New Peer-Reviewed ID Paper — Deconstructing the Dawkins WEASEL

_{William Dembski

March 9, 2010

Evolution, Informatics, Intelligent Design}

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

Winston Ewert, George Montañez, William A. Dembski, Robert J. Marks II, “Efficient Per Query Information Extraction from a Hamming Oracle,” Proceedings of the the 42nd Meeting of the Southeastern Symposium on System Theory, IEEE, University of Texas at Tyler, March 7-9, 2010, pp.290-297.

Abstract: Abstract—Computer search often uses an oracle to determine the value of a proposed problem solution. Information is extracted from the oracle using repeated queries. Crafting a search algorithm to most efficiently extract this information is the job of the programmer. In many instances this is done using the programmer’s experience and knowledge of the problem being solved. For the Hamming oracle, we have the ability to assess the performance of various search algorithms using the currency of query count. Of the search procedures considered, blind search performs the worst. We show that evolutionary algorithms, although better than blind search, are a relatively inefficient method of information extraction. An algorithm methodically establishing and tracking the frequency of occurrence of alphabet characters performs even better. We also show that a search for the search for an optimal tree search, as suggested by our previous work, becomes computationally intensive.

[ IEEE | pdf ]

Comments

WinstonEwert at 27, I said, “Where does nature get a good oracle”. Certainly we can treat nature as something of an oracle. But its not a good oracle. The NFLT tells us that across the space of all possible oracles all search algorithms have equally bad results. Actually, the No Free Lunch theorems tell us that a particular algorithm will perform no better than blind search on average across all possible search spaces. That's completely immaterial to this discussion, however. If you want to model biological evolution as a search, then you must use the known laws of physics and chemistry and the observed environment of the organisms under consideration as the "search space." There is no "search for a search," the fitness landscape, while dynamic, is a given. It turns out that, in the search space of our real world, the mechanisms identified by modern evolutionary theory work sufficiently well to allow populations to change in response to feedback (in the form of differential reproductive success) from the environment. This isn't a surprise -- if those mechanisms weren't useful in this particular search space, we would see different, more useful mechanisms instead (or we wouldn't be here to discuss it in the first place). We have no reason to suppose that an evolutionary strategy will be able to use a nature-based oracle in an effective manner. Of course we do. It's called empirical evidence. We can see these mechanisms working.Mustela Nivalis_{March 11, 2010
March
03
Mar
11
11
2010
06:41 AM
6
06
41
AM
PDT}

Cable at 25, I think you are missing the point. Stating the Oracle is nature is about as empty a statement as can be made. Any emptiness is a limitation of the model, not of the observed ability of the mechanisms identified by modern evolutionary theory to modify subsequent populations in response to feedback from the environment. What in nature provides feedback about an intermediate state i.e. one that does not provide an immediate benefit. Natural selection cannot give feedback about an intermediate state since there is no survival benefit at this point. True, as far as it goes. However, neutral and even mildly deleterious mutations can persist and spread through a population. Therefore it is essentially a blind search. This is incorrect and does not follow from your premises. Since the search space in so massive mutation and natural selection by themselves simply don’t have the time/resources/opportunities to find beneficial states without an Oracle helping or essentially reducing the search space. The mechanisms identified by modern evolutionary theory do not search the whole space of possible viable organisms. They can be modeled (and again, it's important to realize that this is just a model) as searching the immediate neighborhood of known-viable organisms. Empirical evidence shows that this is a useful strategy.Mustela Nivalis_{March 11, 2010
March
03
Mar
11
11
2010
06:31 AM
6
06
31
AM
PDT}

Mr Ewert, If the model is that the oracle is a channel through which information is returned in response to queries, then I would say that it is inaccurate to say that nature is its own oracle. Rather, differential survival is the oracle. Differential survival is a noisy channel. There is no guarantee that it is a "good oracle". But nature doesn't have a choice of oracles.Nakashima_{March 11, 2010
March
03
Mar
11
11
2010
04:00 AM
4
04
00
AM
PDT}

scordova at 18 & 19 The same argument works against an Intelligent Designer. If an ID creates a set of DNA that works in one environment and that environment changes so much the DNA doesn't work anymore, the ID will have to either design some updated DNA or watch his creation die.warehuff_{March 11, 2010
March
03
Mar
11
11
2010
02:25 AM
2
02
25
AM
PDT}

@William Dembski Amusing thought: the remarkably good performance of the FOO Hamming oracle algorithms for the Hamming oracle results in a much worse performance of this algorithm for other oracles - an obvious conclusion of the New Free Lunch theorem. @Winston Ewert: using the standard notation helps. But at least be consistent - your variance of ES(1+1) is introduced as the Rachtet Strategy in this paper, and was called Optimization by mutation with elitism in Conservation of Information.DiEb_{March 11, 2010
March
03
Mar
11
11
2010
01:23 AM
1
01
23
AM
PDT}

Cable @23,
What in nature provides feedback about an intermediate state i.e. one that does not provide an immediate benefit.
From our point of view, i.e., of life, nature is not a series of quantum states. Our fitness in our environment is not measured every hour, season, anniversary, or at any fixed interval. While a mathematical model has a frequency of testing for fitness, it is not applicable to real-time life. Any mutation that does not negatively impact an organism's fitness will survive just like any other neutral characteristic.
If we use the model of an oracle-led search (keeping in mind that it is a model) then the oracle is nature.
Please keep in mind as was mentioned by Mustela Nivalis, we are dealing with a model and we should treat it as such.Toronto_{March 10, 2010
March
03
Mar
10
10
2010
02:51 PM
2
02
51
PM
PDT}

I said, "Where does nature get a good oracle". Certainly we can treat nature as something of an oracle. But its not a good oracle. The NFLT tells us that across the space of all possible oracles all search algorithms have equally bad results. We have no reason to suppose that an evolutionary strategy will be able to use a nature-based oracle in an effective manner.WinstonEwert_{March 10, 2010
March
03
Mar
10
10
2010
02:02 PM
2
02
02
PM
PDT}

Off Topic: Now this is cool: Leading Intelligent Design Advocate Challenges Former President of American Association for the Advancement of Science (AAAS) to Debate: The Discovery Institute has invited Dr. Francisco Ayala to debate the thesis of the book Signature in the Cell: DNA and the Evidence for Intelligent Design with the book's author, Dr. Stephen Meyer. http://www.evolutionnews.org/2010/03/leading_intelligent_design_adv.htmlbornagain77_{March 10, 2010
March
03
Mar
10
10
2010
01:59 PM
1
01
59
PM
PDT}

@20 Mustela Nivalis I think you are missing the point. Stating the Oracle is nature is about as empty a statement as can be made. What in nature provides feedback about an intermediate state i.e. one that does not provide an immediate benefit. Natural selection cannot give feedback about an intermediate state since there is no survival benefit at this point. Therefore it is essentially a blind search. Since the search space in so massive mutation and natural selection by themselves simply don't have the time/resources/opportunities to find beneficial states without an Oracle helping or essentially reducing the search space. This is what Weasel and Avida do. So where does nature get the Oracle?Cable_{March 10, 2010
March
03
Mar
10
10
2010
01:22 PM
1
01
22
PM
PDT}

WinstonEwert at 8, Where does nature get a good oracle? If we use the model of an oracle-led search (keeping in mind that it is a model) then the oracle is nature. That is, the determinant of fitness is the laws of physics and chemistry, the environment, and even the rest of the organism's population. What we observe is that the mechanisms identified by modern evolutionary theory are capable of communicating the results from that "oracle" such that they are represented in subsequent populations. These mechanisms don't have to be optimal, just good enough.Mustela Nivalis_{March 10, 2010
March
03
Mar
10
10
2010
10:01 AM
10
10
01
AM
PDT}

How did you determine that the oracle is good? The oracle is good because it is possible to extract a large amount of active information from it. What is nature’s search space and target? The search space is the space of all biological forms. The simplest way of looking at the target is to view it as a moving target always targeting something better then the current point. However, appealing to a moving target is really not going a help a search process. At any rate, if you are going to take that position it means that most if not all computer simulations of evolution are invalid. But we can pick any deterministic natural process, declare that the outcome is the target, and we have boatloads of easy active information. In doing so, you've create a target and search algorithm pair which match exactly. The search algorithm essentially knows what the target is. You don't the equivalent of taking a negative. You already knew the outcome of the process, you've just transformed that into a target. That is a neat trick you can pull but its not going to help evolution find a way to let the deer escape from the lion. The whole point of a search process is that you don't know what the target is. Hopefully, you'll be able to recognize what you are looking for when you see it.WinstonEwert_{March 10, 2010
March
03
Mar
10
10
2010
09:14 AM
9
09
14
AM
PDT}

Evolution, if you will, gives the appearance of "search," and that itself requires explanation. --WmAD]
Thank you for your reply. I agree 100% that it gives the "appearance" of a search. If we're going to model evolution as a search algorithm though, we really need to get closer to what it really is which is a search for multiple dynamic targets, coupled to multiple feedback mechanisms, with a varying population size. Now something like that would give both sides a model they could work with.Toronto_{March 10, 2010
March
03
Mar
10
10
2010
06:45 AM
6
06
45
AM
PDT}

scordova @19,
But a moving target implies selection will no longer favor a solution once it is found. What works today, is rejected tomorrow.
Selection "implies" the target,meaning that the target is not explicitly defined by some other process. Selection allows a mutation to survive as a match to a different target, and this is what is wrong with modeling evolution as a single target search algorithm. A proper model of evolution as a search algorithm would be multiple varying targets with varying population sizes, coupled with feedback. In real life, a predator's food supply is inversely proportional to his success in getting it, and that is reflected in the oscillating population sizes of predators and their prey. Both predator and prey actively modify each other's selection criteria in an arms race. Longer legs in the prey might help it outrun it's predator but would also allow it to walk further into a pond to hunt for it's own food. Here we have two areas a mutation would be selected for even though the original mutation/selection was due to only one of them. None of this type of feedback is in the search algorithm, yet it will impact on selection. While eventually they may, currently, Dembski-Marks don't go far enough in their work to say it pertains to evolution.Toronto_{March 10, 2010
March
03
Mar
10
10
2010
04:52 AM
4
04
52
AM
PDT}

I started to have a closer look at the paper (section I, II, III A, III B, III C).DiEb_{March 10, 2010
March
03
Mar
10
10
2010
04:00 AM
4
04
00
AM
PDT}

Let me elaborate a little on my last comment. Darwinian evolution presumes that complexity is added piece by piece, that once a solution is found selection will continue to favor it and then favor subsequent improvements. But a moving target implies selection will no longer favor a solution once it is found. What works today, is rejected tomorrow. For example, the idea that a piece of what will be eventually part of a bacterial flagellum will be discovered and preserved will fail in a moving target search. The reason the piece was accepted today would not be grounds for it to be accepted in the future, in fact, it would be selected against since the newly favored target would cause rejection of the previously selected target, and hence the pre-cursor is disposed of like an expired password.scordova_{March 9, 2010
March
03
Mar
9
09
2010
11:09 PM
11
11
09
PM
PDT}

Toronto: If you could come up with a mathematical analysis of a search algorithm that took into account a changing target, it would be a much better model of what evolution does.
The moving target search is doesn't help the Darwinian argument. What this would imply is once evolution finds a solution, then next day it is invalid because what worked yesterday, won't today. This is like finding a password and then immediately it becomes useless for the next go around. And even granting this is the way evolution works, it is no way to accumulate complexity, but it rather guarantees only the simplest solutions will be found since the wealth of prior evolutionary search discoveries become invalidated by the moving target. The solution you propose doesn't help the Darwinian case.scordova_{March 9, 2010
March
03
Mar
9
09
2010
10:13 PM
10
10
13
PM
PDT}

Mr. Ewert, First of all, congratulations on this paper, as well as the Avida paper.
Where does nature get a good oracle?
How did you determine that the oracle is good? What is nature's search space and target? One point that may bear noting: Talk of oracles and problem-specific knowledge may give some people the mistaken impression that active information necessarily comes from an intelligent source. But we can pick any deterministic natural process, declare that the outcome is the target, and we have boatloads of easy active information. Non-randomness implies causality, not necessarily sentience.R0b_{March 9, 2010
March
03
Mar
9
09
2010
09:49 PM
9
09
49
PM
PDT}

bornagain77, the “beneficial adaptation” always comes at a loss of information that was already present in the genome That is probably true, although I wouldn't state it quite so universally. There is difference though. In biology there is no easy way to improve something. It is very difficult to come up with a cluster of mutations to implement a new and useful feature. On the other hand, breaking something in a useful way is much easier. Weasel only has to change one letter to be correct. As a result, WEASEL is not really bound by that restriction. Is this what you find in your model as well? Or am I off base again? Weasel does not get a large benefit from multiple mutations since it can proceed one mutation at a time. The Avida simulation does have multiple muations however. Avida has a much higher mutations rate then biology with a much smaller genome. The result is that it can accumulate larger mutation sizes then a realistic biologic case.WinstonEwert_{March 9, 2010
March
03
Mar
9
09
2010
09:23 PM
9
09
23
PM
PDT}

WinstonEwert, Thanks for that bit of clarification, as I am lost as to exactly what has been accomplished in the paper. If you could would you clear up a few points for me? From all evidence I have been exposed to, the principle of Genetic Entropy holds for all beneficial biological adaptations, that is that the "beneficial adaptation" always comes at a loss of information that was already present in the genome ,,, I am guessing that to correctly model what we actually see in life, the information available for a search after a successful "beneficial" adaptation should be decreased within/from the oracle??? Am I correct to think this? Or am I off base? also of note; Behe has clearly clarified in "The Edge" that all known successful evolutionary searches for a beneficial adaptation are severely restricted to a couple of coordinated point mutations,,, Is this what you find in your model as well? Or am I off base again?bornagain77_{March 9, 2010
March
03
Mar
9
09
2010
08:09 PM
8
08
09
PM
PDT}

Collin:
But it is ID friendly and I can definitly see the scaffolding of an argument emerging from it. Perhaps an argument that Evolution is so inefficient that it cannot sustain positive mutations under the weight of all of the negative mutations.
If so, then that's an interesting twist, given Marks & Dembski's previous statements. In Life's Conservation Law, they said that information comes only from intelligence. Since Darwinian evolutionary processes have active information, they must be designed. Are they now saying that evolutionary processes are designed, but not designed well enough to hit the target in the allotted time?R0b_{March 9, 2010
March
03
Mar
9
09
2010
04:51 PM
4
04
51
PM
PDT}

at II C: Results: Figure 2(B) shows the active information per query for the ratchet strategy given different alphabet sizes and message lengths. Increasing the message length does not appear to significantly change the efficiency of active information extraction. However, increasing the alphabet size has a rather noticeable effect. One should thinks so : As I calculated earlier - when discussing Conservation of Information in Search - Measuring the Cost of Success - the expected number of queries for the One Child Ratchet Algorithm (a more common name would be ES(1+1)) with exactly one mutation per child is E(Q) = (N-1) * H_(1-βL)* L so that the average information per query is I = log(N)/(N-1) * 1/H_(1-βL) (β: rate of correct letters to start with H_k: k-th harmonic number) The effect of log(N)/(N-1) is noticeable, the effect of of 1/H_(1-βL) less so.DiEb_{March 9, 2010
March
03
Mar
9
09
2010
03:06 PM
3
03
06
PM
PDT}

"so is every paper in GECCO 2009. (Paging Jerry…)" Having just looked in briefly before taking off for South America, I have to say yes. Every paper in genetics is an ID paper and all those geneticists are supporting the ID paradigm whether they know it or not . Let me know when you find one that is not. So is everybody here who is anti ID supporting ID, especially Nakashima. Why Nakashima? Because he tries harder than most and still turns up nothing. No one has been able to provide an alternative explanation yet. Not even Richard Dawkins or Jerry Coyne in their recent books.jerry_{March 9, 2010
March
03
Mar
9
09
2010
01:12 PM
1
01
12
PM
PDT}

Nakashima, Yeah, maybe this isn't an ID supporting paper per se. But it is ID friendly and I can definitly see the scaffolding of an argument emerging from it. Perhaps an argument that Evolution is so inefficient that it cannot sustain positive mutations under the weight of all of the negative mutations. Maybe something like John Sanford's "Genetic Entropy" argument.Collin_{March 9, 2010
March
03
Mar
9
09
2010
12:36 PM
12
12
36
PM
PDT}

I haven't read the paper in detail yet. And it's a little bit hard to figure out the number of queries needed on average for various combination of parameters from the three-dimensional pics of the Active information per query. I'd like to compare these numbers with mine for the algorithm mutating children with a fixed mutation rate for L=28 and N=27 and L=100 and N=2. An of course the numbers for K children each with a single mutation, again for L=28 and N=27! It's good that the exchange of information from oracle to program is looked at now - that got muddled up earlier.DiEb_{March 9, 2010
March
03
Mar
9
09
2010
12:26 PM
12
12
26
PM
PDT}

Sincere congratulations to Ewert, Montañez, Dembski, and Marks. Having read it quickly, I'd say that this is the most interesting of the EIL papers for me. I was most interested in section V. I wish I had time to replicate those results and explore the characteristics of the winning functions. It seems that finding a function that achieves an optimal balance of performance and algorithmic simplicity could be a fruitful area of research. The paper also highlights a point of terminological confusion: Although active information is sometimes described as a measure of problem-specific information, this isn't strictly correct. The relevant item of information in this paper is the knowledge that the oracle is a Hamming oracle, but the resulting active information depends very much on how that information is used. Section V shows that we can, for simple cases, find the maximum amount of active info achievable with this knowledge (given values for L and N), so we can uniquely quantify problem-specific information using the active info upper bound.R0b_{March 9, 2010
March
03
Mar
9
09
2010
12:03 PM
12
12
03
PM
PDT}

I’m pretty sure you mean “single parent” not “single child” No, the point there is that we are first considering the case of a single child and then extending to that the k children case. It might also be nice at this point out that this is a standard ES( 1, lambda ) evolution strategy if your readers would like to pursue the literature on these algorithms. If it would helpful, we'll endeavor to include it in future. I've always found that particular syntax just odd and cryptic, but if it is a standard we should use it. I’m glad to see that extracting information from an oracle has replaced smuggling in information The fact that we are extracting information from an oracle means that we have an oracle which has a large amount of active information. Information is being put into the simulation by the use of this oracle. The smuggling has not been replaced, the smuggling is in the availability of an oracle. Evolution can and does work, but inefficiently. The point is that for any given oracle there is a variety of ways to extract the information from it. One of these ways is optimal. FOOHOA shows that there is a lot of information available from the oracle showing that it is very powerful. Search succeeds as a result of the oracle, the oracle is the source of information. We know where this oracle came from, humans designed it. Where does nature get a good oracle?WinstonEwert_{March 9, 2010
March
03
Mar
9
09
2010
09:30 AM
9
09
30
AM
PDT}

Dr Dembski, From the bottom of p 292, left column: We first consider the case where there is a single child and then when K children are generated and the best fitness among the children is kept for the next parent. I'm pretty sure you mean "single parent" not "single child", if you get a chance to correct the manuscript. It might also be nice at this point out that this is a standard ES( 1, lambda ) evolution strategy if your readers would like to pursue the literature on these algorithms. I think it would also be standard form to give a citation to Avida when you mention it. I'm glad to see that extracting information from an oracle has replaced smuggling in information. The fitness function is decoupled from the search algorithm, this is progress. Also nice to see the citation on antenna design. I'm beginning to sense the shape of the argument being built here, and in the previously mentioned Ewert 2009. Evolution can and does work, but inefficiently. ES, which like the real world only knows the previous generation, loses in efficiency of information extraction compared to a strategy that keeps knowledge of the entire history of queries. But I'm stll not seeing how to shift from "some algorithms are more efficient than others" to ID. If this is an ID paper, so is every paper in GECCO 2009. (Paging Jerry...)Nakashima_{March 9, 2010
March
03
Mar
9
09
2010
08:49 AM
8
08
49
AM
PDT}

Maybe when he tries to buy a football team.tribune7_{March 9, 2010
March
03
Mar
9
09
2010
08:32 AM
8
08
32
AM
PDT}

According to Wikipedia's article on Dembski: "Computer scientist and number theorist Jeffrey Shallit states in an expert report that despite common claims in the popular and religious press, Dembski is not a scientist by any reasonable standard, has not published any experimental or empirical tests of his claims, submitted his claims to the scrutiny of his peers or published in a scientific journal." I wonder when this will get corrected, now that Dembski keeps cranking out these papers with Robert Marks's Evolutionary Informatics Lab. Any bets? I'd give it another five years.kibitzer_{March 9, 2010
March
03
Mar
9
09
2010
08:02 AM
8
08
02
AM
PDT}

Wasn't Atom involved in that research too?osteonectin_{March 9, 2010
March
03
Mar
9
09
2010
07:31 AM
7
07
31
AM
PDT}

Prev 1 2 3 4 Next

You must be logged in to post a comment.

Leave a Reply