But none of the letters in the book change. What are the odds of this happening when they are allowed to change?

It is actually possible to figure that out. I created a weasel program that works as Dawkins described (with no latching), but it also counts the number of times that a generation loses a character that had already been found.

Out of 50,000 runs of the program 211 runs had at least one generation where a character was lost. So the odds that Dawkins would stumble on one of the cases where a character was lost is about 1 in 236.

Actually the odds are even worse then that since Dawkins only showed every tenth generation or so, so even if there was a character lost it might not show up in the sampled generations.

Methinks the program “Methinks it is like a weasel” is like a weasel!

First, I have put up links for both Monash programs. I suggest you follow them.

Ah, thank you for pointing me to the GA version of weasel. I was wrong in saying that Monash has only one program. I apologize.

The second program is labeled as similar to Dawkins latter program.

I’m confused as to which you’re calling “first” and which you’re calling “second”. Monash calls them “Richard Dawkins’ Weasel” and “Genetic Algorithm Weasel,” but they each differ from the algorithm in TBW in different ways. The GA Weasel employs crossover, which results in very different behavior from the algorithm in TBW.

So I do not understand the point that the letters are different in the book from the random set that the Monash programs starts with.

That’s not the point. The point is that the first two generations of the TBW algorithm yield almost identical strings, while the first two generations of the Monash algorithm yield almost completely different strings.

It seems to me that the Monash program should actually converge quicker than one where only a limited number of letters can change.

You have to take into account population size along with mutation rate.

But the odds point to some fixed model.

How do you figure?

Also there is no indication that the mutation rate is the same for correct vs. incorrect letters.

And there’s no indication that the mutation rate is the same for the letter A vs. the letter B. But if there were more than one mutation rate, you would think he would say so in his description. Instead, he says, “It now ‘breeds from’ this random phrase. It duplicates it repeatedly, but with a certain chance of random error — ‘mutation’ — in the copying.” Of course this doesn’t prove wrong those who assume that Dawkins employed multiple mutation rates, but that assumption doesn’t seem very parsimonious.

Except that the correct letters never vary in any of the examples and in some of the latter iterations over 20 letters are correct so there would seem to be a high probability of one or more changing out.

That’s true only in the case of a small population and/or high mutation rate.

The most amusing thing about this is why people try to defend one position versus the other like it was life threatening or really meant something.

Agreed. And yet here we are defending our positions. But I’m not under the illusion that this discussion, or any of the other discussions on this site, really mean something. This is just an entertaining diversion.

Well maybe they should. If the letters represent traits then it is unlikely a trait will switch out to something not functional which is what changing the letters would indicate. There are error correcting mechanisms in the genome that prevent this so to lose a trait would be highly unlikely. So once a letter (trait) is at a desirable place, biological processes put a strong hold on it to remain the same.

So someone thinking biology when looking at this example, might naturally think once correct it would be fixed or highly unlikely to switch out. Thus a fixed model or one that is essentially fixed is more accurate.

If the letters are supposed to be amino acids, then I guess changes in each letter may be appropriate. Maybe the biologists should think it out. Either way, the actually simulation is amusing but nothing else.

Two things.

First, I have put up links for both Monash programs. I suggest you follow them. The second program is labeled as similar to Dawkins latter program. Whether Dawkins had more than one program is a trivial question (and a really unimportant one since the program has nothing to do with simulating evolution so to argue over it is pointless.)

Second, the program starts with a random set of letters so they will not be the same as the Dawkins book. The Blind Watchmaker uses three sets of an initial 28 letters and it is unlikely that you will ever see that same set of letters in any future simulation. So I do not understand the point that the letters are different in the book from the random set that the Monash programs starts with.

Because as you said the mutation rates are different for the fixed Monash program it is not the same as for the Dawkins program but I fail to see where it would make much of a difference for the fixed program. The non fixed letters are fungible so where they start each time seems to be irrelevant. It seems to me that the Monash program should actually converge quicker than one where only a limited number of letters can change. Because every non fixed letter has a chance of becoming fixed while if only a few changed each time then only these few could potentially become fixed in the next round.

I believe the second Monash program tried to mimic the program that Dawkins eventually had. Whether the latter Dawkins’ program is the same as his initial program is impossible to tell from the examples in the book. But the odds point to some fixed model.

“Dawkins’ version, which he describes in the book, has multiple offspring per generation, and there is no indication that the mutation rate is any different for correct vs. incorrect letters.”

Also there is no indication that the mutation rate is the same for correct vs. incorrect letters. Except that the correct letters never vary in any of the examples and in some of the latter iterations over 20 letters are correct so there would seem to be a high probability of one or more changing out. I understand that only every 10 iterations are listed but it seems unlikely one out of this many would switch out. So if one was examining the examples from the book, it would make sense to assume these letters were fixed.

As far as I am concerned, the discussion is over. As I said it is pointless to discuss this frivolous example. The most amusing thing about this is why people try to defend one position versus the other like it was life threatening or really meant something.

Monash had two Weasel programs. The first one fixes a letter once it was chosen and the second one allowed the letter to vary.

Monash currently has one program, which implements the algorithm that Dembski calls a “partitioned search,” complete with latching. I don’t see any indication of any other program on their site, although they do acknowledge that their algorithm differs from Dawkins’.

So one can see if the second program was used why one would think the first one was the actual program.

In [107] I showed the first two generations reported in Dawkins’ book, compared with the first two generations from the Monash version. It’s immediately apparent that they aren’t the same algorithm.

The book also shows an extremely rapid convergence on the target string. Is such a rapid convergence possible with the second program?

You bet. It all depends on the population size and mutation rate.

Also notice that Dawkins was extremely sloppy in the book. The first example starts out with 27 letters and the second example starts out with 29 letters when the target is 28 letters. If you look at the second line of each example you can determine the missing letter in example 1 and the added letter in example 2.

Yeah, there were definitely some typos.

The book certainly suggests the first program from Monash or a similar program was used where the letters are fixed and the second program was used in latter instances.

Monash’s version has only one offspring per generation, and has a mutation rate of 100% for incorrect letters and 0% for correct letters. Dawkins’ version, which he describes in the book, has multiple offspring per generation, and there is no indication that the mutation rate is any different for correct vs. incorrect letters.

There was a lot of that kind of analysis carried during the quasi-latching war, and more is available over at AtBC. You would actually have to write extra code to make the letters fix, the text is describing a simple algorithm, and would have need to go on and say explicitly that letters fix if that was the behavior desired. In general, these algorithms, called mu,lambda evolution strategies, do not fix individual parts of the genome.

Really, you should go over to AtBC, you are famous there!

But none of the letters in the book change. What are the odds of this happening when they are allowed to change?

“The scant evidence in the book is consistent with the algorithm described in the text. ”

There is nothing in the book that says the letters can be deselected and in fact none are. A latter example on television showed some of the letters were deselected. The data in the book is more in sync with the fixed algorithm.

“There is no need to assume the program was changed to a different algorithm. There are many kinds of evolutionary algorithm, and the one described can have rapid convergence to a solution on such a simple problem as Weasel.”

I asked about this but no one provided parameters that even got close to 41 or 43 iterations. The fixed program does this often.

If one was analyzing the data from the book, the obvious conclusion was that a fixed program was used.

But whether is was one or the other is meaningless since the algorithm is meaningless. It makes nice cocktail party discussion but nothing more.

There were a couple threads here a few months ago on this and the total comments were close to a thousand or more. Most of it seemed to be over whether the first Monash program or something similar was used or whether the second program was used. The discussion was complete and utter inanity.

Here are the links to the two Monash programs. Play with them and have fun.

Fixes letters:

http://vlab.infotech.monash.ed.....in-weasel/

There is a link on the above page that takes you to another program that allows for letters to vary once selected and allows for other inputs such as mutation rate and population size. Here is that link

http://vlab.infotech.monash.ed.....hm-weasel/