The Simulation Wars

_{William Dembski

March 26, 2009

Intelligent Design}

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

I’m currently writing an essay on computational vs. biological evolution. The applicability of computational evolution to biological evolution tends to be suspect because one can cook the simulations to obtain any desired result. Still, some of these evolutionary simulations seem more faithful to biological reality than others. Christoph Adami’s AVIDA, Tom Schneider’s ev, and Tom Ray’s Tierra fall on the “less than faithful” side of this divide. On the “reasonably faithful” side I would place the following three:

Mendel’s Accountant: mendelsaccount.sourceforge.net

MutationWorks: www.mutationworks.com

MESA: www.iscid.org/mesa

Comments

Onlookers (and Mr Kellogg et al): Pardon some direct remarks, preparatory to getting the discussion back on track: I am pretty well convinced at this stage that most of what Mr Kellogg is saying is distractive, rather than substantial. I in particular note that to date he has not cogently addressed the primary issue with Weasel, whether in 1986 form or in whatever newer forms: it is an example of foresighted, targetted search that illustrates intelligent design not any reasonable BLIND watchmaker. As such, it ducks the essential issue posed by Mr Hoyle and many others. Namely, there is no probabilistically credible materialistic pathway to first life and body plan level innovations, once we have identified the functional, tightly integrated, information-rich complexity of the nanotech of life. In that context, Weasel -- sadly -- has since 1986 been a grand exercise in question-begging, and the many personalities-tainted rhetorical attacks here and else where over latching -- especially in light of the refusal to address the cogency of the responses and demonstrations -- have been a grand red herring exercise. A red herring led out to strawman mischaracterisations that were soaked in ad hominems and ignited to cloud and poison the atmosphere. (The latest recirculated one being complaints that I am long and unclear, when in fact the real truth is that it takes time and words to answer objections at a responsible level.] Onlookers, let's set the record straight (yet again). And, pardon the necessity of a bit of a 101 level tutorial; which will have a certain unavoidable length. First, though, we can -- yet again [see what, sadly, I mean about failing to responsibly address cogent, easily accessible evidence . . . ? (Cf. 313 just above) ] -- roll the tape back to December last -- yet again -- where you will see the issues I raised as primary and as associated:
[107:]the problem with the fitness landscape [i.e. as envisioned for the biological world] is that it is flooded by a vast sea of non-function, and the islands of function are far separated one from the other. [Notice how this has never been seriously addressed: getting to body plan no 1, with credibly 600 k bits or so of bio-information as the threshold of functionality, i.e a config space ~ 10^180,617; and onward for body plans requiring 10's - 100's of mega bits of increments of functional information] So far in fact — as I discuss in the linked in enough details to show why I say that — that searches on the order of the quantum state capacity of our observed universe are hopelessly inadequate. Once you get to the shores of an island, you can climb away all you want using RV + NS as a hill climber or whatever model suits your fancy. But you have to get TO the shores first. THAT is the real, and too often utterly unaddressed or brushed aside, challenge. [Notice, what is central to the issue, right from the outset.] [111, excerpted paragraph used by GLF in his threadjack:] Weasel [i.e. as published in 1986] sets a target sentence then once a letter is guessed it preserves it for future iterations of trials until the full target is met. [If you doubt this, simply observe the o/p . . . ] That means it rewards partial but non-functional success, and is foresighted. Targetted search, not a proper RV + NS model.
Now, per 101 . . . 1 --> It should be very plain that the latching concept was there from the outset, as an explanatory concept primarily relative to the o/p behaviour evident form Weasel 1986. 2 --> The terms explicit and implicit latching were descriptive summaries [so is Mr Dembski's ratcheting] of that concept. 3 --> Then, when issues over mechanisms came up, I posited three T1 -- random search, T2 -- explicit latching, and T3 -- implicit latching. (And, by the lucky noise principle, random variations can in logical principle mimic any pattern in the world; the issue is that when the odds against the lucky noise alternative are sufficiently long, it becomes implausible. This is a premise of science and that common sense reason that says we live in an intelligible world, not a chaos in which all happens by accident and/or by blind necessity. Down that chance + necessity only road lies self referential incoherence of the deepest and most irretrievably self-defeating kind. [For more details, kindly read appendix 7, my always linked.]) 4 --> All of this, Mr Kellogg knows, or should know. So -- sadly -- his objection above is objection for the sake of objection and/or -- which would be even worse -- intended to annoy and to create distractions from the issues on the primary merits, and even the secondary issue through tactics of repeated red herrings, strawmen, and ad hominems. 5 --> To deal with that, we should realise that those who repeatedly distract attention from a subject and the framework of evidence and reasoning that point where it is heading do so as a rule because they may be deeply confused and/or fearful and/or deeply hostile to where that subject is heading. 6 --> In this case, evolutionary materialism is a deeply held view that dominates the academy and many other influence or power centres of our civilisation, and gives comfort to radical secularism and its fellow travellers in their cultural-moral agendas, that they are "scientific." But in fact, science is being held captive to teh censorship of evolutionary materialistic philosophy,as Mr Lewontin's confession in the NY review of Books, 1997, so abundantly documents, and as has in recent years been enforced upon science education, courtrooms and even parliaments. Let's remind ourselves:
Our willingness to accept scientific claims that are against common sense is the key to an understanding of the real struggle between science and the supernatural. We take the side of science in spite of the patent absurdity of some of its constructs, in spite of its failure to fulfill many of its extravagant promises of health and life, in spite of the tolerance of the scientific community for unsubstantiated just-so stories, because we have a prior commitment, a commitment to materialism. It is not that the methods and institutions of science somehow compel us to accept a material explanation of the phenomenal world, but, on the contrary, that we are forced by our a priori adherence to material causes to create an apparatus of investigation and a set of concepts that produce material explanations, no matter how counter-intuitive, no matter how mystifying to the uninitiated. Moreover, that materialism is absolute, for we cannot allow a Divine Foot in the door. [NY review of books, 1997]
7 --> So, we must determine to restore science to being an unfettered (but intellectually and ethically responsible) pursuit of the truth about our world in light of the empirical evidence. 8 --> In fact, the truth is that evo mat has no good -- empirically well warranted -- origin of life account, once we see the incredible complexity and information basis for cell based life, multiplied by the fine-tuned complexity of the physics that undelries such a cosmos that facilitartes such life. Similarly, it has no good account for the origin of body plan level biodiversity that elaborates on teh complexities of such life. [Cf my always linked for a reasonable 101 level discussion of this.] 8 --> Thirdly, it has no good grounds for the credibility of our minds and consciences, i.e it is self referentially absurd at its core. [Cf the Appendix 7 the always linked for a 101 level discussion on this. (And just perhaps this sort of thing and the sort of thing that appears in the correctives and the glossary at UD are why we see the sort of rhetorical tactics that have been evident in recent weeks. The future of our civlisation is at stake,and who holds the academic high ground is a big part of it. Worse, some pretty indefensible positions have been taken by the evo mat magisterium, and have been defended by tactics that will not stand the light of day in too many instances.)] 9 --> In that context, we can see why the Hoylean challenge was ducked through a question-begging exercise. The issue is -- pace Wiki's attempted defense of Weasel etc. -- not whether or no RV + NS as BLIND watchmaker can climb hills to the peaks of Mt Improbable, the issue is to first credibly get to the shorelines of the island of function on which the mt sits. For, the islands of function are tiny, hard to find, dots in a vast sea of plainly non-functional configs. [ . . . ]kairosfocus_{April 12, 2009
April
04
Apr
12
12
2009
02:32 AM
2
02
32
AM
PDT}

Hazel: Pardon a few direct words: you are beginning to recycle already long since adequately answered objections. So I simply repeat: That approach which simply takes Mr Dawkins' TBW at face value and does a straightforward letterwise, explicitly latched partitioned search is conceptually and programmatically simpler to do than one that has to balance per generation size, mutation rates and a filter towards proximity to the target. The very length of the exchange on the latter is a strong enough proof of that. So is th fact that consistently, the first resort is to do a partitioned search program, and the implicitly latched, quasi-latched etc cases come later, and often with apparently much more program development effort. GEM of TKI PS: Clive, Mr Kellogg et al have a problem with seeing that we are dealing with a tailed distribution [the binomial] for the letters, with a significant fraction being no-change under the relevant odds of random change per letter. So, with a population N sufficiently large to bring out the bulk pattern of the distribution, but small enough that the double and triple mutation cases required for the proximity filter to pick up substitutions of a newly correct letter for a letter that reverts to incorrect status as champion, of probabilities p, will be PRACTICALLY impossible [effectively unobservable, N* p -/-> ~ 1], the Weasel program will implicitly latch as no change and single step changer champions dominate the runs to target; and just beyond that range quasi-latch will occur, as only very rarely will we see the relevant reversions and substitutions. Not even the provision of actual runs of such an implicitly latched case [cf above thanks to Atom] has proved sufficient to move them from their preconceptions. At the root, the problem is that the Law of large numbers on observable patterns of samples from a population, is fatal to their underlying evolutionary materialism.kairosfocus_{April 12, 2009
April
04
Apr
12
12
2009
02:31 AM
2
02
31
AM
PDT}

David Kellogg, "That phrase is likely to conserve correct letters but does not have to in any particular generation." I'm sorry David, this makes no sense to me. The phrase conserves letters?Clive Hayden_{April 11, 2009
April
04
Apr
11
11
2009
10:12 PM
10
10
12
PM
PDT}

hazel, Thanks for that clarification. So there is a target phrase that is being approximated to. I see. The program is making an effort at getting these letters in their correct places in order to make the phrase, right?Clive Hayden_{April 11, 2009
April
04
Apr
11
11
2009
10:07 PM
10
10
07
PM
PDT}

kairosfocus insists that he distinguished between explicit and implicit latching from "the outset." Certainly those terms were used from the outset of this thread, but the discussion precedes this thread by some months (and, in typical kairosfocus fashion, by tens of thousands of words -- over 10,000 from kairosfocus in this thread alone!). Those who examine that earlier history, however, will find that the term latching arose from the primeval soup of words sometime after kairosfocus was asked to back up a claim that Weasel searched for and fixed correct letters. The species L. explicitus and L. implicitus evolved sometime after that, with L. implicitus giving rise to the closely related subspecies (some would say it's the same species) L. quasitus.David Kellogg_{April 11, 2009
April
04
Apr
11
11
2009
07:17 AM
7
07
17
AM
PDT}

Atom wrote that one (or maybe Apollos), but they had confused other factors into their response. Let's not dig up the past. How is if the letter is incorrect, p(mut) = p if the letter is correct, p(mut) = 0 simpler than p(mut) = p?hazel_{April 11, 2009
April
04
Apr
11
11
2009
06:31 AM
6
06
31
AM
PDT}

hazel: The programmers among us have testified to that fact, that an explicitly latched version of Weasel is simpler than an implicit one. And, remember that we do need to do a bit of parameter tuning through runs to get the implicit latching effect too. GEM of TKIkairosfocus_{April 11, 2009
April
04
Apr
11
11
2009
06:12 AM
6
06
12
AM
PDT}

Onlookers (and DK): Let's clarify. 1 --> Is there o/p latching in Weasel 1986? [Me -- per LOLN, yes! DK et al originally, no -- Hazel being a notable exception. Evidence of Atom's sims: very probably so.] 2 --> Are there various ways to explain such o/p latching, e.g. by explicit and implicit latching? (Yes -- but DK et al thought that mechanism T3, IMPLICIT latching was non latching. [Period. Onlookers cf 39 and 41 supra.]) 3 --> Do explicit letterwise and implicit latching differ? (Obviously: the former partitions letterwise and explicitly masks off successful guesses, letter by letter. The latter -- by definition -- arrives at o/p latching [what was to be explained . . . ] by co-tuning of pop per generation, mutation rate and proximity only filter, in light of the binomial statistical distribution under LOLN. [Far skirt members regularly or reasonably are expected to show up only when there is enough in a sample to make that reasonable. Criterion N.p --> 1 or so.]) 4 --> So why is it now a big thing to pretend and argue as though the undersigned needs to concede, that this difference [which I pointed out from the outset in defining the two mechanisms] exists? 5 --> ANS: Simple, onlookers: We have now demonstrated implicit latching by actual simulation so there can be no debate on its reality and credibility. This moves beyond arguments to what is more persuasive and hard to deny: directly observable fact. Facts that in fact underscore that a much objected to argument was correct all along. 6 --> So, do we see a serious response to the now confirmed balance on the merits? (Sadly, no. History is being re-written counterfactually before our eyes.) 7 --> So let's roll the tape briefly:
DK, 39: I think “implicit latching” is a way to avoid saying “non-latching.” SK, 41: So-called “implicit latching” (which as David points out really means “non-latching”) . . . DK, 315: kairosfocus, again I say: what you call “implicit latching” means non-latching at the mutation level. Can you bring yourself to say that? Say it yourself: There is no latching at the mutation level.
8 --> In short, what was never in dispute about how implicit latching works [i.e why it is implicit as opposed to explicit!], is now trotted out as a qualification ex post facto, and put rhetorically up as though I would have to now concede it. Sad. And sadly revealing. GEM of TKIkairosfocus_{April 11, 2009
April
04
Apr
11
11
2009
06:09 AM
6
06
09
AM
PDT}

KF writes, "Additionally, it [explicit latching] is the easiest to develop an algorithm for and to code." NO! NO! NO! In 136, I wrote, and you eventually agreed, that the only difference between the two cases is
In the implicit case, for each letter p(mut) = p In the explicit case, for each letter if the letter is incorrect, p(mut) = p if the letter is correct, p(mut) = 0
How in the world can you say that the explicit case is simpler when it includes a longer conditional statement where the implicit case has a simple one-line rule? I dare you to respond to just this one question, preferably in under 300 words. Take this on as a challenge, kairosfocus - an exercise in restraint. Don’t mention any other topic, all of which you have posted on at least a dozen times, and just address this issue. Good luck.hazel_{April 11, 2009
April
04
Apr
11
11
2009
05:59 AM
5
05
59
AM
PDT}

kairosfocus, again I say: what you call "implicit latching" means non-latching at the mutation level. Can you bring yourself to say that? Say it yourself: There is no latching at the mutation level. Feels good, doesn't it? I maintain that "implicit latching" is an obscuring term that can do nothing to explain how the program (a) does not latch at the mutation level, and (b) works anyway. Throughout this discussion, critics of Weasel have repeatedly failed to understand it because they have failed to understand in what sense it (however crudely) models evolution. To do what it says, Weasel must model: 1. Random variation at the level of the gene 2. Nonrandom selection at the level of the individual 3. Competition among individuals in a population. The only form of Weasel that satisfies these is one where the letters don't latch. Why make up the term "implicit latching"? The only thing kit does is confuse points (1) and (2). Repeat: in Weasel, mutation is and remains random at the level of the letter. That's the only thing that makes sense.David Kellogg_{April 11, 2009
April
04
Apr
11
11
2009
05:30 AM
5
05
30
AM
PDT}

PS: I need to amplify a point. I should add that the best explanation per the evidence in hand for the difference between the published Weasel program o/p 1986 and tfhe videotaped run of 1987 remains that Mr Dawkins modified the program. Specifically, on the model of implicit latching circa 1986, that the program in 1987 -- which behaves radically differently form that of the 1986 o/p [i.e. reversions are frequent, as opposed to credibly absent, and it seems to run for a much larger number of generations] -- is that the parameters for population size, mutation rate were shifted in ways that detune the program. [Cf the various runs I did above.] And that is not an "accusation" -- much less an unwarranted accusation -- on my part, it is a reasonable explanation on evidence of radically different behaviour.kairosfocus_{April 11, 2009
April
04
Apr
11
11
2009
04:11 AM
4
04
11
AM
PDT}

h --> I have, finally, ALWAYS said that the latching question is a secondary defect of weasel, but has significance as it is so obvious and points to the PRIMARY defect: Weasel is a case of targetted search that fails to properly address the need to get to the shores of islands of complex information rich function, in a sea of non-function. i --> Indeed, from the original thread where this all came up last December, I stated [and as was cited repeatedly in the current multi thread exchange, including at 64, 177 and 268 supra, so DK has little excuse for this]:
[107:]the problem with the fitness landscape [i.e. as envisioned for the biological world] is that it is flooded by a vast sea of non-function, and the islands of function are far separated one from the other. [Notice how this has never been seriously addressed: getting to body plan no 1, with credibly 600 k bits or so of bio-information as the threshold of functionality, i.e a config space ~ 10^180,617; and onward for body plans requiring 10's - 100's of mega bits of increments of functional information] So far in fact — as I discuss in the linked in enough details to show why I say that — that searches on the order of the quantum state capacity of our observed universe are hopelessly inadequate. Once you get to the shores of an island, you can climb away all you want using RV + NS as a hill climber or whatever model suits your fancy. But you have to get TO the shores first. THAT is the real, and too often utterly unaddressed or brushed aside, challenge. [Notice, what is central to the issue, right from the outset.] [111, excerpted paragraph used by GLF in his threadjack:] Weasel [i.e. as published in 1986] sets a target sentence then once a letter is guessed it preserves it for future iterations of trials until the full target is met. [If you doubt this, simply observe the o/p . . . ] That means it rewards partial but non-functional success, and is foresighted. Targetted search, not a proper RV + NS model.
3] Hazel, 304: There is no mention [in TBW] of an explicit latching rule. There are statements that pretty well support that understanding, as may be seen from looking at 285 and Joseph's remarks. Per Weasel 1986, absent the testimony as reported to us in recent days, explicit latching is a very reasonable understanding. (Indeed, the Monash University biologists, naturally understood it that way until Mr Elsberry "corrected" them.) And of course the point is, that while latching is very much a secondary defect of the Weasel 1986 program, it points like a signpost to the primary defect: targetted search without reference to a reasonable threshold of funcitonality. Intelligent design, with foresight built in, being presented in a rhetorical context that seeks to make a claimed BLIND -- non foresighted -- watchmaker seem credible. In short, the entire exercise in 1986 was fundamentally misleading. [Cf 285 for why I say that.] I think that a better expenditure of effort on the part of those who have laboured long and hard to try to justify it, would be to acknowledge that Weasel is fundamentally misleading, was known to be so from the outset [cf 285 above] and should never have been used. It should be withdrawn -- and, frankly, apologised for. 3] Atom, 311:In nature, the “current” fitness function is based on the organism itself, the random environmental factors, the other organisms in its biosphere, etc. It is a function of many inputs. So for a specific organism, which is just a “permutation” of possible traits, we can assign a “fitness” value to it: more or less how many organisms it successfully brings into the next breeding generation. Perhaps the “fitness function” at the time can assign two different values for the same “permutation”, which is where random events come into play, but that can just be modelled as the moving between two similar fitness reward matrices. Physically intantiated fitness functions change over time, sometimes in complex ways, and can be density dependent, etc. Yes. I just note that the further complication is that this starts from a functioning organism and body plan. The first question is to get TO that functioning body plan, from the very first one. And in the light of the credible degree of complexity that obtains, e.g. as measured by the bit capacity of the DNA, about 600 k bits. And, onward, to get to new body plans, 10's - 100's of mega bits is credible. This, in my considered opinion [cf my always linked], is the central problem/weakness with the whole Darwinist scheme for macroevolution and the associated spontaneous origin of life models. GEM of TKIkairosfocus_{April 11, 2009
April
04
Apr
11
11
2009
03:56 AM
3
03
56
AM
PDT}

Onlookers: It seem that some further remarks are neeeded, in light of onward fairly sharpish comments by esp. DK, and some telling associated silences on key points that were up to recent days central objections [e.g. on law of large numbers]. Of course, let us first underscore: the MAIN problem with Weasel is that it is targetted search, which does not properly reckon with the need to first achieve complex functionality before one can hill-climb to optimal function, so it ducks the Hoylean challenge. That is, natural selection etc, are not capable of creating function that does not exist, they can only describe what happens when populations with differential fitness to environment interact, i.e. the currently fittest survive and reproduce themselves. Weasel, as Mr Dawkins knew from the outset, sadly, is adn was always fundamentally misleading. [Cf my remarks on his statements at 88 and as reproduced at 285; which document precisely what I just claimed.] A few footnotes on further comments: 1] DK, 305: What people here call “implicit latching” turned out to be trivial once it was clear what you were talking about (that is, no latching at the mutation level) Of course, a main objection maintained for weeks was that there was no latching in Weasel 1986's o/p much less anywhere else. And, my adverting to the significance of the Law of Large Numbers to infer to o/p latching in the data as published by Mr Dawkins in 1986, was supposedly illustrative of my ignorance. VANISHED, unacknowledged, as though it never were, now that thanks to Atom, we have been able to replicate the circumstances and show that this is a very reasonable conclusion on the published o/p circa 1986. (But in fact, had the import of LOLN been acknowledged from the outset, this would not have ever been an issue.) Next, I have for weeks been repeatedly very, very explicit and specific that the issue is to explain the Weasel 86 o/p latching, for which mechanisms T2 -- EXPLICIT, and T3 -- IMPLICIT latching were reasonable candidates. On the data in hand circa 1986, explicit latching is a very reasonable interpretation, and it is the simplest way to account for what was said in the text as can be seen at 285. Additionally, it is the easiest to develop an algorithm for and to code. Also, if we conceive of the target on a letterwise basis, achieving it by random search per letter with proximity filtering is not materially different form doing so on a phrasewise basis. And, indeed the letterwise interpretation fits in very well with "cumulative selection" that "chooses the one which, however slightly, most resembles the target phrase." However, on being told that c. 2000 Mr Dawkins said that he did not explicitly latch on a per letter basis, we have accepted that T3, implicit latching is the best explanation, and we have been able to materially replicate the situation again courtesy Atom. 9the existence and capacity of such implicit latching was also objected to, it was hardly a trivial matter. Indeed, DK is on record, among others, as saying, in effect, that implicit latching = non-latching, specifically, that it was allegedly a way to AVOID saying non-latching. [Cf 39 and e.g. Skeech's response at 41]) It is thus fair comment to say that the above has not been a "trivial" exercise. Implicit latching is real -- currently observed, and [as described and predicted by the undersigned and Joseph] it arises from the interaction of per letter mutation rate, population size and filter that rewards mere proximity. And, of course, it is precisely because it arises from interaction across three factors that it is different from use of a mask that once letters achieve their target, they will be preserved directly and explicitly. In short, the excerpt being commented on is an attempt to word a concession as if it were a victory. Sad. 2] explicit latching radically misconstrues what the program was trying to show . . . . The accusation for years was that Weasel explicitly fixed letters. kairosfocus insisted this was the best explanation, accused Dawkins of modifying the program between 1986 and 1987, and only backed down (kind of) when told Dawkins never used explicit latching. Recently he’s said the issue is unimportant, but it sure was important for him earlier. Highly misleading, ad hominem-laced mischaracterisation of what has happened over the past weeks: a --> It is a very reasonable, natural and simple understanding of Weasel 86' o/p, that it latches letters; especially given where 200+ of 300+ potentially changeable sampled letters go right then stay right. b --> That this is so was stoutly resisted for weeks and, now that it has been abundantly vindicated thanks to Atom's simulation, there is a studied silence on the matter, especially of the significance of the law of large numbers. (Just scroll up and look at the way my darts and charts thot expt was objected to.) c --> On looking at the evidently latched o/p, the question that arises is mechanism, and when that came up as an issue, I proposed two: T2 (explicit) and T3 (implicit) latching. [T1 of course, was the "select all at once" option, which in principle [per lucky noise] can replicate any other pattern of champions in a run; in praxis, it most likely will never achieve the target in any reasonable length of time -- its relevance lying in that the 28 latter phrase is well below the reasonable threshold of first life function, 600 k bits. In short what Mr Dawkins dismissed as "single step" mutation, is the most biologically credible case of getting to FIRST function. Hill climbing can only begin when you already have function.] This of course sets up an inference to best, empirically anchored explanation exercise. d --> As I will give more details on in responding to Hazel, the simplest explanation on the published run excerpts and statements of Mr Dawkins circa 1986 [cf 285 supra and focus on the implications of rewarding the slightest increment in proximity multiplied by remarks on cumulative selection and on the pattern of o/p where we see not one reversion form a letter hitting target], is that the program explicitly latched. e --> However, implicit latching was still a feasible mechanism [as an alternative explanation], and when a statement was presented to us that Mr Dawkins c 2000 denies explicitly latching the program, it has become the best explanation on preponderance of evidence. f --> This is, of course, not a definitive proof. (Only credible code would be decisive; e.g. Apollos has been able to show that an explicitly latched version of Weasel can show reversions, once they are written in.) g --> Nor is such a response to evidence per the provisionality of empirical reasoning properly characterised as making [presumably, ill-founded] accusations. [ . . . ]kairosfocus_{April 11, 2009
April
04
Apr
11
11
2009
03:55 AM
3
03
55
AM
PDT}

iskim labmildew, You seem to be getting ahead of the argument, at least in terms of what is presented in the Weasel 2.0 documentation. First, there is a formal/mathematical point and later a physical/biological interpretation. My goal with Weasel is to explore the formal structure of fitness space in hopes of perhaps later shedding light on the physical implications for biology. But lots of work and experiments needs to be done before we get to that step. I should begin by explaining my nomenclature more fully. As you know, a function simply takes a set of input and maps it to a set of outputs. In our case, we have a set of string permutations as inputs, and we can assign each permutation a "fitness" value between 0 and F. Let's start simple, with a two bit string and a two value fitness function. This creates a mapping, which I call a Reward Matrix (since we're rewarding strings based on this value and since it is a matrix, similar to a truth-table row in Symbolic Logic. You can also refer to the rows as "reward vectors", which may be more descriptive, but for now I'll refer to each row as a reward matrix, since you can write them in two-dimensional matrix form as well.) The complete table for two bit strings is this: 11 10 01 00 Function --------------------------------- 0 0 0 0 (2)f0 0 0 0 1 (2)f1 0 0 1 0 (2)f2 0 0 1 1 (2)f3 0 1 0 0 (2)f4 0 1 0 1 (2)f5 0 1 1 0 (2)f6 0 1 1 1 (2)f7 1 0 0 0 (2)f8 1 0 0 1 (2)f9 1 0 1 0 (2)f10 1 0 1 1 (2)f11 1 1 0 0 (2)f12 1 1 0 1 (2)f13 1 1 1 0 (2)f14 1 1 1 1 (2)f15 As you can see from the table, for a two bit string with two possible fitness values, you have exactly 2^(2^2) = 16 possible reward matrices. (Each row represents the mapping of the inputs to a set of outputs, using one fitness function.) I labeled the fitness functions as follows: (fitness_values_base) f Vector_Value The first part in parentheses is the number of possible fitness values, 0 to F. In our case it is (2). Then "f" signifies it is a fitness function, and the number that follows is the reward matrix number in that base. For example, (2)f5 would mean: take the base-2 representation of 5, which is 0101, and this is the output row of assigned fitness values. This notation works for all your permutation numbers and bases; if your input string gets more characters, you simply write them in alphabetical order, with the "lowest" string permuation on the far right column. (In our case 00 is the lowest.) If we want to specify the base of our input string in the notation, you could like so: (4,2)f0. If not, then we assume that all values to the left of the most significant bit will be zero. For example, when looking at the (2)f0 row for a two bit string (0000), this is identical to the (2)f0 row of a four bit binary string, except there are zeros for the leftmost, higher order bits (00000000). So noting the base of the input string is unnecessary, though you can note them. My notation scheme is arbitrary, but it allows us to talk about specific fitness functions and the "fitness space" in a formal way. We can get any function, for example (4)f15, and construct the unique reward matrix row from this: 0033. Now that we see all the possible fitness value assignments, we can ask "How many of these improve our search? How many of them hinder it?" This is what Weasel 2.0 allows us to begin to explore (and some other sims in the works will help explore in an easier manner.) So far the problem is strictly formal/mathematical and calls for numerical and/or empirical analysis. Now, you bring up the question of physically instantiated fitness functions. This is a different question and is beyond what my work at this point is ready to fully address. But in general, given any number discrete inputs, there are a fixed number of possible fitness value assignements, if your fitness values are discrete and bound by a limit. (For example, 0 through F.) In nature, the "current" fitness function is based on the organism itself, the random environmental factors, the other organisms in its biosphere, etc. It is a function of many inputs. So for a specific organism, which is just a "permutation" of possible traits, we can assign a "fitness" value to it: more or less how many organisms it successfully brings into the next breeding generation. Perhaps the "fitness function" at the time can assign two different values for the same "permutation", which is where random events come into play, but that can just be modelled as the moving between two similar fitness reward matrices. Physically intantiated fitness functions change over time, sometimes in complex ways, and can be density dependent, etc. So at each moment, we can see that one of the possible fitness functions will be instantiated, and will assign the corresponding fitness values to organisms ("permutations") present. (This is not entirely correct, since the organism itself does the reproducing, thus the "assignment" of fitness value, but the environment and other organisms play the part in controlling how successfully the organism can survive and reproduce.) Sticking with our metaphor, you examine a particular insect and the shape of its reproductive organ and see that it must have a certain shape to be able to successfully copulate with members of the opposite sex; so at that time, given the input of that organisms traits (permuation), and the traits of females as additional input, that should be assigned a high fitness value. However, if the shape of the female becomes different, then that same trait permutation will get a low fitness value. So fitness functions change over time and the same trait can be assigned a high value or a low value depending on the fitness function in place at that time (which is dependent on the other inputs, the laws of physics, etc.) So far I have not disagreed with you, I have just clarified. So are some reward matrices uninstantiateable? Perhaps. This is an empirical question that should be explored. If, however, we are limited in the number of physically instantiateable fitness functions, that that may directly limit what traits we can effectively select for, if not every fitness function allows us to have a successful search for a target. The more primary question is this: How many of the possible fitness functions allow for success in a specific search? If we find over and over again that only a few fitness functions will help us find our target (namely those with high information about the specific target), then this would have obvious implications for biology. As of right now, the more basic research hasn't been done (as far as I know.) I simply don't know the answer to a lot of those questions, but I'm beginning to investigate them. Hopefully Weasel will allow others to begin exploring and experimenting with me. Atom PS Sorry for the long post PSS I refer to the output matrix rather than simply talking about fitness functions, since for any input there are an infinite number of functions that can map it to that same output. (Just keep adding useless parts to your function/circuit. This is why we do circuit reduction, since a simpler circuit may be able to acheive the same logic, meaning map the same input to the same output.) Talking about the mappings themselves, therefore, is more precise.Atom_{April 10, 2009
April
04
Apr
10
10
2009
10:04 AM
10
10
04
AM
PDT}

Clarification: "It" in the "It preserves" [309] refers to Weasel, not to Clive's question.David Kellogg_{April 10, 2009
April
04
Apr
10
10
2009
10:02 AM
10
10
02
AM
PDT}

Clive [307], your question makes no sense. It preserves the closest phrase among the progeny. That phrase is likely to conserve correct letters but does not have to in any particular generation. That's how Weasel works.David Kellogg_{April 10, 2009
April
04
Apr
10
10
2009
09:46 AM
9
09
46
AM
PDT}

Clive, I don't know whether you have been paying close attention to the discussion or not, but everyone agrees that it knows what is a correct letter by comparing each phrase to the target phrase. There has never been any controversy about this aspect of the situation. The issue has been about the details of how and when in the program this happens.hazel_{April 10, 2009
April
04
Apr
10
10
2009
09:20 AM
9
09
20
AM
PDT}

David Kellogg, "Of course it will tend to conserve correct letters. Again, that was part of the point." How does it know what is a correct letter? How does it know to "conserve" any letter? It seems to me that if you didn't already have the whole phrase in mind, you wouldn't get any correct letters, because you would have nothing to approximate towards "conserving."Clive Hayden_{April 10, 2009
April
04
Apr
10
10
2009
09:06 AM
9
09
06
AM
PDT}

Correction: "I submit that the longstanding ~~focus~~ insistence on what we’re now calling explicit latching demonstrates that people didn’t understand what Dawkins was saying." Obviously those who wanted to correct it had to focus on it to do so.David Kellogg_{April 10, 2009
April
04
Apr
10
10
2009
07:11 AM
7
07
11
AM
PDT}

What people here call "implicit latching" turned out to be trivial once it was clear what you were talking about (that is, no latching at the mutation level). That's what makes the program work: it moves the phrase (not the letters) toward the target at the selection level. Why was refuting explicit (mutation-level) latching important? Because explicit latching radically misconstrues what the program was trying to show. Explicit latching both (a) runs counter to the plain sense of the BW text, and (b) misrepresents the function of this sort of (admittedly minor) exercise. The accusation for years was that Weasel explicitly fixed letters. kairosfocus insisted this was the best explanation, accused Dawkins of modifying the program between 1986 and 1987, and only backed down (kind of) when told Dawkins never used explicit latching. Recently he's said the issue is unimportant, but it sure was important for him earlier. I think it became unimportant when he had to give up the claim of explicit latching. I submit that the longstanding focus on what we're now calling explicit latching demonstrates that people didn't understand what Dawkins was saying. Now, I don't always agree with Dawkins: in the fights between Dawkins and Gould, I was usually more on Gould's side. But obscurity is not one of his faults. Whatever else he is, Dawkins is a clear writer. On his worst day he writes more clearly than kairosfocus at his best. I submit that ID proponents' long-term insistence on explicit latching and the current obfuscation over "implicit latching" suggest a greater failure of understanding. In short, Weasel is hard for you to understand as an analogy for evolution because you don't understand evolution. I wish it were otherwise. I wish the problems were merely ones of interpretation. But I think your repeated attention to the "gnat" of Weasel obscures the big conceptual log you are ignoring.David Kellogg_{April 10, 2009
April
04
Apr
10
10
2009
07:02 AM
7
07
02
AM
PDT}

Joseph - 1. Dembski and Marks assumed a latching rule was in place 2. Here is what people have provided to show that a latching rule is not what Dawkins used a) Direct quotes from BWM where Dawkins describes the process. There is no mention of an explicit latching rule. b) Programs, including Atom's, that show that explicit latching is not necessary to produce the kind of results shown in BWM c) Quotes from elsewhere in the book (See 263) that show that Dawkins knows that mutation is random in respect to fitness. With this said, how can you possibly say,
To date not one person has provided anytrhing that would show that the inference made by Dembski and Marks is wrong.
Clearly people have provided a great deal of evidence that Dembski was wrong. You can say that you don't accept the arguments provided, but you can't possibly say that no one has provided any.hazel_{April 10, 2009
April
04
Apr
10
10
2009
06:58 AM
6
06
58
AM
PDT}

Joseph Appreciated. Just, we need to maintain an example of civlity, if we are to have a reasonably merit based forum. [Believe you me there are any number who would want to trot up crying hypocrite at us if they could.] GEM of TKI PS: Hazel: per the substance of what we have from 1986, Joseph is right. It is because of additional information, i.e. a report of testimony, that I take implicit latching on preponderance of evidence; which does not explicitly lock up letters, but the dynamics I just explained -- again -- to DK -- show why implicit latching will latch. So, Joseph is right again on that mechanism: once a letter goes right under the right circumstances it will not normally be knocked off its perch -- you have to push out into the skirts far enough for that to happen with Quasi-latching. (Notice, the preponderance conclusion is not at all "beyond reasonable doubt" or "to moral certainty" or the like strongest degree of empirical warrant.)kairosfocus_{April 10, 2009
April
04
Apr
10
10
2009
06:35 AM
6
06
35
AM
PDT}

KF, My aplogies I do not have your patience. As a hockey player I just want to jam. But that is me...Joseph_{April 10, 2009
April
04
Apr
10
10
2009
06:15 AM
6
06
15
AM
PDT}

To date not one person has provided anytrhing that would show that the inference made by Dembski and Marks is wrong.”
I have. Repeatedly. And others have.
Can you reproduce those quotes? I haven't read any.
You keep asking whether people have read the book, and challenging people to provide quotes that support their position.
That is how it goes.
I challenge you to provide a direct quote, from the book, that makes you think that the program explicitly locks letters so that once a “best fit” phrase is selected in a generation the correct letters in that phrase will always be passed to its children (Dawkins calls them progeny), without exception.
I take it you have reading comprehension issues. That was never my point. My point is and always has been that with cumulative selection once someting (useful) is found the search for it is (essentially) over. And that is what a patitioned search is.Joseph_{April 10, 2009
April
04
Apr
10
10
2009
06:13 AM
6
06
13
AM
PDT}

Joseph writes, “To date not one person has provided anytrhing that would show that the inference made by Dembski and Marks is wrong.” I have. Repeatedly. And others have. You keep asking whether people have read the book, and challenging people to provide quotes that support their position. I challenge you to provide a direct quote, from the book, that makes you think that the program explicitly locks letters so that once a “best fit” phrase is selected in a generation the correct letters in that phrase will always be passed to its children (Dawkins calls them progeny), without exception. Please quote the exact text, with page numbers. Thanks.hazel_{April 10, 2009
April
04
Apr
10
10
2009
06:04 AM
6
06
04
AM
PDT}

PS: Joseph, thanks on the substantial matter, but please . . . on words.kairosfocus_{April 10, 2009
April
04
Apr
10
10
2009
05:37 AM
5
05
37
AM
PDT}

Joseph @283
I challenged anyone to provide the reference quote from TBW which would show that with a cumulative selection process once a matching letter is found the program keeps looking for it. To date not one person has provided anytrhing that would show that the inference made by Dembski and Marks is wrong.
That statement does not reflect the facts of the matter. David Kellogg has posted a link to a line by line refutation of your claim. Further, several people, including myself, have noted the key lines from The Blind Watchmaker:
It now 'breeds from' this random phrase. 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.
This is the essence of Dawkins' description. There is no reasonable interpretation of these words that suggests that letters are "latched." In fact, the term "random error" suggests exactly the opposite. Joseph @284
Your selective quoting of TBW didn’t capture what he stated after-
I quoted the entire description of the Weasel program. No quote mining, nothing "selective."
That being that cumulative selection is a process of slight improvements- however slight.
Please provide the quote to which you are referring and explain how it is pertinent to this discussion. Dawkins' words about the Weasel algorithm are very clear, and they show that your interpretation is untenable.
IOW Jay you are being very deceptive which is to be expected.
Someone is certainly trying to be deceptive here, on that we agree. Joseph @286

How can you blame Dawkins for NOT mentioning things that his program was NOT doing, or for not having the foresight to anticipate that Dembski would mistakenly make incorrect assumptions about the program.
Did you read the book? He makes it clear t5hat once something is found the search is over.
Provide the quote supporting your claim or retract it. After all, you wouldn't want to be seen as lacking in intellectual integrity or being deceptive, would you? JJJayM_{April 10, 2009
April
04
Apr
10
10
2009
05:36 AM
5
05
36
AM
PDT}

DK: Ignoring for the moment a long list of issues you have to address in light of the runs in evidence above, I think you need to look a bit closer:
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.
1 --> We are dealing with discrete state things, where the element is the 27-state character, from the set {A, B, C, . . . Z, * [for space]} 2 --> In that context, the slightest increment possible is that there is a one-letter improvement in proximity to the target. 3 --> next, we are dealing with runs where we take 40+ and 60+ as published, which means that about 1/2 the time, no-change wins. [That no change may do such we can see from the runs of close proxies to Weasel c 1986 on implicit latching above.] 4 --> We may freely infer that no-change cases are frequent in the underlying per generation populations of "mutant NONSENSE phrases" from which the champions come; at least for the relevant published runs and close enough cases. 5 --> thus only improvements of one or more letters, or substitutions [cf. cases in point above] may win by competition with the no-change cases. (As well they may occur together: substitution plus improvement.) 6 --> however, we are dealing with a Bell or reverse J distribution. So, multiple correct mutants will be relatively rare, far skirt cases, and though substitutions are possible, they too will be rare, comparatively. 7 --> So for a certain range, we will see latching, and as pop rises and/or mutation rate, we can then see quasi-latched cases then much less cumulative cases. [In one run it was amusing to see how it was like several times a letter would go right then another in parallel would lose its correct state,and then the program would flounder until the next substitution on and on till finally we got the last le3tter in place.] 8 --> In short, letterwise latching is there in the text of TBW, just look carefully in light of the fact that we are dealing with a discrete state, i.e. digital context. GEM of TKIkairosfocus_{April 10, 2009
April
04
Apr
10
10
2009
05:33 AM
5
05
33
AM
PDT}

David Kellogg:
Of course it will tend to conserve correct letters. Again, that was part of the point.
And that is what I have been saying. Therefor the inference is once a matching letter is found the search for it is (essentially) over. In a partitioned search once a matching letter is found the search for it is (essentially) over. IOW I fail to see the big issue that have the anti-IDists pissing themselves.Joseph_{April 10, 2009
April
04
Apr
10
10
2009
05:32 AM
5
05
32
AM
PDT}

Of course it will tend to conserve correct letters. Again, that was part of the point.David Kellogg_{April 10, 2009
April
04
Apr
10
10
2009
05:26 AM
5
05
26
AM
PDT}

Prev 1 2 3 4 5 … 13 Next

You must be logged in to post a comment.

Leave a Reply