The battle over cumulative selection and Dawkins’ Weasel program has raged on for some months [years?] here at TSZ and across numerous threads. So can it possibly be that we now, finally, have a definitive statement about cumulative selection?
Mung: And whether or not my program demonstrates the power of cumulative selection has not been settled…
To which keiths responded:
keiths: Anyone who understands cumulative selection can see that it doesn’t, because your fitness functions don’t reward proximity to the target — only an exact match. The fitness landscapes are flat except for a spike at the site of the target.
So there you have it. You need a target and a fitness function that rewards proximity to the target.
Imagine my surprise when I discovered that I had said the exact same thing nine months ago.
Mung: 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.
Now perhaps I have misunderstood keiths here. Perhaps he did not really say, or really mean, what I think he said, or what it appears like he said. So I’d like to hear his response.
Is it possible that keiths has agreed with me all along while expending every effort possible to make it seem otherwise?
Just so there’s no mistake, here he is again saying the same thing:
keiths: Mung,
Besides failing in your attempt to code a Weasel and contradicting yourself regarding your intent, you also failed to demonstrate the power of cumulative selection in your program.
1) Your program doesn’t evolve a phrase; it evolves individual letters, one after the other, latching each one when it matches.
2) There is a separate fitness function for each letter.
3) The fitness functions don’t reward proximity to the target — they only reward an exact match for a single character.
The only thing your program demonstrates the “power” of is latching, not cumulative selection.
It’s a remarkable display of incompetence.
Perhaps. But it served its’ purpose. keiths admits I was right all along. So incompetence? Perhaps not.
You need a target. You need a fitness function that rewards proximity to the target. Is that your story keiths, and are you sticking to it? Weasel out of this!
I predict keiths will try to make this about my program and what it does or does not demonstrate rather than his revelation about cumulative selection.
I just want to be sure to give keiths his due for being honest about what needs to be present to demonstrate the power of cumulative selection.
But has he agreed with me all along about them or is this a sudden change of mind on his part, because for all this time I’ve been thinking that keiths does not understand Weasel or cumulative selection.
Fallacy fallacy, Mung. If someone in a discussion makes a less than perfect argument, it doesn’t make your side correct.
Your one and only evident skill is finding loopholes in other people’s definitions.
Definitions are always incomplete, even professionally written dictionary definitions.
One can accumulate fitness without a specific target.
Battle? What battle?
Oh, were you having a battle with yourself?
Faint praise for keiths there petrushka. 🙂
If my argument was less than perfect when I set forth the elements I thought needed to be present in order to demonstrate the power of cumulative selection, does it follow that I was wrong, or does this only work one way?
However, you’re not just saying his argument was less than perfect, you are claiming it is flat out wrong. But let’s not lose sight of the essential point of the OP, which is that keiths and I actually agree on something!
😀
Mung, what is the target in the travelling salesman scenario?
Selves, Neil. I am legion.
Seriously, petrushka? I mean, are you being serious?
http://bfy.tw/7eQP
So, your entire argument is over Keiths’ incorrect use of the word “target” rather than using something like “end result” or “currently observed trait”.
We both know what Keiths meant by his poorly chosen words. If not, you have not been paying attention to anything anyone has said. But I guess when one side has as little going for it as ID does, clutching at straws is all you have left.
Maybe you should write a paper about this and submit it to Biocomplexity. Your paper would double the number of publications for 2016.
Another mind reader. You and who else know what he meant?
Words may be poorly chosen and thus may not at all be a reliable indicator of what someone was thinking. Heck, perhaps Acartia’s words were poorly chosen!
I do find myself in a bit of a quandary though. I mean, who can resist the argument that keiths chooses his words poorly?
Why don’t you share with all of us what keiths really meant, substituting “end result” or “currently observed trait” where he actually meant to use those terms rather than the words he actually used.
keiths is being pretty consistent here. I don’t think we ought to be claiming he was just choosing his words poorly.
Is this what ID science looks like?
By your own poorly chosen words, lol.
It’s the science of poorly chosen words where everyone knows what keiths really meant in spite of what he actually said.
Mung,
I don’t think I’ve been following it as closely as you. I don’t think anyone has. Times are hard when the rules lawyer is chasing ambulances.
EV, Richardthughes, EV.
It finally paid off, and it sucks for keiths and his sycophants.
All these threads on GA’s and you haven’t been following along.
Whatever.
Aw, Richardthughes. What a poor choice of words!
How to disagree with keiths without actually appearing to have disagreed with keiths.
Because keiths and I agree.
Who will we throw under the bus?
Poor keiths.
Who are these sycophants of which you speak?
The name “newton” comes to mind. Anything to avoid the issues presented in the OP.
A standard keiths fail, of the kind we’ve come to expect from keiths!
But now we know at least two things:
Cumulative selection requires a target.
Cumulative selection requires a fitness function that rewards proximity to the target.
All that sounds vaguely familiar.
Mung: I’ve never doubted the power of cumulative selection.
Neil: I have. And I still do. That’s why I say that I am not a Darwinist.
click here
WTF Neil? Maybe this isn’t the thread for you.
Let’s just chalk this up to a poor choice of words!
keiths has been consistent.
A target.
A fitness function that rewards proximity to the target.
Wow. A full-blown Sunday night Mung meltdown.
You should have waited until tomorrow. It would have been more alliterative.
Mung,
Your mistake is obvious, as usual. Cumulative selection doesn’t require a target, but Weasel does.
You told us that your program was a Weasel:
And you told us that it was intended to demonstrate the power of cumulative selection (while trying to downplay your failure at implementing a Weasel):
You failed at both, and now you’ve created yet another failing thread denying your failures. It’s a remarkable display of incompetence.
The lesson should be obvious: when you go up against Weasel, you fail. Again and again.
If you learn nothing else from your failures, could you at least learn that? Weasel won, Mung. It succeeds where you have failed. Accept that and move on.
And who said ID does not make predictions? Eh? Eh?
Irony ,thy name is mung
What is necessary to demonstrate the power of cumulative selection? I don’t think that it is necessary for there to be a fitness that is a function of proximity to a target. A few points:
All this discussion is in the context of a model which has a single genotype in the population. Population geneticists have come to call this an “origin-fixation model”. That is to be contrasted with models that have a population which has multiple genotypes.
The whole reason that Weasel programs arose is to refute a creationist falsehood. Creationist debaters make a big point of trying to convince their audiences that evolution “is random”. They know that people will refuse to believe that adaptations can arise by a purely random process, so if they can make the “random” trope stick, they can get people to reject natural selection by conflating it with a tornado-in-a-junkyard hypothesis. Dawkins’ program was put forward to refute that false characterization of evolutionary theory.
To have a program demonstrate the power of cumulative selection, it is necessary to have a fitness (or a stand-in for fitness which measures the degree of adaptation by using some other number). The fitness has to have property that it changes smoothly enough that once one has a higher fitness, one has a path upwards to even higher fitnesses. Dawkins’ Weasel uses a measure which involves distance to the target, but the fitness does not have to be set up that way.
There does not have to be a target. The original Weasel had one, and Dawkins acknowledged that having a target was not necessary in a model of evolution, and that this was a limitation of his model. If you have different fitnesses of different genotypes, the change will tend to go uphill, but there need be no explicit target. If someone wants to say “the target is there, it is simply higher fitness” then that would be semantic game-playing as far as I am concerned, not worth much discussion.
The demonstration of the power of cumulative selection is that the process can increase fitness extraordinarily faster than one could do by drawing genotypes one at a time at random from the whole space, tornado-in-a-junkyard style.
My conclusions about this thread: Mung is wrong about there needing to be a target. Mung is wrong about there needing to be a fitness that “rewards proximity to the target”. But … Mung is right that his program is a Weasel. It is latching version which has a bizarre mutational process that acts one letter at a time, moving on to the next position once that site latches. The “fitness” in Mung’s program is basically the same as in Dawkins’ Weasel, so I fail to see Keiths’s point when he criticizes it.
Am I supposed to write Dawkins’ or Dawkins’s when I describe the Weasel?
It’s still a fallacy fallacy.
Even if Keiths erred, an error made in an argument does not affect the claims being made.
I have no interest in figuring whether keiths actually erred. It is irrelevant.
Joe,
I don’t either, contrary to Mung’s tendentious characterization.
For Weasel to demonstrate the power of cumulative selection, its fitness function must reward proximity to the target, but that’s because Weasel already has a target.
In other circumstances, such as in Rich’s “math genome” threads, there is no specific target, but cumulative selection still operates.
That alone makes it a non-Weasel, and things just get worse from there.
No, it’s completely different.
1) Weasel’s fitness function looks at the entire phrase; Mung’s doesn’t.
2) Weasel’s fitness function is static; Mung’s changes, looking at different characters over time.
3) Weasel’s fitness function rewards proximity to the target; Mung’s doesn’t, rewarding only an exact match.
4) Weasel demonstrates the power of cumulative selection; Mung’s program doesn’t — it’s just a series of random latching searches for individual characters.
There’s no consensus, so you’re pretty much free to choose either without damage to your grammar cred.
So fine. Let’s see how much difference there is. Let’s do something dramatic: take Mung’s program and replace its fitness-proxy function with the one from Dawkins'[s] Weasel.
And … lo and behold, the result is exactly the same. Mung’s program changes one character at a time. And Dawkins'[s] fitness-proxy rewards that, even if it is looking at the whole genome. Nothing is happening, yet, in the rest of the genome so that is the same as looking at only that one character.
Joe,
Now do the opposite: Take Mung’s concept of fitness, which rewards only exact matches, and use it in Dawkins’s Weasel. You now have a vast, flat fitness landscape with only a single spike in it at the exact location of the target phrase. It’s effectively a random search. The program will churn uselessly, never locating the target.
The reason is clear. The “power” of Mung’s program doesn’t come from cumulative selection, It comes from the fact that he’s randomly searching very small spaces and then latching the matches.
Expand the character set so that there are as many characters as there are possible strings in the Weasel universe, and Mung’s program will never get past the first character. Random search is a poor strategy when the space is large.
Mung’s program isn’t a Weasel, and it fails to demonstrate the power of cumulative selectiion.
Well, people can judge this one for themselves. I would just note that applying Mung’s fitness scheme to a different case than Mung uses is not relevant, as far as I can see. For his own case, Mung used a scheme that did not do anything different than Dawkins'[s] scheme.
We could go ’round and ’round on this forever.
Joe,
Yet you were the one who suggested that we apply Dawkins’s fitness function to Mung’s program. If one is relevant, then so is the other.
Mung’s program uses random search to find individual characters which it then latches. Unlike Weasel’s, his fitness function provides no hill to climb. The character either matches or it doesn’t, and all non-matching characters have equal fitness. The program just keeps taking random stabs at it until a match is obtained. That isn’t how cumulative selection works.
Mung’s program isn’t a Weasel, and it doesn’t demonstrate the power of cumulative selection.
An amusing factoid on the apostrophe-s issue:
Keiths, I will leave you with the last word on whether Mung’s program is a Weasel. Which is not say I agree.
Joe,
If you disagree, then feel free to say why.
To me, a program that
a) fails to implement the Weasel algorithm that Dawkins so clearly described,
b) varies the fitness function as it runs,
c) uses random search instead of hill-climbing, and
d) doesn’t demonstrate the power of cumulative selection, which is what Weasel was intended to do,
clearly isn’t a Weasel.
You yourself wrote:
Mung’s program is a random search program. It randomly searches for the first character, latching when it stumbles upon a match. Then it randomly searches for the second character, latching it when it stumbles upon a match, and so on. Only by keeping the search space extremely small — the size of the character set — is the program able to make progress toward the target.
And:
Mung’s fitness function doesn’t have that property. It’s all-or-nothing. Non-matching characters have equal, low fitness. The matching character, and only the matching character, has high fitness. Thus the fitness landscape is flat, except for a spike at the location of an exact match. Rather than hill-climbing, Mung’s program wanders around the fitness landscape until it hits the spike.
It isn’t a Weasel, and it doesn’t demonstrate the power of cumulative selection.
keiths,
There is no power of cumulative selection, if you don’t decide beforehand what you are selecting for.
That’s why evolutionary algorithms are so laughable.
Far be it from keiths to admit he made a mistake.
From the OP:
keiths:
But he wasn’t really talking about cumulative selection, because cumulative selection doesn’t need anything that the Weasel program needs.
LoL!
Completely and utterly avoiding the point at issue, which is cumulative selection, not weaseliness.
Go back and read the OP.
That makes two of you. 🙂
Is you point about cumulative selection or about keiths?
Which just happens to converge on the target phrase. Is it magic, or did I just get lucky?
Yet you clearly claimed otherwise. Were you wrong then or are you wrong now?
So my program does have a target. Check.
I was wondering whether you were even willing to grant that much, lol!
Just saw this. I Don’t pay attention to flame wars, so I don’t even know what keiths said. How can I say it is wrong?
My point is generic. And it applies to anyone who addresses people rather than arguments. Finding an error in some’s argument doesn’t say anything about the correctness of either side.
Which is why it is silly to get worked up about such things.
You’re amusing keiths, but no longer credible.
It is clear that my program has a target, even you admit it does. It even manages to mutate the candidate string until the candidate string matches the target.
And if it has more than one fitness function it would be nice if you could point out for everyone to see just where in my code I hid all those other additional invisible fitness functions.
My program rewards proximity to the target, you just don’t appear to understand how it does so., which I find utterly mysterious, given that you can see Dawkins’ program approach the target. So it is with mine as well. Did you even bother to run it? I’m betting no.
By the way, could please address the OP?
By the way, keiths, could you pick a story and stick with it?
Are you asking us to believe that when you say “cumulative selection” what you really mean is “the Weasel program”? So when you say I don’t understand Weasel and I don’t understand cumulative selection you’re just being redundant? Nah, that can’t be right.
You’re claiming that cumulative selection operates in programs without targets, so you’re contradicting yourself.