Two forthcoming peer-reviewed pro-ID articles in the math/eng literature

_{William Dembski

January 20, 2009

Informatics, Intelligent Design}

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

The publications page at EvoInfo.org has just been updated. Two forthcoming peer-reviewed articles that Robert Marks and I did are now up online (both should be published later this year).*

——————————————————-

“Conservation of Information in Search: Measuring the Cost of Success”
William A. Dembski and Robert J. Marks II

Abstract: Conservation of information theorems indicate that any search algorithm performs on average as well as random search without replacement unless it takes advantage of problem-specific information about the search target or the search-space structure. Combinatorics shows that even a moderately sized search requires problem-specific information to be successful. Three measures to characterize the information required for successful search are (1) endogenous information, which measures the difficulty of finding a target using random search; (2) exogenous information, which measures the difficulty that remains in finding a target once a search takes advantage of problem-specific information; and (3) active information, which, as the difference between endogenous and exogenous information, measures the contribution of problem-specific information for successfully finding a target. This paper develops a methodology based on these information measures to gauge the effectiveness with which problem-specific information facilitates successful search. It then applies this methodology to various search tools widely used in evolutionary search.

[ pdf draft ]

——————————————————-

“The Search for a Search: Measuring the Information Cost of Higher Level Search”
William A. Dembski and Robert J. Marks II

Abstract: Many searches are needle-in-the-haystack problems, looking for small targets in large spaces. In such cases, blind search can stand no hope of success. Success, instead, requires an assisted search. But whence the assistance required for a search to be successful? To pose the question this way suggests that successful searches do not emerge spontaneously but need themselves to be discovered via a search. The question then naturally arises whether such a higher-level “search for a search” is any easier than the original search. We prove two results: (1) The Horizontal No Free Lunch Theorem, which shows that average relative performance of searches never exceeds unassisted or blind searches. (2) The Vertical No Free Lunch Theorem, which shows that the difficulty of searching for a successful search increases exponentially compared to the difficulty of the original search.

[ pdf draft ]

—————

*For obvious reasons I’m not sharing the names of the publications until the articles are actually in print.

Comments

Sal Gal and Mark: As usual, very interesting discussion with you. I offer my counter arguments to what you say, and in defense of my "thought simulation". Sal Gal: "But you are have not spoken to death of virtual organisms." Death can well be included in the original programmimg of the replicators. You can program the replicators as you like, provided that you don't use any explicit or implicit form of frontloading, which could be anyway easily revealed by an impartial scrutiny of the original software. "There are an estimated 5 X 10^30 bacteria on earth at present. I won’t guess how many elementary self-replicators there might have been prior to cellular life, but will suggest that they may have reproduced more rapidly than bacteria. In any case, our computational resources are relatively tiny." Well, I am not so sure they are so tiny, although I have not made the calculations. Even if we work with less replicators, in a digital environment we can greatly increase the times of variation and of replicating fitness evaluation. And anyway, I am not saying that such a simulation is easy, quick, or that it could be done with small computational resources. I am only saying two things: a) It is possible in principle to realize that simulation. b) It is the only appropriate kind of digital simulation for darwinian theory. "As long as the virtual environments are uninteresting, there is no reason to expect virtual organisms to be anything but the same. One thing that makes an environment “interesting” is that there are many ways to garner resources and reproduce." You can make the operating system as interesting and varied as you like, provided that it contains no specific code to recognize and ecourage any specific thing which you want to attain. In other words, the selection must be self-selection, deriving from the interaction between the functional replicators and the rules of the environment, and must in no way be programmed by the desginer of the simulation. "This is not sensational, but it is qualitatively what you said you wanted to see." I have no problem in admitting that GAs can do a lot of things, like any other software can. The only thing they cannot do is simulate the logical mechanism of darwinian evolution. What I wanted to see is some form of evolution of CSI in simulation of the kind I have suggested. It is obvious that GAs, being intelligently designed problems, can find solutions which the designer did not know in advance. If a computing software is designed to solve an equation, that's exactly because the designer does not know in advance the solution. But that does not make it a simulation of darwinian theory. And finally, I do believe that you should play a little with real proteins. It's rather easy, and it could be an amazing experience. Mark, next post is for you.gpuccio_{January 24, 2009
January
01
Jan
24
24
2009
03:24 AM
3
03
24
AM
PDT}

But you are have not spoken to death of virtual organisms. You also have not addressed the massive parallelism of life. Gpuccio Sal Gal is spot on. Any simulation of RM+NS has to deal with both RM and NS. RM is random in the sense that the mutation is independent of the selection criteria. But NS, which is the simulated by the fitness function, is far from random. What confuses the issue is that the fitness function for real species are very complicated, hard to determine, and keeps on changing over time. Nevertheless it is not random. If the criteria for survival in each generation had no relation to the criteria for survival in the previous generation then evolution would not get off the ground. GAs typically specify a simple and unchanging fitness function. This is like a world where there is only one attribute that matters in the struggle to survive. For example, imagine a world where the only organisms to survive were those with the greatest power/weight ratio. Then the mechanism of RM+NS would lead to species with ever greater P/W ratios. This world has just as much knowledge of its target as a GA, but would still be a Darwinian process. You see something like this when artificial selection limits the fitness function to one or two criteria desired by man. You seem to be demanding that the fitness function arise spontaneously in the simulation. But, just like the real world, any simulation must include a process for selection based on something.Mark Frank_{January 24, 2009
January
01
Jan
24
24
2009
01:46 AM
1
01
46
AM
PDT}

gpuccio, I have agreed with your general argument, though not the details, for years. But you are have not spoken to death of virtual organisms. You also have not addressed the massive parallelism of life. There are an estimated 5 X 10^30 bacteria on earth at present. I won't guess how many elementary self-replicators there might have been prior to cellular life, but will suggest that they may have reproduced more rapidly than bacteria. In any case, our computational resources are relatively tiny. Various folks, including me, who are interested in both information and evolution think of organisms as modeling their environments. As long as the virtual environments are uninteresting, there is no reason to expect virtual organisms to be anything but the same. One thing that makes an environment "interesting" is that there are many ways to garner resources and reproduce. I should mention that GAs have discovered bugs in application-specific software the GA developers had not examined. That is, the GAs came by fit individuals that exploited errors made by programmers other than those who wrote the GAs. This is not sensational, but it is qualitatively what you said you wanted to see. P.S.--Thanks for pointing me to SCOP and BLAST. I'm interested, but I just can't follow up at the moment.Sal Gal_{January 23, 2009
January
01
Jan
23
23
2009
05:52 PM
5
05
52
PM
PDT}

gpuccio[62] How about this? A 256 bit "DNA" based fractal generator. char DNA[32]; where DNA[0] to DNA[15] select any one of f1() to f16() in any order with the parameters from DNA[16] to DNA[31]. x=f1(DNA[16],0){yadadyada..}; x=f2(DNA[17],x){yadayada...}; x=.... r=f16((DNA[31],x){yadayada...}; The result r contains the seed and fractal generator. The final output, determined at birth, is used either to generate a video bitmap or fed to a DA converter for sound output which means each DNA/life, doesn't know what it's generating as an output, audio or video. Lifeforms die based on voter input. If too many visitors on the website don't like the song or picture, it's removed from the environment. If an output gets a lot of thumbs up, it survives and modifies a bit. If it survives long enough, (100 votes?), it gets to mate with another successful DNA/life whose offspring have attributes of both parents. In the US, enough generations might produce a picture of an eagle, while in Italy, you might end up with something that sounds like opera.Toronto_{January 23, 2009
January
01
Jan
23
23
2009
04:51 PM
4
04
51
PM
PDT}

R0b: I will try to clarify. 1) The system must not be designed to run the simulation, just as the environment (in darwinian theory) is not designed to produce life (or any higer form of life). The system will obviously have its rules of functionality (its "fitness functions"), but those rules must not have been written in view of the simulation. Just as it would have happened in natural history, according to darwinian theory. 2) Yes, we are simulating biological evolution in a computer environment. Obviously, a computer system cannot have the characteristics of our terrestrial environment, neither in my simulation nor in any other GA simulation. And, least of al, can the digital replicators have the characteristics of true biological replicators. But that's not the point. The point is that we are simulating the model according to which random noise, plus a natural selection deriving from the unprogrammed interaction between modified replicators and a fucntional environment, can build up new functional information and generate more complex and functional replicators. That's the point of the simulation. Even if the digital context is obviously different from the biological one (and there is no escape for that, in any digital simulation), the logical model is the same. So the simulation, although not perfect, is appropriate, while traditional GAs are not appropriate because they are not modeling a blind natural selection, and therefore those models are logically completely different from what they are trying to simulate. 3) The point is exactly to verify if what you call "the replicators that are better aligned to the environment" will really arise in this model. I believe and expect they never will. 4) You say: "I don’t know where the line is between “more functional” and “new function”, so that’s something you’ll need to define for me.". There is no line. Any new replicator which is better aligned to the environment and spontaneously expands in the environment to the expense of the previous forms is more functional and has developed a new function. There is no difference between the two concepts. 4) Personally, I do believe that those more functional replicators will never arise in such a system, as I believe that they have never arised in natural history by a similar causal mechanism. But I am ready to analyze any possible outcome of such a simulation. Morever, if a new form of replicator is really selected, it would be easy to verify what the new function is, how it gave the new replicator a reproductive advantage, and above all what is the statistical boundary which has been overcome (how many functional bits have been added). According to darwinian theory, even if the first advantages may be trivial (within the range of statistical expectations), in time the accumulation of such useful variations should give functional variations of the order of CSI (500 bits). My belief and expectation, on the contrary, is that we would never ackowledge even the first, microevolutionary steps, or if we do they will be absolutely trivial variations, and they will stop at that level. I hope that is clear. I really believe that, if we want to simulate the logical model of darwinian theory, we have to proceed that way. Otherwise, we are only playing games, doing one thing and claiming we are doing another. And, by the way, that's what I meant when I said that GAs don't model anything. I meant that they don't model anything even distantly related to the logical model of darwinian evolution. But it is true that they can model an intelligently designed way to intelligently find a solution.gpuccio_{January 23, 2009
January
01
Jan
23
23
2009
03:44 PM
3
03
44
PM
PDT}

gpuccio, first of all, I don't understand what you mean when you say that the system must not be designed to run the simulation. What are we simulating? Biological evolution? If so, then the system should have the characteristics of our terrestial environment. It needs to simulate whatever it's supposed to simulate. Regardless, selection will be determined by the rules and characteristics of the system. The replicators that are better aligned to the environment will reproduce more. Such replicators are more functional, in terms of that environment, than other replicators. I don't know where the line is between "more functional" and "new function", so that's something you'll need to define for me.R0b_{January 23, 2009
January
01
Jan
23
23
2009
03:12 PM
3
03
12
PM
PDT}

R0b: "Active information is defined in terms of a target. What is the target of the proposed simulation?" That's exactly the point: my proposed simulation has no target, out of verifying if a simulation which has no target can obtain some new functional target. That's exactly the point, and that's exactly what darwinian evolution is believed to be: a process with no target, which can find some extraordinarily functional targets. I don't believe that can happen. But darwinists believe exactly that. So why don't they try to simulate that kind of process, instead of trying to realize processes which see very well, and then pretending that they are simulations of a blind process? Again, and to be clear to the extreme, I admit that the simulation can exercise natural selection, but not intelligent selection. In other words, the selection must arise spontaneously from the interaction between the replicator and the system, without any previous programming. In that way, and only in that way, a true (blind) natural selection can take place versus any possible new useful function which may arise from true blind random variation. AQny possible useful new function is the target. No specific target at all, the hreatest possible target of all: any possible new useful function. True blindness, true random variation, true blind natural selection. Isn't that what darwinists have been preaching for all these years?gpuccio_{January 23, 2009
January
01
Jan
23
23
2009
02:29 PM
2
02
29
PM
PDT}

gpuccio:
GAs know a lot about what they want to find, and how to find it.
I said that GAs typically know nothing about the location of the target. But yes, what they are trying to find and how to find it is incorporated into the GA.
The weasel is just a gross example, but wouldn’t you agree that the weasel algorithm knows exactly the solution it is searching for?
No, I wouldn't, but I think our disagreement is semantic. You seem to see the objective function as part of the GA, while I see the two as separate. The weasel search algorithm, for instance, starts with a random string precisely because it has no idea what the target string is. The fitness function (oracle), on the other hand, knows the target string.
I am not specially interested in GAs, being absolutely convinced that they are completely useless, and that they don’t model anything, except for the ability of their programmer.
And yet GAs are able to find solutions that their programmers don't know beforehand. And I'm not sure what you mean when you say that they don't model anything. If you're referring to biological evolution, then obviously nobody claims that computers can model that with any degree of fidelity, although they can provide us with some insight into the variation+selection process on a small scale.
...and, above all, no active information introduced in the system.
Active information is defined in terms of a target. What is the target of the proposed simulation?R0b_{January 23, 2009
January
01
Jan
23
23
2009
12:41 PM
12
12
41
PM
PDT}

R0b: "GA’s typically know nothing in advance about the location of optima, but they do act on the assumption that the landscape is reasonably smooth." I don't agree. GAs know a lot about what they want to find, and how to find it. The weasel is just a gross example, but wouldn't you agree that the weasel algorithm knows exactly the solution it is searching for? Other GAs may not know exactly the solution, but they do know a lot of other precious information. They don't certainly know (or assume) only "that the landscape is reasonably smooth". I am not specially interested in GAs, being absolutely convinced that they are completely useless, and that they don't model anything, except for the ability of their programmer. But many times, here and elsewhere, I have declared what a true computer simulation of darwin's theory should be, usually receiving no answer or comments. IMO, such a simulation should work more or less as follows: 1) Take a digital environment: a computer running some operating system, and any software we like. The digital environment can be stable or change, but the important point is that it should in no way be programmed for the simulation. 2) Take a program which generates digital replicators, and let it run in the system. The program can incorporate some system of random variation, which can be regulated as we like, but it has to be completely random, and we can apply any random probability distribution we want, but again then important point is that no programming must be introduced which has any relationship with the simulation. The whole system will so be blind to the simulation, exactly like darwinian theory assumes. 3) Just let the system and the software run, and wait. For what? For any variation in the replicator which is spontaneously selected by the system as useful. That would really be an evolutionary simulation. You have everything: a system which is not designed to run the simulation (in other words, a blind system), but which has its own rules and characteristics, like any true landscape; replicators which start as functional (we are not simulating OOL here), and are subject to random variation, adjustable as we like; and, above all, no active information introduced in the system. That, and only that, would be a real simulation of darwinian evolution. Can you see the differences with existing GAs?gpuccio_{January 23, 2009
January
01
Jan
23
23
2009
12:07 PM
12
12
07
PM
PDT}

Joseph:
2- Whether or not “evolution” is goal-oriented is being debated.
Interestingly, Marks and Dembski's work cannot address that question, because you have to know that there is a goal and know what it is before you can even begin an analysis in their framework.
3- In any goal-oriented scenario the goal is being searched for.
Actually, we need to distinguish between a goal in the codomain of the objective function vs. a goal in the domain. If, as in a prototypical optimization problem, the goal is defined in terms of the codomain, it makes sense to describe the process as a search for a solution to a problem. If the goal is defined in terms of the domain, then we're not searching for anything, but simply moving toward a goal whose location is already known. Correct me if I'm wrong, but ID scenarios seem to imply the latter. That is, the designer was not using evolution to find a solution to a problem, but rather to instantiate an already-known solution.R0b_{January 23, 2009
January
01
Jan
23
23
2009
12:05 PM
12
12
05
PM
PDT}

Joseph[50]
2- Whether or not “evolution” is goal-oriented is being debated.
gpuccio @[48] seems to agree that the process of evolution has no goal.
..because a true evolutionary model must know nothing in advance of the solution it finds.
If evolution had a goal it would be a form of ID.Toronto_{January 23, 2009
January
01
Jan
23
23
2009
11:13 AM
11
11
13
AM
PDT}

gpuccio:
The concept is probably that the author of a simulation of evolutionary search would want, and indeed does want, to hide how his algorithm performs well only because of the specific information about the solution which has gone into the programming of the algorithm.
The active information metric measures the algorithm's performance relative to the given problem. To say that an algorithm performs well because of X bits active information is to say that it performs well because it performs well.
The idea is simple: that makes the algorithm efficient, but it makes it a very bad evolutionary simulation, because a true evolutionary model must know nothing in advance of the solution it finds. Is that clear?
Actually, no. Efficient algorithms have lots of active information, by definition. Does that make them bad? "True evolutionary models", assuming they perform better than random sampling, have plenty of active information. GA's typically know nothing in advance about the location of optima, but they do act on the assumption that the landscape is reasonably smooth. The correctness of that assumption constitutes active information.R0b_{January 23, 2009
January
01
Jan
23
23
2009
10:51 AM
10
10
51
AM
PDT}

Thanks Prof. Olofsson. On reading II.B, I see that I was completely misinterpreting their vertical NFLT examples. It appears to be perfectly general. That'll teach me to try to speed-read technical papers.R0b_{January 23, 2009
January
01
Jan
23
23
2009
10:08 AM
10
10
08
AM
PDT}

Prof. Olofsson: "As for the kind of papers you ask for, you can check out Rick Durrett’s publications for a start." Does he show how background noise (chance) combined with an arbitrary collection of laws (set of laws put together absent any consideration for future results -- absent foresight) will produce a system of signs and the mechanisms to process them, a search space amenable [where functions aren't spaced too far apart] to a ratcheting filter (natural selection), and the laws necessary to create environments that will allow life to evolve to ultimately highly improbable and functional machines and intelligence (systems with the ability to model the future and generate targets). Basically does he show that life and evolution to produce intelligence will occur from any arbitrarily chosen set of laws and initial conditions? I would definitely be interested in seeing such research.CJYman_{January 23, 2009
January
01
Jan
23
23
2009
09:45 AM
9
09
45
AM
PDT}

ROb[44], As you can see, there is no mention of fitness functions in the "search for a search" paper. A "search" is defined as a probability measure over the search space (see the construction on page 2, B); this probability measure may or may not involve the use of fitness functions.Prof_P.Olofsson_{January 23, 2009
January
01
Jan
23
23
2009
07:19 AM
7
07
19
AM
PDT}

Joseph[51], You're avoiding th issue. If you claim these papers to be "pro-ID" you ought to demonstrate what bearing they have upon biology. As for the kind of papers you ask for, you can check out Rick Durrett's publications for a start.Prof_P.Olofsson_{January 23, 2009
January
01
Jan
23
23
2009
07:06 AM
7
07
06
AM
PDT}

djmullen: No, things are very different from what you are saying. You say: "But every living organism that exists is the product of organisms that are already in that tiny proportion of the possible genomes and if they successfully reproduce then their DNA is already in it too." Absolutely not. First of all, we cannot speak of the whole genome (that would be beyond any possible analysis). We have to analyze some single portion of genomes. As I have said (see my answers to Sal Gal) the only portion we can really compare (because it is the only portion we really understand) is the protein coding genes. Now, while there are some similarities between some proteins in different species, others are completely different. Practically each species has proteins which are absolutely peculiar to that species. I quote here form one previous post of mine: "The proteins we do know (and we know a lot of them) are really interspersed in the search space, in myriads of different and distant “islands” of functionality. You don’t have to take my word for that. It’s not an abstract and mathematical argument. We know protein sequences. Just look at them. Go, for example, to the SCOP site, and just look at the hyerarchical classification o protein structures: classes (7), folds (1086), superfamilies (1777), families (3464). Then, spend a little time, as I have done, taking a couple of random different proteins from two different classes, or even from the same superfamily, and go to the BLAST site and try to blast them one against the other, and see how much “similarity” you find: you will probably find none. And if you BLAST a single protein against all those known, you will probably find similarities only with proteins of the same kind, if not with the same protein in different species. Sometimes, partial similarities are due to common domains for common functions, but even that leaves anyway enormous differences in term of aminoacid sequence." Just to give you an idea: bacteria have about 500-2000 protein genes, while humans have 20000-25000. Drosophila has about 14000. Each protein, and each protein fucntionality, is a different island in the sae of possib le sequences. We have hundreds of thousands of different proteins. Some of the most complex proteins are more than 2000 aminoacids long. So, maybe you are not familiar with biology (no problem there), but when you say: "This keeps the results of their so-called “search” either inside the sweet spot or very close to it. No other “search strategy” is necessary." you really don't know what you are speaking of.gpuccio_{January 23, 2009
January
01
Jan
23
23
2009
06:12 AM
6
06
12
AM
PDT}

To all of those who doubt these papers are pro-ID: Can you point to any peer-reviewed papers that demonstrate the power of blind, undirected processes pertaining to biology?Joseph_{January 23, 2009
January
01
Jan
23
23
2009
05:11 AM
5
05
11
AM
PDT}

Evolution is a blind, unguided process. ID is a goal-driven process. Since ID has it’s goal and evolution doesn’t care for one, neither are searching for anything.
1- "Evolution" is not being debated 2- Whether or not "evolution" is goal-oriented is being debated. 3- In any goal-oriented scenario the goal is being searched for.Joseph_{January 23, 2009
January
01
Jan
23
23
2009
05:08 AM
5
05
08
AM
PDT}

What I don't understand is what does searching have to do with evolution? It looks to me like the ultimate concern of the authors is searching through all possible genomes for the tiny proportion of those genomes that will construct and operate an organism that can successfully reproduce. But every living organism that exists is the product of organisms that are already in that tiny proportion of the possible genomes and if they successfully reproduce then their DNA is already in it too. In practice, they do this by making only small changes to the DNA they hand down to their offspring. This keeps the results of their so-called "search" either inside the sweet spot or very close to it. No other "search strategy" is necessary. Evolution says that every living organism since the first self-reproducing molecule is already in the genomic sweet spot and all they have to do is keep their offspring in it too.djmullen_{January 23, 2009
January
01
Jan
23
23
2009
01:30 AM
1
01
30
AM
PDT}

R0b: "What I find interesting is the idea that someone would want to hide how well their algorithm performs." The concept is probably that the author of a simulation of evolutionary search would want, and indeed does want, to hide how his algorithm performs well only because of the specific information about the solution which has gone into the programming of the algorithm. The idea is simple: that makes the algorithm efficient, but it makes it a very bad evolutionary simulation, because a true evolutionary model must know nothing in advance of the solution it finds. Is that clear?gpuccio_{January 22, 2009
January
01
Jan
22
22
2009
10:45 PM
10
10
45
PM
PDT}

Bill Dembski, When you first announced your publications here at UD, you linked them to complex specified information. Yet I can find no explicit reference to CSI. And I genuinely do not see it between the lines. Would you explain the connection to CSI?Sal Gal_{January 22, 2009
January
01
Jan
22
22
2009
09:32 PM
9
09
32
PM
PDT}

gpuccio:33 "Well, NS is a very limited oracle, and it can only judge if a reproductive advantage has been achieved or not." Whether an agent reproduces or not is always a binary event, obviously. How NS differnetiates between agents that reproduce or not is not a binary event, it is a probability. Roughly speaking the fitter the individual the greater their probability of reproducing and, depending on the type of GA, higher fitness individuals have a greater chance of producing greater numbers of offspring. "NS can do only two things: expand a genome if a perceptible reproductive advantage has been achieved (positive selection); or eliminate it if there has been a significant loss of function (negative selection)." Genomes with a lower fitness than their parents still have a probability of reproducing, it is just lower. Genomes with the same fitness as their parents have the same probability of reproducing, even if they are different. And that all assumes that the fitness landscape stays the same, which in biology and in some GA implimentations is not always the case - you can end up fitter than your parents because of an environmental change even if at birth you were less fit. Either way we are talking (and the paper is talking) about search algorithms and my point still stands - A binary fitness evaluation presents no slopes, even if they actually exist in the fitness landscape. Removing the ability to detect any graduated differences in fitness between individuals or iterations effectively converts a variable fitness landscape into a flat one. A hill climbing algorithm is ONLY an hill climbing algorithm if there is a hill to climb and a GA is ONLY a GA if there is some way of ranking individuals. Otherwise you just have an individual taking a random walk, or a population of individuals taking a random walk. To imply that these strategies are 'adding information' is like arguing that an aeroplane is no better at flying than a car because the aeroplane can't fly without wings.Laminar_{January 22, 2009
January
01
Jan
22
22
2009
05:04 PM
5
05
04
PM
PDT}

A comment on the last conclusion of the 1st paper:
To have integrity, search algorithms, especially computer simulations of evolutionary search, should make explicit (1) a numerical measure of the difficulty of the problem to be solved, i.e., the endogenous information, and (2) a numerical measure of the amount of problem-specific information resident in the search algorithm, i.e., the active information.
So whoever publishes an algorithm for solving a particular problem should make explicit (1) the size of the problem space and (2) the efficiency of the algorithm applied to that problem. What I find interesting is the idea that someone would want to hide how well their algorithm performs.R0b_{January 22, 2009
January
01
Jan
22
22
2009
04:39 PM
4
04
39
PM
PDT}

Not that anyone cares, but I answered my own question from #21. The piece I was missing was the fact that the amount of active information changes over the course of the search. The fact "the fitness function is smooth" contains no active information initially, since by itself it provides no guidance as to the location of the target. That's why the first query of genetic algorithms is typically random. It's only after we start querying that the distribution starts to become biased. So maybe the vertical NFLT is general enough to handle all kinds of information about the target and fitness landscape. I would be interested to see it formally applied it to a smooth fitness landscape.R0b_{January 22, 2009
January
01
Jan
22
22
2009
03:41 PM
3
03
41
PM
PDT}

Sal Gal: Long time no see, friend. It's always a pleasure to meet you. You say I may have missed your point, but I don't believe that. I have purposefully avoided to discuss Dembski and Marks'points, mainly for lack of competence on my part. But I have commented on what you were saying. I quote again: "There is huge inconsistency among IDists on the matter of functionality of DNA sequences." That's not true, as I have tried to show. We do believe that, in general DNA sequences are functional, but in different ways: protein coding genes are functional as "storage" of aminoacid sequences; non coding DNA is functional as a regulatory information, at present poorly understood, and probably much more complex than the "simpler" protein coding information. Let's say that protein coding DNA codes for the "effectors", while non coding DNA codes, in some mysterious way, for the procedures. Or for part of them (other unknown components will probably be discovered in time). Let's remember that at present we know almost nothing about the procedures. We don't know how transciption is regulated, how and why cells differentiate in myriad of forms, how multicellular plans are achieved, how intercellular communication is controlled at higher levels of integration, and so on. Procedures. In any software, those are the real thing. I quote again: "But when they want to portray evolution as utterly improbable, they say that a sequence of bases is categorically fit if it codes for a prespecified protein, and is categorically unfit otherwise. Now which is it? Can a non-coding sequence contribute to fitness, or not?" As already said, we reason on the protein coding genes, because that is the part we understand (and that "we", for once, includes both darwinists and IDists). We will reason about the regulation part when we will understand it. And believe me, that will not be good news for darwinists. So, leaving alone Dembski and Marks, where is it that I have missed your point? And is my point clear? To answer explicitly your question, a non coding sequence can certainly contribute to fitness, but only to the fitness related to its specific function, probably a regulatory one. It cannot certainly contribute to the fitness inherent in a protein coding gene, which is another thing, works in another way, has another symbolic code, and so on. So, for protein coding genes, we have to go back to the only possible oracle if we are to believe in unguided, non designed evolution: NS. I have dealt with that in my previous post, so I refer you to that. In other words, all "warmer or colder" effect must be attributed only to natural selection (or to design). And it is a very "threshold" effect: only if the message is "very warm" (the new replicator replicates better than the old ones, and can expand in the population) can the variation in genome be fixed, and we have something different from a random search on a uniform distribution. Or, in alternative, if the message is "very cold", and the new genome is terminated. In all other cases, there is no oracle, and no other supporting or miraculous force. We are back to random search over an uniform distribution of possibilities. Or to design.gpuccio_{January 22, 2009
January
01
Jan
22
22
2009
03:08 PM
3
03
08
PM
PDT}

pubdef: "What I’m trying to say is: notwithstanding the characterization of DNA as “digital information,” isn’t it really a physical configuration of molecules that interacts with other physical objects/particles, leading to a far flung range of physical consequences? Another stab at it: DNA seems to be viewed as software, but isn’t it really hardware?" As Upright BiPed has already said, DNA is a support for information: you cannot have pure software, you always have software written on a hardware support. The "physical configuration of molecules that interacts with other physical objects/particles" is only the biochemical structure of the DNA molecule. But the special sequence of nucleotides which, in a symbolic code, encodes the sequence of aminoacids in a specific protein is pure digital information. There is no physical or biochemical law which determines that sequence. The sequence is preserved in DNA, and it must have been created in some way. Darwinists believe that it was created through RV + NS. We (IDists) believe that it is the product of designe. But in no case it can be the product of necessity. There is no law of necessity which can output the correct sequence of nucleotides which corresponds to the aminoacid sequence of, say, myoglobin, according to a definite symbolic genetic code (the genetic code we observe in biological beings). Darwinists, and even many biologists, are really confused when they speak of non uniform distributions, or of landscapes, or of fitness functions, as though those concepts could apply to the random generation of new nucleotide sequences. There are no laws which can generate randomly a sequence of nucleotides in a pattern which is really distant from an uniform distribution, least of all in a pattern which may have anything to do with the genetic code or with functional aminoacid sequences. All that talk about evolutionary algorithms is just smoke in your eyes. The truth is that any component in darwinian theory which can apply to the possible generation of information in a way which differs from a random search on a uniform distribution must be related to NS. And NS is a form of necessity. Call it evolutionary algorithm or any other name, it is only NS which can make a difference. But NS is not an omnipotent principle. As all laws of necessity, it must be modeled according to necessity. It is indeed a potential oracle, but an oracle of which we know all too well the limits: it can only expand genomes with a reproductive advantage or eliminate genomes with important loss of function. Out of those limits, NS is non existent. So, again, darwinists should be able to show where NS can act to "join" informational leaps which are in the range of what a random search over an uniform distribution can empirically accomplish. If they cannot do that, they have nothing: not a model, not a theory, nothing. And believe me, they cannot do that.gpuccio_{January 22, 2009
January
01
Jan
22
22
2009
02:47 PM
2
02
47
PM
PDT}

P.S. to 39: Free associating on Toms -- that's Schneider, not Ray.Sal Gal_{January 22, 2009
January
01
Jan
22
22
2009
01:36 PM
1
01
36
PM
PDT}

gpuccio, Long time no see, friend. I think you may have missed my point, which was a criticism of modeling the fitness of a DNA sequence as all-or-nothing. I am saying that I think it's inappropriate for Dembski and Marks to attribute all "warmer or colder" information to an "oracle" or an "assistant."Sal Gal_{January 22, 2009
January
01
Jan
22
22
2009
01:26 PM
1
01
26
PM
PDT}

So if you’ve got a problem with the applicability of the research at the Evolutionary Informatics Lab to real-life biological evolution, take it up with Schneider and Pennock.
Evolution of Avida programs is not combinatorial optimization. When the running time for Avida is bounded, proceeding from shorter programs to longer programs is generally superior to proceeding from longer programs to shorter programs. Avida affords itself of that free lunch. Only you and Bob, to my knowledge, have tried to bring "no free lunch" results to bear on Avida. So I believe it is reasonable to take up the issue with you. And I'm curious as to whether you and he elected to remove the attacks on ev and Avida from the first paper, or if the reviewers required it. As for ev, Ray has contended that you have not understood what he meant to demonstrate. I have not reread his paper to see if he's right, and I won't bother, because it seems to me that you're beating a dead horse. Computational studies today are more sophisticated than ev.Sal Gal_{January 22, 2009
January
01
Jan
22
22
2009
01:10 PM
1
01
10
PM
PDT}

Prev 1 … 6 7 8 9 10 Next

You must be logged in to post a comment.

Leave a Reply