Uncommon Descent Contest Question 10: Provide the Code for Dawkins’ WEASEL Program

_{Denyse O'Leary
August 26, 2009

Darwinism

2}_{Categories
Darwinism}

Share: Facebook; Twitter/X; LinkedIn; Flipboard; Print; Email

Special invitation for Richard Dawkins – but any civil person is entitled to enter.

There’s been some discussion here and elsewhere whether the the recent IEEE article by Dembski and Marks correctly characterizes Richard Dawkins’ famous METHINKS IT IS LIKE A WEASEL program.

Does the program ratchet correct letters or does it let them vary? One is a partitioned or stair-step search, the other a more realistic evolutionary search. From The Blind Watchmaker, where Dawkins describes the program, its performance suggests that it could be either of these options (though he doesn’t say).

On the other hand, from a (video-run of the program , go to 6:15), it seems to be the latter.

It’s easy enough to settle this question: Make the code for the program public. Perhaps Richard Dawkins himself or his friends at RichardDawkins.net can finally provide this code (apparently a program written in BASIC).

The prize is a copy of either Stephen Meyer’s new Signature in the Cell or Richard Dawkins’ soon-to-be-out The Greatest Show on Earth.

Should the winner choose the latter, I will ask Dawkins’s publicist to mail the copy. Given that at his site, he calls himself “the most formidable intellect in public discourse,” I would assume that if he signs the copy, it will be worth millions.

But wait. Let’s see that code first.

Comments

I propose the following experiment:
Take 100 programmer. Show them the two pages of Dawkins book. Ask them to realize the algorithm in the language of their choice in an hour of time.
Then, we can see which percentage understands the algorithm to latch explicitly, which fitness function is used in general, etc...DiEb_{August 29, 2009
August
08
Aug
29
29
2009
06:44 AM
6
06
44
AM
PDT}

Sorry about the bad link. Please go to the EIL site and compare the behavior of "Partitioned Search" and "Proximity Reward Search". I lowered the mutation rate to 3% to ensure I still get kf's so-called quasi-latching behavior with longer Search Phrases, and I had a lot of fun watching the Proximity Search get worse and worse (relative to the Partitioned) as I increased the length of the Search Phrase. Don't trust either me or kf, onlookers, go and try it for yourself. D&M's paper maintains that these searches are the same...DNA_Jock_{August 29, 2009
August
08
Aug
29
29
2009
06:37 AM
6
06
37
AM
PDT}

Onlookers: Simply compare what I wrote and DJ's critiques. See where the balance on the merits is for yourselves, and who is dealing with specifics vs who is making simple dismissals based on snippets out of context and caricatured. GEM of TKIkairosfocus_{August 29, 2009
August
08
Aug
29
29
2009
06:18 AM
6
06
18
AM
PDT}

Yowser. With reference to kf’s first post, where he points out (in his language) that Weasel uses tons of ‘active information’, i.e. it is better than a random search. He is making CRDs point for him. Thank you. In the second post, kf tries to deduce a lot from the fact that D&M’s exemplar goes from 2 hits to 7 in a single generation. He claims this is “rather unlikely” and concludes that both of the following must be true 1) it is clearly a didactic example, and not a real sample 2) it shows that their must be a “very large pool of mutant progeny would have had to be created so that far-tail highly improbable outcomes of multiple correct letters would show up among the generation of mutants” Obviously, these two cannot simultaneously be valid conclusions, but, not to worry, the premise is wrong: the chances of getting an iteration this good are one in 419, well within the bounds of reasonable ‘data selection’, given the time and computing power available…. But the third post is the winner. Here kf argues that equation 22 describes the probability that the Qth query is correct, rather than the cumulative probability.
But, there is an evident subtlety there. This can be seen in that each of the L letters is free to come home at any point up to and including the Qth query. (So, it seems we are working on the odds of hitting the target AT the Qth try, not the cumulative odds up to that point.) This is fine for the context of comparative purposes as is shown in EQN 23 — on the similar odds to get to the target on simple random search — and following but it makes for a subtlety if one is expecting the cumulative odds to that point
Wow. Wow. Wow. This is so obviously wrong I don’t know how to explain it. Here goes: since (1-1/N) is less than one, (1-1/N)^Q tends towards zero as Q gets very large. Thus equation 22 tends towards 1 as Q gets very large (for any fixed value of L). It’s a cumulative probability distribution. As D&M make really clear in the text. To repeat : eqn 22 shows the cumulative probability that a partioned search will hit its target. It cannot be adjusted to account for a generation size greater than one. Go to EIL and compare “Proximity Reward Search” (Weasel) with “Partitioned Search”. Due to the way they cite TBW in their paper, D&M are claiming that these two beasts are the same, when they clearly know better. 2,740 words, all heat and no light. I guess I was wrong about Morton’s Demon.DNA_Jock_{August 29, 2009
August
08
Aug
29
29
2009
05:42 AM
5
05
42
AM
PDT}

III: Minor issues Several debate points have been raised that require a footnote or two: 1] Latching must only be explicit: We have of course demonstrated that it can be implicit, especially when one is looking at showcased "good" runs. That is, if there is a big enough population and a small enough mutation rate that we have a high probability fo no-change members in each generation, and single step changes are otherwise significant, then with an appropriate filter, we can fairly easily observe cases where once a letter goes correct, it will remain so all the way to the target. (The first such demo used 4%, 50 on Atom's adjustable weasel.) In short, latching, an aspect of ratcheted, cumulative progress to target, can be achieved explicitly and implicitly. 2] What about quasi-latching? Since we are looking at showcased runs, it is the cases that show no reversions that become what we need to explain. The fact that -- since, probabilistic barriers are inherently porous -- other runs on the same parameters may show occasional reversions therefore makes no relevant difference. (I used the term quasi-latched to predict such cases, which have ALSO been observed.) 3] Which Weasel shows implicit latching per Atom's lists of algorithms? Already answered but recirculated as though unanswered: PROXIMITY REWARD SEARCH. 4] Parsing EQN 22, p. 1055: In introducing the expression, M & D note: "Assuming uniformity, the probability of successfully identi-fying a specified letter with sample replacement at least once in Q queries is 1 - (1 - 1/N)^Q." So, in this analysis, they are seeing letters in parallel columns of search, each in effect independent of the others. N is number of letters in the alphabet, and Q the number of queries completed. The probability of being right are 1 - that of being wrong after Q tries. Eqn 22 extends this to L letters, L the length of the phrase. But, there is an evident subtlety there. This can be seen in that each of the L letters is free to come home at any point up to and including the Qth query. (So, it seems we are working on the odds of hitting the target AT the Qth try, not the cumulative odds up to that point.) This is fine for the context of comparative purposes as is shown in EQN 23 -- on the similar odds to get to the target on simple random search -- and following but it makes for a subtlety if one is expecting the cumulative odds to that point. 5] Partitioned search and weasel are different beats! Not so fast, pardnuh! Up to 2000, the information the public has had in hand was the showcased runs of 1986, and the description of cumulative targetted search. on this premise, it is reasonable -- thus legitimate -- to construct a weasel that uses the obvious mechanism, explicit latching. Subsequent to that time, we have had statements from Mr Dawkins and co that Weasel did not explicitly latch at any stage -- but mo credible c. 1986 code. (And BTW, commenter Apollos on this site has shown that explicitly latched weasels can be programed to show reversions. ONLY CREDIBLE CODE WILL BE DEMONSTRATIVE, IN THIS LIGHT.) We have accepted the testimony on the presumption of charity, and have demonstrated that latching is feasible on an IMPLICIT basis. (Note, this makes no difference to the already discussed point that weasel demonstrates the power of active information, not chance variation and natural selection based on complex function.) An explicitly latched weasel is a legitimate interpretation of the information given. And that is part of the point of the EIL demonstration: many weasels are possible and they give various patterns of behaviour, sometimes convergent, sometimes divergent. And in particular, implicit latching is a point of convergence between explicitly latched partitioned search and proximity reward search. 6] What about the BBC Horizon 1987 program? This has often been held to demonstrate that Weasel as originally coded did not latch. Of course, it now is apparent that the winking effect is possibly due to the video highlighting the members of generations, not the generational champions. A result that -- as was acknowledged above by the commenter who raised it -- is entirely compatible with: IMPLICIT LATCHING. Beyond that, the gap between 1986 and 1987 together with the demonstrated impact of shifting parameters on a proximity reward algorithm without EXPLICIT latching -- we can get latching, quasi-latching and far from latched behaviour -- indicates that the 1987 run does not ground a claim that Weasel c 1986 as showcased did not latch and ratchet to the target in a cumulative process. 7] Somebody was banned! This occasionally happens at UD, generally for abusive commentary; i.e for good -- though regrettable -- reason. (For comparison, there was a recent case where Mr Barrett Brown -- a columnist at HuffPo -- made some slanderous remarks there. A UD thread was put up to discuss the mater, and it turned out that at HuffPo: (i) comments were very restricted as to length [250 words in response to nearly 3,000], (ii) critical comments were subject to suppression on non-transparent grounds, with no defensible reason, (iii) slander and abuse were tolerated and even encouraged there. On balance UD is the better forum by far.) ______________ So, much of the above shows how Darwinist talking points can make the weaker appear the stronger case, to the unwary onlooker. So, let us take note. GEM of TKIkairosfocus_{August 29, 2009
August
08
Aug
29
29
2009
03:50 AM
3
03
50
AM
PDT}

Now, too, there are a few mopping up points that need to be addressed: II: The Dembski-Marks "Algorithm": Above, much has been made of an alleged algorithm developed by M & D, and how it so far differs from the Weasel 1986, to the point where issues of academic misrepresentation have been hinted at in purple passages. However, especially given the fact that EIL hosts Atom's adjustable weasel that makes its source code available -- hint, hint, Mr Dawkins -- and covers the various possible weasels admirably, this is a case of making a rhetorical mountain out of a mole-hill. A glance at the section of p 1055 of the IEEE paper will help us see why: _____________ >> Partitioned search [12] is a “divide and conquer” procedure best introduced by example. Consider the L =28 character phrase METHINKS ? IT ? IS ? LIKE ? A ? WEASEL. (19) Suppose that the result of our ?rst query of L =28 charac-ters is SCITAMROFN ? IYRANOITULOVE ? SAM. (20) Two of the letters {E, S} are in the correct position. They are shown in a bold font. In partitioned search, our search for these letters is finished. For the incorrect letters, we select 26 new letters and obtain OOT ? DENGISEDESEHT ? ERA?NETSIL. (21) Five new letters are found, bringing the cumulative tally of discovered characters to {T, S,E, ?,E, S,L}. All seven char-acters are ratcheted into place. The 19 new letters are chosen, and the process is repeated until the entire target phrase is found. >> _____________ 1 --> First, the example is explicitly given to illustrate what partitioned search is like, and starts with a set of letters in 20, then immediately proceeds to vary all incorrect letters, giving us 21 with several more correct letters [itself rather unlikely and a clue that this is didactic, not an algorithm in progress], then says "the process is repeated until the entire target phrase is found." 2 --> Let us, just for argument, accept that this is meant to be a realistic algorithm, with partitioning and 100% variation of letters remaining. 3 --> Now, in sampling with replacement, the idea is that once a letter may vary, it can take up any of the 27 available states. Of these, one is correct, and 26 are not. So, immediately, to plausibly gain a five-letter increment in proximity to target, a very large pool of mutant progeny would have had to be created so that far-tail highly improbable outcomes of multiple correct letters would show up among the generation of mutants. 4 --> And, a proximity to target filter would then be required to select the closest to target, which –voila! -- has five more correct letters. 5 --> So, immediately, if an alleged algorithm is being shown, it is credibly based on a very large population per generation and a proximity to target filter. (And, as the demonstration at 237 in the March 26 Wm A Dembski thread in this blog on simulation wars shows, such a case would race home to target through picking multiply correct mutants. For, there, we may see how a 999 pop, 8% all-letters may vary case runs home in 22 gens.) 6 --> Of course, such an imaginary algorithm would have to be explicitly latched. For as the case F from 239 in the thread -- a 999 pop, 25% case -- shows, with those high rates, reversions will otherwise make it hard to close the deal. 7 --> "GOTCHA!" Nope: STRAWMAN! 8 --> For, already, we have seen that the projected D & M algorithm shows that a ratcheting action will close to target, and that it implies targeted, proximity filtering based search that uses generations of sufficient size to see far tail population effects such as multiple go-correct mutations. And, that the projection is based on an illustrative pedagogical example rather than a sample run. So, there is plainly a caricature being set up to be knocked over here. 9 --> On a more reasonable, less rhetorically loaded discussion, M & D used a didactic illustration of what partitioning means, and how a partitioned search ratchets, then analysed what that implies about active information and its impact on success. 10 --> Now, we have long since seen from 234 in the Sim wars thread that such partitioning can be achieved not only EXPLICITLY, but also IMPLICITLY. Similarly, EIL just happens to sponsor Atom's adjustable weasel, which provides explicitly latched search and a second algorithm that under certain cases -- as in 234 Sim wars thread, etc -- will implicitly latch as pop per generation, mut rate per letter and filter interact. [Observe how those objectors who so stridently claim to see an algorithm convenient to their debating points in the IEEE paper glide by the provision of the adjustable weasel as quickly as they can, even repeating silly questions on which algorithm was used to do the implicitly latched demonstrations even after it had not only been obvious but explicitly answered.] 11 --> So, once we respect context, we can easily see that the illustration at p 1055 in the IEEE paper is didactic rather than a description of a serious algorithm. 12 --> But also, we see that the illustration even taken as a caricature is actually going to have to imply a very large population per generation and aggressive proximity filtering that can capture cases where multiple letters advance. 13 --> this is suggestive o0n a second point of debate as already addressed: what a "Query" means in the context of p. 1055. the answer is again plain, i.e my earlier analysis in what is now 131 is correct:
. . . let us now observe the lead in to Eqn 22 [on p. 1055]:
. . . the probability of identifying all L characters in Q queries is . . . .
1 –> That is, the natural point of completion of a query is the point where the issue of distance of approach to target is definitively addressed as at that time. 2 –> This, plainly, is not the point where mutants in any given generation have simply been generated by the seed with the per letter mutation rate applied [that just gives us a cluster of n mutants]. 3 –> Instead, it is the point where you have applied the selection filter in the generation and have chosen a champion, identifying its Hamming distance to the target as the least. Probability metric q (the subject of Eqn 22) is then assessing the likelihood that one is then on target, i.e. ALL L LETTERS are then correct after Q queries. 4 –> In short, just as I have already observed, the proper understanding of the point where a query is completed by the relevant algorithms is the point where you have a definite generation champion at a definite nearest distance to target so far.
14 --> So, there are no real -- as opposed to strawman --barriers to looking at the M & D analysis as an analysis of partitioned, ratcheted, cumulatively progressive search for a target, however achieved: explicitly or implicitly. 15 --> And that is what M & D present, and in so presenting they highlight the impact of active information coming from intelligent designers -- including of course Mr Clinton Richard Dawkins. [ . . . ]kairosfocus_{August 29, 2009
August
08
Aug
29
29
2009
03:49 AM
3
03
49
AM
PDT}

Mrs O'Leary (and onlookers): It is clear that no authentic code by CRD c. 1986 in Basic or Pascal will be forthcoming. So, we will have to work with the showcased print-offs and with the descriptions he has given. I: The impact of Active Information CRD's remarks in BW, 1986, bring out a remarkable key fact that we might easily miss in the clouds of digital ink above on all sorts of side issues and side-tracks. Namely, in 1986, Mr Dawkins inadvertently testified to the power of intelligent design over blind chance and necessity, and in so doing also testified to the impact of active information provided by intelligence in making a search for functionally specific complex information become feasible: ___________________ >> It [Weasel] . . . begins by choosing a random sequence of 28 letters … it duplicates it repeatedly, but with a certain chance of random error – ‘mutation’ – in the copying. The computer [indirectly, the programmer!] examines the mutant nonsense phrases, the ‘progeny’ of the original phrase, and chooses the one which, however slightly, most resembles the target phrase, METHINKS IT IS LIKE A WEASEL . . . . What matters is the difference between the time taken by cumulative selection, and the time which the same computer, working flat out at the same rate, would take to reach the target phrase if it were forced to use the other procedure of single-step selection [i.e. "more than a million million million times as long as the universe has so far existed"] . . . >> ___________________ a --> Weasel c. 1986 (and maybe 1987 too) is cumulatively progressive, targetted search that on the showcased sequences of generational champions, ratchets forward to the target [latching the already achieved successful letters, even if they are in "nonsense phrases"], through a filtering mechanism that picks the closest approach to target in each generation as the seed for the next. b --> It is therefore unsurprising that when CRD makes qualifying statements, he admits that Weasel "is misleading in important ways. One of these is that, in each generation of selective 'breeding', the mutant 'progeny' phrases were judged according to the criterion of resemblance to a distant ideal target." c --> So, Weasel is by its author's direct admission, misleading because it is foresighted, targetted search, search that rewards increments in proximity rather than of function [note his "nonsense phrases remark]. d --> Likewise, he admits that if instead Weasel had to produce functionality before any hill climbing could begin, it would fail, exhausting the available search resources. e --> And so, we see what accounts for the relatively remarkable capacity of cumulative, targetted selection on proximity: information abovut warmer/colder has been given, allowing for proximity-climbing. f --> i.e., the active information injected by the intelligent designer of the program has had a dramatic impact on the capability of Weasel. g --> It is equally worth noting that CRD acknowledges: "In real life, the criterion for selection is always short-term, either simple survival or, more generally, reproductive success." That is, present functionality is critical to the working of natural selection, and so the dismissal of "single step selection" rather begs the main question, and helps Weasel achieve an unwarranted degree of plausibility and influence of making it seem that chance variation and natural selection can credibly create novel complex functionally specific information. h --> So, Weasel needs to be put out to pasture as another exploded icon of evolution. (But ID-ers can then hire it back out of retirement as evidence of the power of active information based intelligent design.) ___________ A bit of an irony, nuh? [ . . . ]kairosfocus_{August 29, 2009
August
08
Aug
29
29
2009
03:48 AM
3
03
48
AM
PDT}

Oh, and btw, as far as I can see a few posts from DeLurker have been removed from this forum. This usually means that DeLurker has been banned, Joseph. So, you are probably arguing with an empty chair. This is easy (for most people, some even then mess it up!), but not much fun I suppose.Indium_{August 28, 2009
August
08
Aug
28
28
2009
09:03 AM
9
09
03
AM
PDT}

Joseph, yes, evolution tends to keep positive mutations in the population and so does Weasel. From the output in the 1987 video you can see that this is not true for every individual member of the population though. This is in line with the explanation of Weasel in TBW: The mutation rate is independend of the resulting fitness. This is, btw, the most important difference between Weasel and the partitioned search from the D+M article.Indium_{August 28, 2009
August
08
Aug
28
28
2009
07:21 AM
7
07
21
AM
PDT}

BTW by saying non-telic processes can mimic agency, that is very deceptive. Cumulative in a non-telic scenrio can only refer to "increasing by succesive additions".Joseph_{August 28, 2009
August
08
Aug
28
28
2009
07:19 AM
7
07
19
AM
PDT}

DeLurker, We cannot consider what we cannot observe. All we have to go by is what is in the book. That is why your comment in 41 is irrelevant.Joseph_{August 28, 2009
August
08
Aug
28
28
2009
07:17 AM
7
07
17
AM
PDT}

DNA_Jock: Be careful! You have just proven that evolution is wrong! ;-)Indium_{August 28, 2009
August
08
Aug
28
28
2009
07:15 AM
7
07
15
AM
PDT}

DeLurker:
Nonetheless, the Weasel algorithm as described can, in fact, take a step back.
Not as described by Dawkins in TBW.
You continue to ignore the fact that Dembski and Marks mischaracterized the Weasel algorithm in their IEEE paper
That is not a fact. And as a matter of fact the only inference one can take from TBW is that cumulative selection, as described by Dawkins and illustrated by the weasel program is a ratcheting process.Joseph_{August 28, 2009
August
08
Aug
28
28
2009
07:15 AM
7
07
15
AM
PDT}

Good idea, Indium, to point people to the EIL site, where they can see for themselves that partitioned search and Weasel are quite different beasts. My personal favorite - take parameters that produce kf's "quasi-latching", say gen size 100 and mutation rate of 10%, but then increase the length of the target string...At 57 characters, the partitioned search needs a median of 117 queries, but Weasel has gone past 36,000 (x100/gen = 3.6 MM offspring) without success. It is left as an exercise for the reader to find Weasel parameters that produce "apparent latching" but query runs over a thousand generations. Now, how do I adjust eqn22 to model these searches? That's a puzzler.DNA_Jock_{August 28, 2009
August
08
Aug
28
28
2009
07:09 AM
7
07
09
AM
PDT}

kairosfocus:
The proposal by Rob [that the BBC Horizon video is showing all pop members] would explain the winking effect on observed reversions
The winking effect has been explained several times by various people in the WEASEL discussions. Are you just now taking note of it?
but at the price — as he implies — of removing it as evidence that implicit ratcheting/latching is presumably “not” happening.
No, it doesn't remove it as evidence that implicit latching is not happening, because it never was such evidence.
Hitherto, the 1987 video was usually cited by objectors tot he concept of latching as “proof” of non-latching behaviour.
And we're exactly right, because we have always used the term latching to refer to what you call explicit latching, and the video does prove that the letters are not explicitly latched.
the proper interpretation of Q is as number of generations of query
Earlier it was "tell me just what part of . . . Q queries . . . it is that you do not understand." Now you're saying that a query isn't actually a single inquiry of the oracle, but rather a whole generation. Of course, that's what you meant all along, right? But you seem to have not put much thought into this. How do you explain the fact that M&D's math doesn't take into account the population size and mutation rate? Let's do the same math as #125 with Q=generations: q = (1-(1-(1/N))^Q)^L = (1-(1-(1/27))^45)^28 = .003 but it should be .5.R0b_{August 28, 2009
August
08
Aug
28
28
2009
07:02 AM
7
07
02
AM
PDT}

Joseph, admitting an error (even if it is not your own) is not usual here on this board, I applaud your honesty in #150. Cheers!Indium_{August 28, 2009
August
08
Aug
28
28
2009
07:00 AM
7
07
00
AM
PDT}

OK, let's imagine, just for the sake of illustration, that ref[12] described a Weasel program with an explicit latching mechanism. D&M would still wrong to equate ref[12] with a partitioned search. Eqn 22 describes a partitioned search. It cannot be used to describe a process that includes the selection of a generational champion. This has been pointed out to you a number of times, and yet you continue to assert.
ONCE RATCHETING EXISTS, THE MARKS AND DEMBSKI ANALYSIS APPLIES.
Stripping away the rhetoric, your argument appears to be "I cannot see the difference between eqn 22 (partitioned search) and a Latching-Weasel, therefore there is none." This sounds like an argument from personal stupidity, but I'll give you the benefit of the doubt here and blame Morton's Demon.DNA_Jock_{August 28, 2009
August
08
Aug
28
28
2009
06:55 AM
6
06
55
AM
PDT}

The Dembski/ Marks paper pertaining to a partitioned search uses a selection process similar to Dawkins in "WEASEL". They also use a population of 1. They also appear to be using a very high mutation rate- 5 correct letters appearing in one whack. They do not describe the starting sequence. IOW yes I see issues with their algorithm if it was supposed to exactly mimic Dawkins' "weasel". That said I don't believe that was their intent. Dawkins was trying to illustrate a different idea than Dembski/ Marks.Joseph_{August 28, 2009
August
08
Aug
28
28
2009
06:52 AM
6
06
52
AM
PDT}

kairosfocus#141
As to he case of the 43 gen program, I think the sample was nearly 90 letters in the sample that go correct and in the samples never revet. I suspect that a 90 or so sample will be representative, of the program as a whole
and Joseph#147
I know only a sample of outputs were shown in the book. In that sample we never observe a character reversal.
You are both ignoring my #41 and #78 in which I show that you are considering only approximately 0.047 percent of the strings generated by a typical run of the Weasel algorithm. Looking only at the most fit string every 10 generations is unlikely in the extreme to show reversions. Look at the algorithm description. It is very clear.DeLurker_{August 28, 2009
August
08
Aug
28
28
2009
06:52 AM
6
06
52
AM
PDT}

In the meantime you have ignored my question in two additional posts of yours, btw, kf. I will add a bonus question: Look at the number of queries Atom tracks for both algorithms. Do they match your definition = finding one champion is one query? This is rather irrelevant for the question wether D+M misrepresent Dawkins but it shows very nicely that you just make up your definitions as you go along.Indium_{August 28, 2009
August
08
Aug
28
28
2009
06:43 AM
6
06
43
AM
PDT}

DeLurker, I know only a sample of outputs were shown in the book. In that sample we never observe a character reversal. That is a fact. Take that with the definition of "cumulative" and his description of cumulative selection, there is no reason to think that character reversals are possible. Yes intlligent agencies can take a step back in order to gain ground going forward. But the weasel program is not imitating intelligent agencies. IOW your deception is duly noted. And if deception is all you have to make your case then you have already lost.Joseph_{August 28, 2009
August
08
Aug
28
28
2009
06:40 AM
6
06
40
AM
PDT}

KF: I was trying to point out that the search will work equally well for any equivalent gradient function. For this example the target is known and fixed, but that is not a requirement: The search will work equally well even if a specific target is unknown (granted we might not know immediately when the search is completed). My comments perhaps stray too far from the topic at hand, so I won't pursue this line of discussion. However ... It seems that the definition of latching and ratcheting being used is that of any successful search. By this standard any algorithm capable of climbing a gradient is latching and ratcheting, and that seems to included all possible search algorithms. What would it take to have a search algorithm that does not, implicitly or explicitly, ratchet and latch? I would suggest that an algorithm with non-zero probability of back-sliding (moving down the gradient instead of up) is not doing either. I confess some confusion about the concept of "active information". It appears to me like any use of information from the local or global search environment is being considered to as "active". This would then include all search algorithms other than a "blind" random walk, and I have a hard time considering that to be any sort of a search. Is there such a thing as a search algorithm that does not use active information?Tomato Addict_{August 28, 2009
August
08
Aug
28
28
2009
06:38 AM
6
06
38
AM
PDT}

Ok, let´s try again. Which of Atoms algorithms is the partitioned search and which one Weasel?Indium_{August 28, 2009
August
08
Aug
28
28
2009
06:28 AM
6
06
28
AM
PDT}

PS: Onlookers, do the same. is it fair to say that I have ignored "every argument and question"?kairosfocus_{August 28, 2009
August
08
Aug
28
28
2009
06:20 AM
6
06
20
AM
PDT}

Indium: I am not ignoring questions, I have answered them on the merits. Just, my answers do not fit your preconceptions. Why not pause and review the thread above back to say 131, and then go way back up to 66 and come back froward? (Not to mention my always linked, appendix 7.) While you are at it, reread pp. 1055 - 6 in the paper. Then come back on the issues. Okay, I really have to go now. GEM of TKIkairosfocus_{August 28, 2009
August
08
Aug
28
28
2009
06:18 AM
6
06
18
AM
PDT}

No, sorry, it doesn´t help at all because you just ignore every argument and question.Indium_{August 28, 2009
August
08
Aug
28
28
2009
06:11 AM
6
06
11
AM
PDT}

DL: Passing by. For the sake of onlookers, i will add some pointers. You have summarised implicit latching as though you disagree with me. As to he case of the 43 gen program, I think the sample was nearly 90 letters in the sample that go correct and in the samples never revet. I suspect that a 90 or so sample will be representative, of the program as a whole, especially as CRD used it to illustrate cumulative selection, which is of course a near synonym for ratcheting. In short, you are indulging here in selective hyperskepticism. Indium: Not at all, you are reading INTO the remarks on p 1055, an algorithm that is not there. the analysis is of ratcheting behaviour as observed, not of how it may arise -- apart from the associated effect of partitioning [which is produced by latching whether implicit or explicit]. This can happen explicitly and implicitly, i.e. the analysis of ratcheting is not algorithm-specific, but observation responsive. Implications of observed ratcheting, not inference to best explanatory mechanism. AND, you need to realise that for the past 6 or so months there have been wave after wave of objections tot he reality and significance of implicit latching, including some aspects of your remarks. Onlookers, hope this helps. G'day again GEM of TKIkairosfocus_{August 28, 2009
August
08
Aug
28
28
2009
06:04 AM
6
06
04
AM
PDT}

kf, you are so fixated on the implicit or explicit ratcheting or latching, it´s amazing. Please acknowledge that everybody has more or less understood this latching behaviour. From the output strings in the BW we can´t decide which internal behaviour the algorithm has. So, for the moment let´s pretend the latching issue is solved. But, what we can see very easily just from looking at EQ. 22 is that the analysis from Dembski and Marks is not correctly describing Weasel. The formula, plain and simple, is an incorrect description of Weasel. I don´t even need to do any calculation to prove it, I just have to note that the behaviour of Weasel strongly depends on mutation rates and population sizes, which is of course not true or EQ 22 or the partitioned search D+M in general. So, D+M use a completely different algorithm, the math is plainly incorrect for Dawkins Weasel. It is correct only for the partitioned search they describe in the text. You can check this easily once you finally let me know which of Atoms algorithms is Weasel and which one the partitioned search. But since you have ignored this questions about 10 times now I guess you will do it an eleventh time, too, right?Indium_{August 28, 2009
August
08
Aug
28
28
2009
05:52 AM
5
05
52
AM
PDT}

kairosfocus#135
You are confusing a description of the EFFECT of ratcheting with the mechanisms that generate it. As has been shown, ratcheting can be achieved explicitly OR implicitly. ONCE RATCHETING EXISTS, THE MARKS AND DEMBSKI ANALYSIS APPLIES.
The behavior of ratcheting only exists for certain values of population size and mutation rate, and even then only probabilistically. As Indium points out in #134:
Dawkins looks for a champion of a pool of sequences he builds by copying a parent string with random mutations (which are independent of the letter being correct or not). The chance to have the complete target string reproduced differs greatly from Eq. Nr. 22 from the D+M paper and obviously depends on population size and mutation rates, which are not used in the partinioned search D+M apply.
The Weasel algorithm documented by Dawkins and the the partitioned search algorithm documented by Dembski and Marks are completely different and not amenable to the same type of mathematical analysis. Further, your statement
Note that one of Dawkins’ examples, the 43 gen run, starts with three correct letters that never revert.
is incorrect. You have no way of knowing if those letters revert because only the best daughter strings from every tenth generation are shown. With Dawkins' clearly documented algorithm, it is mathematically possible for a reversion to occur. It is impossible for a reversion to occur in Dembski and Marks partitioned search. This brings us back to the core issue, namely that Dembski and Marks did mischaracterize Dawkins' algorithm in their IEEE paper.DeLurker_{August 28, 2009
August
08
Aug
28
28
2009
05:36 AM
5
05
36
AM
PDT}

DNA Jock: I see you can repeat the idea hitman talking points quite well. (And please, turn off the strawman and red herring factories: it should be clear that I am asking for a fair reading on the grounds that people of a certain background have a reasonable level of basic competence, not appealing to blind submission to the authority of peer review. [See what troubled waters reading to object lands you in?]) Now, please observe what has been pointed out, most recently in 131, and explain how ratcheting is observable apart from after a single representative of each generation has been selected, and for sufficient generations that we may reasonably see non-reversion. And, tell me how the generation of a cluster of mutants from a seed and selection based on closest approach does not generate a SINGLE champion as representative per generation; as we may see listed in the printoffs from Weasel 1986. The D & M analysis occurs in the context of such observable ratcheting, and addresses the question in the first instance of probability of hitting home by Query number Q. And, the focus of that is the identification of the member of the latest generation which has least Hamming distance to target. Att hat point, once we have a viable mechanism to selecty a champion per generaiton, we have a basis for an analysis of ratcehting behaviour, which is what Dembski and marks undertake. They have said precisely nothing about how we get to the closest-approach string, only that it is in the context that such strings, in succession ratchet towards the target. It is others who have asked how can that be, and our answer, with demonstration -- facilitated by Atom's adjustable weasel produced and supported by the same EIL -- is that Weasel algorithms can be latched explicitly or implicitly, and with algorithms that latch implicitly, that depends on matching of generation pop size, mutation rate per letter and proximity selection filter. And, we can show marches of implicitly latched champions, which would be the result of queries in the relevant sense, as discussed in 131. once we see the ratcheting, cumulative march to target in action, the M & D analysis applies. And so, onlookers, we again see the sort of problem Galileo faced: many of his critics refused to look through his telescope and examine the actual evidence on the merits as a basis for informed discussion. their idea hitman talking points were quirte persuasive to many who were schooled in their view of things, but lacked grounding in the evident facts. And here, the latest red herring and strawman models intended to dismiss the M & D analysis hinge on a distorted reading of what a query is in teh M & D analysis. of courese, all of this is w=ever so far form the real issue: 1 --> M & D have produced a significant analysis that introduces Active Information solidfly to the world of peers. 2 --> In so doing, they are accounting for the impact of itnelligence on the dcreation of fucntionaly specific complex informaiton 3 --> this extends the existing analysis of ID. 4 --> Last and least, Weasel c 1986 aptly illustrates the impact of such active information. indeed, without realising it, Dawkins says as much in BW:
It [Weasel] . . . begins by choosing a random sequence of 28 letters … it duplicates it repeatedly, but with a certain chance of random error – ‘mutation’ – in the copying. The computer examines the mutant nonsense phrases, the ‘progeny’ of the original phrase, and chooses the one which, however slightly, most resembles the target phrase, METHINKS IT IS LIKE A WEASEL . . . . What matters is the difference between the time taken by cumulative selection, and the time which the same computer, working flat out at the same rate, would take to reach the target phrase if it were forced to use the other procedure of single-step selection . . .
5 --> that difference of course comes out of active information. Onlookers, I think enough has been said for now to help spot the gaps in the objectors' latest set of idea hitman talking points. G'day. GEM of TKIkairosfocus_{August 28, 2009
August
08
Aug
28
28
2009
05:36 AM
5
05
36
AM
PDT}

Prev 1 … 6 7 8 9 10 … 13 Next