power of WEASEL to dismember ID is the seemingly endless arguments over latching ath UD.
There were some very vocal people there who were concerned that if WEASEL could work without latching, it would actually be a threat to ID.
Of course, what the argument demonstrated was just how stupid some people there really are.
tell us what difference a word would make.
I’m talking about the Weasel program.
Do you think it was designed to demonstrate the power of cumulative selection?
I think that’s the explicit claim Dawkins made for it in his book. I could be wrong.
I had a similar “discussion” with Barry a few months ago. He kept arguing that the probability of a room full of monkeys typing the first verse of some specific Shakespeare tome was 10^??? (I forget the actual number he used, but it was astronomical). I agreed with him but I pointed out that this was not how evolution worked. I argued that evolution does not have a specific target in mind. That a better analogy, although still not a good one, is the probability of the same room of monkeys pounding out a similar sized string of letters that can be construed as English.
Barry, in typical Barry fashion, resorted to name calling and banning.
But my point is simply that there is no specific target in evolution. There are a finite (although large) number of possibilities from any given starting point, any of which may be of adaptive advantage.
Why are you asking me and not Neil?
He thinks evolution is not a search. As such, it’s difficult to see the relevance of any search algorithm to biological evolution. Weasel is an implementation of a search algorithm. That’s what difference a word makes.
I don’t much care.
For one thing, “search” is not well defined anyway.
As I see it, the environment is an important part of evolution. And computer programs are limited in their ability to adequately simulate the environment.
Maybe it’s just the fact that I was underwhelmed by “The Blind Watchmaker”, but I’ve long thought that the amount of effort spent debating Weasel-related topics is puzzling and vastly incommensurate with the importance of the program.
Yes, it was designed to show that cumulative selection could be far more powerful than purely random changes. Creationist debaters were saying that evolutionary biologists were claiming that adaptations arise “just at random”. That is a gross misrepresentation, so Dawkins came up with the Weasel example to counter that.
It does that very effectively. Pure random change of the text takes about
steps, while with natural selection it takes less than 1000 steps.
What it was not designed to do was to make a realistic model of evolution. Or to give us some deep insight as to where adaptive information comes from. That hasn’t prevented creationist debaters from playing “gotcha” by claiming that these were the intentions of the model.
Mung,
You’ll be unsurprised to find that I don’t either (did I mention the genetic code isn’t a code too …?).
An evolutionary algorithm can be used as a search. That doesn’t make the biology from which the method was borrowed a search method itself.
Weasel isn’t really a search anyway – you’re not looking for “METHINKS IT IS LIKE A WEASEL”; you already have it. You just want one of your genomes to say it too.
I think Dembski tried to define evolution as a search because searches are easy to model mathematically.
But “search” is loaded with implications. Mostly, the implication that there is a target or goal.
That makes Dembski’s understanding of biology rubbish, right out of the gate.
Mung:
I am not sure I agree with your conditions for a model to demonstrate the power of cumulative natural selection/
First of all there does not have to be a target. It could be like the 3d and 2d blocks world examples that I cited. One is more fit the faster rightwards one. There does not have to be a single best genotype that is accessible by the system. One need only demonstrate that the movement is much faster rightward than the movement of randomly mutating genotypes.
Yes, one does need a fitness surface that is appropriate. The question is, how hard is it to come up with one. I claim that it is only necessary to not have tight interactions between genotypes at typical loci.
Oops, typo: “… the faster rightwards one moves”
Then perhaps you should stop saying it is employing natural selection.
Cumulative selection implies a goal. Darwinian/ unguided/ blind watchmaker evolution is not a nor does it have a goal. And it cannot be modeled by employing a search heuristic.
So you’re searching for a genome that matches the target. Maybe you just don’t think that’s a “real search.”
Another way to look at it is that you’re searching for the fittest genome where the fitness is defined relative to the target phrase. The fittest genome is the one that matches the target phrase exactly. It’s still a search.
An Patrick thinks IDCists have problems understanding Weasel.
Perhaps there is some way to demonstrate “the power of cumulative selection” using a computer program that doesn’t involve goals or targets. A NotAWeasel program.
Mung,
What real world process would such a thing be modeling?
I think the closest you’ll find is Thomas Ray’s Tierra.
I linked to one. You ignored it.
What real-world process is Weasel modeling?
ETA: Perhaps cumulative selection doesn’t exist in the real world (except in computer programs).
Plant and animal breeding?
Mung,
It’s a very simple demonstration of the advantage of cumulative selection over random selection. It was never intended as any more than that.
So what would your “no goal, no target” GA model? Does Tierra qualify?
Nope. The phenomena we see in nature, and use in mathematical and computational models of nature, also exist in oversimplified models of nature. As here.
Well, look, we’re back to reality again. Chemistry. Physics. Proteins. Mutations that explore all possibilities in certain spaces.
Is it so improbable or impossible to imagine that random steps that work are kept preferentially, overall, given limited populations? I’m sure someone will be along to give you a more technical description but, then again, can’t you check those books you have?
If it does, does that change the way you view ‘Weasel’?