Correspondences between ID theory and mainstream theories

Per Gregory and Dr. Felsenstein’s request, here is an off the top of my head listing of the major theories I can think of that are related to Dembski’s ID theory. There are many more connections I see to mainstream theory, but these are the most easily connected. I won’t provide links, at least in this draft, but the terms are easily googleable. I also may update this article as I think about it more.

First, fundamental elements of Dembski’s ID theory:

  1. We can distinguish intelligent design from chance and necessity with complex specified information (CSI).
  2. Chance and necessity cannot generate CSI due to the conservation of information (COI).
  3. Intelligent agency can generate CSI.

Things like CSI:

  • Randomness deficiency
  • Martin Löf test for randomness
  • Shannon mutual information
  • Algorithmic mutual information

Conservation of information theorems that apply to the previous list:

  • Data processing inequality (chance)
  • Chaitin’s incompleteness theorem (necessity)
  • Levin’s law of independence conservation (both chance and necessity addressed)

Theories of things that can violate the previous COI theorems:

  • Libertarian free will
  • Halting oracles
  • Teleological causation
0

187 thoughts on “Correspondences between ID theory and mainstream theories

  1. Corneel: The whole point of the evolutionary process is to push any genome towards the region of high fitness.

    That may be the point of Weasel, but I don’t think it is “the point” in biology.

    0
  2. One might say that given an incremental path, selection will result in high fitness, but what does that mean?

    There could be hundreds of dimensions of fitness, some of which diminish when another increases.

    0
  3. petrushka: That may be the point of Weasel, but I don’t think it is “the point” in biology.

    I meant the algorithm, but it’s good that you are keeping everybody on their toes.

    petrushka: There could be hundreds of dimensions of fitness, some of which diminish when another increases.

    Yes, there are many components of fitness that may indeed trade off. But let’s not complicate matters. I think most residents here accept that organisms tend to be adapted to their surroundings: their behaviour, physiology and morphology helps them survive and reproduce in their respective environments.

    0
  4. Organisms tend to be descended from an unbroken lineage.

    Just my feels, but I see high fitness as a secular version of all things bright and beautiful.

    Legs are long enough to reach the ground. Populations are as fit as required, unless they aren’t.

    Perhaps I’m just old and grumpy.

    0
  5. I think that it’s inaccurate and misleading to call the Weasel Program an algorithm. It doesn’t compute anything nor does it solve any problem. The thing that one might think of as a solution, the string “ METHINKS IT IS LIKE A WEASEL,” is already known and is used in the program.

    Tom English described it accurately:

    Dawkins wrote a computer program to simulate a monkey at a typewriter. He referred to the program as the monkey/Shakespeare model of cumulative selection, and explained that it was not a model of cumulative natural selection. One of his main points, illustrated by the model, was that cumulative selection is not necessarily a biological process.

    0
  6. Weasel incorporates a set of computational rules. The algorithm could just as easily be applied to a changing target, perhaps one generated by a natural process. It can also produce a fit among a number of targets. It isn’t necessary for there to be a single solution, or one known to the programmer.

    0
  7. Or maybe one could take the output to be the set of all the sequences traversed by the algorithm from the initial random one to the target

    0
  8. Dawkins’s procedure is not guaranteed to halt in a finite number of steps, and thus is not an algorithm.

    That is, for all [non-negative] integers n, if the monkey has completed n selections of a “parental” phrase, then the probability is greater than zero that none of the n “parental” phrases matches the target phrase. This is easy to prove by induction.

    0
  9. Tom English,

    If there is a nonzero rate of mutation from correct to incorrect letters, the process need not stop (unless you impose a rule to stop it). Lacking such a rule, it is a stochastic process that goes on indefinitely, rather than an algorithm.

    By the way, Tom, does a binary search on the real line to find the root of a polynomial ever stop? For most starting points, no, so maybe it isn’t an algorithm either.

    0
  10. petrushka: Just my feels, but I see high fitness as a secular version of all things bright and beautiful.

    Legs are long enough to reach the ground. Populations are as fit as required, unless they aren’t.

    Perhaps I’m just old and grumpy.

    I don’t remember luminance or aesthetics coming into any measurement of fitness I have ever seen done, so I guess you ARE being old and grumpy. 😉

    Usually it is the variation in fitness within populations that interests us, and the changes in genetic composition that occur because of it, not whether a population as a whole is fit enough at some isolated point in time.

    0
  11. Gentlemen,
    I wasn’t talking about Dawkins’ weasel program either but about Eric’s evolutionary algorithm E (which admittedly looks a bit like Dawkins’ weasel). I assumed, naively perhaps, that since it is called “evolutionary algorithm”, that it was actually an .. eh .. algorithm.

    I wasn’t aware that an algorithm is required to do problem solving or perform calculations, or has to halt in a finite number of steps, so I guess I learned something. Still, I am pretty sure that I will be using the word again in the (incorrect) informal sense that many n00bs do. Apologies in advance to all.

    0
  12. Joe Felsenstein: If there is a nonzero rate of mutation from correct to incorrect letters, the process need not stop (unless you impose a rule to stop it). Lacking such a rule, it is a stochastic process that goes on indefinitely, rather than an algorithm.

    As you know, I think it’s important to distinguish the simulator and the simuland. (Dembski, Ewert, and Marks mistake analysis of a simulation for analysis of the simulated process.) For instance, an evolutionary process terminates only on extinction of a lineage. When a simulation of an evolutionary process terminates, it is usually not the case that extinction has occurred in the simulated process. Instead, the simulation terminates when the user of the simulator is no longer interested in observing the simulated process.

    If Dawkins’s monkey never miscopied correct letters, then the process of cumulative selection would enter an absorbing state when all letters in the “parental” phrase were correct (i.e., each succeeding state of the process would be identical to the preceding state). However, this would not imply termination of the process. You refer to imposing a rule to stop “it.” But we have to be careful about the referent of “it.” Does Dawkins’s monkey stop typing when the “parental” phrase matches the target phrase? Or do we stop the simulation of a neverending monkey-at-a-typewriter process when we are uninterested in further observation of it? In other words, is the termination rule part of the description of the simulated process, or is it part of the simulator?

    In the case of Dawkins’s monkey/Shakespeare model of cumulative selection, it doesn’t matter whether the monkey stops typing when the “parental” phrase matches the target, because Dawkins’s main concern is how many generations the monkey requires to first hit the target. You had various other concerns when you interpreted a modified Weasel program as an implementation of a finite-population approximation to a special case of the Wright-Fisher model. Then it was important to understand that the evolutionary process was nonterminating, even though the simulation of it terminated: there is no possibility of extinction in the Wright-Fisher model.

    0
  13. Freelurker and Flint:

    A simulation outputs (digested) observations of the simulated process. It makes good sense to speak of an algorithm for simulation of a process. However, the precise meaning of simulation is tricky, particularly when the simulated process is random and the simulator is deterministic (pseudo-random). I’m ignoring quite a number of difficult details.

    You can make a Weasel program algorithmic by imposing a limit on the number of generations that it simulates in a run. It’s important to understand that this limit is an attribute of the simulator, not the simuland. The simulator surely terminates in a finite number of steps, though the probability is, for any finite number of generations, less than 1 (unity) that the monkey types a phrase matching the target.

    The best way to implement a simulation generally depends on the observations to be output. Suppose, in the case of the monkey/Shakespeare model, that all you want to know about each generation is how many characters in the “parental” phrase are correct. Then it’s possible to produce a very fast implementation that does not operate on phrases. In each generation, the number of correct characters in the current “parental” phrase is randomly transformed into the number of correct characters in the next “parental” phrase. Hopefully Freelurker will see that the problem here is to get the next number from the preceding number (s/he’s right to object to the notion that hitting the target is the problem), and that there are alternative algorithms for solving it. Endless iteration of an algorithm for (pseudo-)randomly updating the number of correct characters is a definite procedure for realizing the infinite, random sequence of numbers of correct characters. The definite procedure is not an algorithm because it never terminates.

    0
  14. Corneel: I wasn’t talking about Dawkins’ weasel program either but about Eric’s evolutionary algorithm E (which admittedly looks a bit like Dawkins’ weasel). I assumed, naively perhaps, that since it is called “evolutionary algorithm”, that it was actually an .. eh .. algorithm.

    Usually the terms “genetic algorithm” and “evolutionary algorithm” are just misnomers — mildly annoying, but no big deal. However, Eric’s argument requires that the “algorithm” actually be an algorithm.

    EricMH: The alleles are 1s and 0s, and the gene G a bitstring of N bits. A gene’s fitness is based on how many 1s it has, so fitness(G) = sum(G). The population consists of a single gene, and evolution proceeds by randomly flipping one bit, and if fitness is improved, it keeps that gene, otherwise it keeps the original. Once fitness(G) = N, the evolutionary algorithm stops and outputs G, which consists of N 1s. We will denote the evolutionary algorithm E, and it is prefixed on an input bitstring X of length N that will be turned into the bitstring of N 1s, so executing the pair on a universal Turing machine outputs the bitstring of 1s: U(E,X) = Y.

    That’s grossly incoherent. Let n be the number of 1s in the initial bitstring X. If n = N, then the fitness of the initial bitstring is maximal, and the generous interpretation of what Eric has written is that the procedure terminates immediately. Otherwise, the probability that a random bit-flip decreases fitness (changes a 1 bit to a 0 bit) is n / N > 0, and the probability that all of the first k bit-flips decrease fitness is (n / N)^k > 0. Put simply, no matter how many mutants the procedure randomly generates, the probability is nonzero that all of the mutants are less fit than the initial bitstring, and that the current parent G remains identical to the initial bitstring X. Thus there is no guarantee that the procedure will ever generate an output.

    In contrast, the expression U(E, X) = Y indicates that the partial function U, representing a universal computer, maps the program-input pair (E, X) to output Y. The expression indicates no randomness whatsoever. The computer does not have access to a source of random bits. There is no input of a random seed for a pseudorandom number generator that might be embedded in program E. It is ridiculous for Eric to have juxtaposed the deterministic U(E, X) = Y with the preceding description of a random process. This is not a matter of a “hurried genius” having neglected to spell out for his intellectual inferiors the bridge from the description of a random process to the description of a deterministic process. There exists no such bridge.

    0
  15. Tom English: This is not a matter of a “hurried genius” having neglected to spell out for his intellectual inferiors the bridge from the description of a random process to the description of a deterministic process.

    Thanks Tom. I’ve expressed concerns that Eric’s models have no bearing on biology, but I assumed his math was reasonable. My bad for making that assumption..

    0
  16. I hope to put comments here on two issues: Dembski’s argument for conservation of Complex Specified Information, and whether a simple model of natural selection acting on genotypes, one that achieves higher fitnesses, can be shown to be unable to do so unless Design Intervention is involved. Let’s start with the first one.

    In my 2007 article on Dembski’s arguments I gave a simple example to show that while information can be conserved, specified information isn’t. The case was a digital image of a Zinnia flower, the black and white pixels being represented by 1s and 0s. The specification was “looks like a flower”. I applied an arbitrarily-chosen permutation to the bits in the image and got a new image. That definitely did not look much like a flower. If I reversed the permutation, which is easy to do if you have a record of it showing where each bit moved to, you get back the flower image. So the amount of SI could either increase or decrease, depending on the case.

    The permutation is a 1-1 transformation. It conserves information, as you can always reverse the transformation and get back to where you were. But if we judge by the original specification, we get less SI when we apply the permutation, and we get more SI when we undo it.

    Note that in Dembski’s argument, natural evolutionary processes are modeled by a 1-1 transformation, in his simplest case (see No Free Lunch, chapter 3, section 3.8, pages 152 and 153). Dembski uses this to show that CSI is conserved. (That’s like having a law that large enough amounts of matter are conserved). How can that be? We have already seen that SI isn’t conserved under action of a 1-1 transformation. Answer: he changes the specification from before to after. Now his set of specified bitstrings is not those that “look like a flower” but those that, when reverse-permuted, look like a flower. So if the original transformation is called f, the new specification is “when the reverse permutation f^{-1} is applied, looks like a flower”.

    Eric Holloway has argued against this, saying that

    It seems you did not read the rest of the argument about conservation of information. Dembski’s point is *not* that a fitter organism cannot come out of an evolutionary process.Dembski’s point is that if this happens, the CSI exhibited by the organism was contained, but not originated, by the evolutionary process.

    But in the Zinnia example, the transformation was an arbitrarily-chosen one, generated by a pseudorandom number generator making a random permutation. It contained no specified information whatever about being like-a-flower.

    I am not arguing that Dembski’s is wrong about the probability of being in the set of like-a-flower bit images being exactly the same as the probability of being, after the 1-1 transformation is applied, in the when-that-transformation-is-reversed-like-a-flower set of images. That is obviously, and trivially, true. The point is that if we imagine that there is a set of like-a-flower images, and we keep that criterion unchanged, 1-1 transformations can move us either into or out of that set.

    Once we have had some discussion about this, I will comment in a second comment about the simple example of genetic simulation putting Specified Information into the genome.

    0
  17. Corneel: Eric, would you answer my question in the meantime? Doesn’t your pencil factory allow for an non-interventionist scenario? Once, the “evolutionary process” is in place, a population can evolve without guidance from an intelligent agent. That has a distinct theistic evolution feel about it.

    Yes, theistic and deistic evolution proper is compatible with ID. The intelligent agency could set up all the dominoes, and then let them all fall of their own accord, maybe in the meantime setting up other domino chains that end up intersecting.

    The ID PR authors write a lot about the theistic evolution *movement* essentially because it is not truly theistic evolution in the way you outline. They deny that empirical measurements can give us any insight into whether an intelligent agent had a hand in our origin and evolution, which as you pointed out is not really doing justice to the actual theistic evolution position, i.e. the pencil factory.

    It is quite consistent with ID to say that once in operation the pencil factory needs no outside intervention (although in reality entropy is always an issue and the factory requires constant maintenance). However, if we saw a pencil factory in operation, we would certainly infer an intelligent designer and not claim the pencil factory sprang into existence through natural causes. The theistic evolution *movement* wants to say that we cannot know anything about the origin of the pencil factory, and inferring a designer does nothing to help us understand the factory further (although blueprints, an operation manual, and troubleshooting guide would certainly help!). That is why the ID authors write so much about the movement.

    0
  18. Joe Felsenstein: The permutation is a 1-1 transformation. It conserves information, as you can always reverse the transformation and get back to where you were. But if we judge by the original specification, we get less SI when we apply the permutation, and we get more SI when we undo it.

    Thank you, it is much clearer what you are saying.

    As I mentioned before, this appears to be the same counter argument as Dr. English’s offered a number of months ago. He titled the argument “non conservation of ASC’. To properly apply ASC (more mathematically rigorous CSI) to your example, you have to apply your transformation function to the original random variable. In which case, you get a new random variable, and the improbability of ASC obtains.

    If you choose to not transform the random variable, you can increase the ASC arbitrarily high, or even just say infinity, as Dr. English’s example shows. High ASC indicates the source of the image is not the chance hypothesis, and since in the case of your example that is true, the source of the image is some other function, then ASC is operating as expected.

    It is only important that ASC be conserved with regard to the chance hypothesis, so that ASC operates as a hypothesis test.

    0
  19. Tom English: In contrast, the expression U(E, X) = Y indicates that the partial function U, representing a universal computer, maps the program-input pair (E, X) to output Y. The expression indicates no randomness whatsoever. The computer does not have access to a source of random bits. There is no input of a random seed for a pseudorandom number generator that might be embedded in program E. It is ridiculous for Eric to have juxtaposed the deterministic U(E, X) = Y with the preceding description of a random process. This is not a matter of a “hurried genius” having neglected to spell out for his intellectual inferiors the bridge from the description of a random process to the description of a deterministic process. There exists no such bridge.

    Again you seem intent on making a big deal about nothing. Just put a long enough random bitstring into E and there is no need for a stochastic process.

    Alternatively, U can be a probabilistic Turing machine. You just choose to interpret it as a mathematical function in order to avoid debating the actual substance of the argument.

    0
  20. Eric,

    As Joe has told you repeatedly, he is not talking about ASC. He is talking about Specified Information, and he has supplied you with specific references.

    It’s time to stop dodging and address his actual argument, even if it’s not the argument you wish he were making.

    0
  21. EricMH: The theistic evolution *movement* wants to say that we cannot know anything about the origin of the pencil factory, and inferring a designer does nothing to help us understand the factory further. That is why the ID authors write so much about the movement.

    There isn’t a ‘theist evolution *movement*.’ Some IDists are trying to make theistic evolution & evolutionary creation into their ‘opponent.’ Theistic evolution is the default and vast majority Abrahamic monotheistic position. IDism is the ideology of the ‘Intelligent Design Movement,’ which is a phrase that IDists at the DI have used about themselves.

    The interpretation of ‘theistic evolution’ as “wants to say that we cannot know anything about the origin of the pencil factory” is inverted. IDists cannot say anything about the when, where, how or who of the supposed ‘design instantiation’ or study *anything* about the actualized ‘designing process.’ This is the required sacrifice IDists make with their semantic sleight of hand.

    0
  22. EricMH:

    However, if we saw a pencil factory in operation, we would certainly infer an intelligent designer and not claim the pencil factory sprang into existence through natural causes.

    Yes, we conclude that a pencil factory is designed, but not because it is complex (which it is), and not because it is specified (which it is, too).

    We conclude this because of the obvious fact that the factory was built, not born.

    This is such a simple, basic concept that somehow always eludes the ID proponents. All those tired analogies about machines on Mars etc. fail on this simple point: we already know that if something with CSI isn’t the outcome of biological reproduction it isn’t a product of evolution, but of manufacturing, i.e. design. Well, duh!

    These analogies fail to provide any more insights into biology (and that is what we are talking about here, isn’t it?) than we all share already.

    0
  23. EricMH: As I mentioned before, this appears to be the same counter argument as Dr. English’s offered a number of months ago.

    To someone like me who is trying to follow the logic of the arguments, it would be very helpful if you expanded on this sentence, in something like the following way:
    1. Give your understanding of Joe’s argument in your own words.
    2. Show how CSI used in this argument maps to ASC. Or link to a paper which shows this mapping, being explicit about what section of the paper provides the required mapping and why it does so.
    3. Show how Joe’s counter-example maps to an argument about ASC as defined in 2.
    4. Refute that argument in 3 more explicitly.

    I understand you likely had some version of 1-4 in mind when you posted your reply. I think making those thoughts explicit is a necessary part of engaging fully in the debate.

    In any event, as far as I can tell this argument is about the mathematical coherence of CSI. It does not have any bearing on whether that math can be used to build a successful biological model.

    0
  24. EricMH: Back to the pencil factory, imagine a pencil factory factory. The pencil factory itself is complex, but a factory to make the factory would be absolutely mind boggling complex.

    FWIW, the analyses of self-replicating machines goes back at least to von Neumann, who was also no slacker when it came to scientifically useful concepts of information (quantum versions of entropy bear his name and he was present at the creation of Shannon “entropy”).

    This book discuss such machines. Chapter 4 addresses DNA replication as an example of a self-replicating machine.

    My text search of the book did not find any hits for the phrase “mind boggling”. Or any kind of boggling for that matter.
    https://www.amazon.com/Kinematic-Self-Replicating-Machines-Robert-Freitas/dp/1570596905/

    0
  25. keiths: andomness tests alone don’t resolve the paradox, because there are patterns — such as one’s SSN — that might qualify as random but nevertheless have special significance to u

    Have you seen this paper? The Miraculaour Universal Distributino. It does not speak directly to the personally-interesting SSN case, but it does cover the all heads case.

    It addresses some topics that Eric name drops in the OP.

    ETA: The authors of the paper wrote the book, so to speak, on Kolmogorov Complexity. I think I got the paper from a citation in one of Shallit’s CSI papers.
    https://www.amazon.com/Introduction-Kolmogorov-Complexity-Applications-Computer/dp/0387339981

    0
  26. keiths: He is talking about Specified Information, and he has supplied you with specific references.

    The same argument applies to CSI, although in this case CSI needs to be normalized via Dr. Montañez concept of a kardis. So, same basic idea, it’s just easier to discuss with ASC b/c the normalization is baked in via the Kraft-McMillan inequality.

    Is there a specific problem you see in my explanation?

    0
  27. faded_Glory: We conclude this because of the obvious fact that the factory was built, not born.

    You are just trying to confound the example. The point of the pencil factory is that it shows we need even more information to create the factory than the pencil.

    I then used it as an explanatory device to address your point and explain the relationship between ID and theistic evolution.

    This has nothing to do with the explanatory filter.

    0
  28. BruceS: The authors of the paper wrote the book, so to speak, on Kolmogorov Complexity.

    Yes, the universal distribution is pretty neat, and is key for Levin’s proof. Li and Vitanyi’s book has taught me a lot, but they also are committed to naturalism, and so they avoid the full extent of Levin’s proof by an offhanded remark that it is ‘too complicated.’ Lol! pot meet kettle!

    0
  29. BruceS: To someone like me who is trying to follow the logic of the arguments, it would be very helpful if you expanded on this sentence, in something like the following way:
    1. Give your understanding of Joe’s argument in your own words.
    2. Show how CSI used in this argument maps to ASC. Or link to a paper which shows this mapping, being explicit about what section of the paper provides the required mapping and why it does so.
    3. Show how Joe’s counter-example maps to an argument about ASC as defined in 2.
    4. Refute that argument in 3 more explicitly.

    Alright, I’m assuming the person making the argument is familiar with standard ID literature, but I’ve been neglecting the rest of the audience. My apologies.

    First, a brief overview of CSI. I’ll assume you are generally familiar with the concept, so won’t go into too much detail.

    CSI measures whether an event X is best explained by a chance hypothesis C, or some specification S. Consequently, it is measured by the difference of two parts: complexity and specification. Complexity is the negative log of the event’s probability according to a chance hypothesis: -log P(X|C). Specification is the number of descriptions (given some description language common to intelligent agents) that are at least as concise as the description of the event: S(X).

    So, CSI(X) = -log P(X|C) – S(X). If this value is greater than 1, then we infer X is most likely the product of an intelligent agent.

    We can then imagine a set of various Xs, some with positive CSI and some without.

    The question is if we take some X such that CSI(X) 0, what does this imply about f? Does f give us CSI for free, or does the transformation require that f itself contain CSI?

    Dembski conservation of information argues that for this transformation to take place f must itself contain CSI. There is never a way to get CSI for free. If CSI(X) 0, then this implies CSI(f) > 0. Further, he argues this is a tight requirement, so that CSI(f) >= CSI(f(X)) – CSI(X). Finally, he argues that randomizing f doesn’t get around this problem. While F as a random variable may range over some f in F with positive CSI, the random variable F itself must contain positive CSI if we are to expect positive CSI by randomly sampling and applying f(X). In other words, it must be the case that CSI(F) >= E[CSI(f(X))] .

    Dr. Felsenstein’s counter argument is that we can take some randomly generated function f and apply it to an X such that CSI(X) > CSI(f(X)). This doesn’t invalidate Dembski’s argument. However, Dr. Felsenstein then claims we can invert f to get f-1, such that f-1(f(X))=X, so we get the situation that CSI(f(X)) = a] <= 2^-a. The key element that makes it work is the prefix-free Kolmogorov complexity, which gives us the Kraft-McMillan inequality. This general result is fairly well known, and has been proven a number of times by various mathematicians before ASC, including Leonid Levin. The innovation of ASC is the inclusion of a context with conditional complexity, and Dr. Ewert's proof is also novel (as far as I know).

    Now in my last response to Dr. Felsenstein, I wanted to address the issue more generally, as Dr. English did, when he proposed his 'non conservation of ASC' theorem, as Dr. Felsenstein's sort of counter argument does not succeed if Dr. English's argument is flawed.

    Dr. English showed that we can take some function and apply it to X to get a Y such that Y is not in the random variable that generated X. Consequently, the probability for Y is zero, and the ASC is infinite:

    ASC(Y,P,C) = -log2 P(Y) – K(Y|C) = -log2 0 – K(Y|C) = oo – K(Y|C) = oo.

    This seems to show that ASC is not conserved. Note this argument is also similar to Dr. Felsenstein's, although Dr. Felsenstein was attacking the specification component, and Dr. English is attacking the complexity/surprisal component.

    The key to refuting both sorts of arguments on a formal basis is to recall what ASC measures. ASC is essentially measuring the probability that X was generated from distribution P. If ASC is positive, especially if very positive, then most likely X came from a different distribution than P. Consequently, if we are transforming X with f to get Y, we don't expect ASC to be conserved, since Y did not come from P. So, the fact that Y can have infinite ASC shows ASC is operating correctly, since a Y with infinite ASC definitively shows Y did not come from P.

    On the other hand, if we transform the random variable that X is sampled from with f, it would be a problem if ASC is not conserved, since we are essentially proposing a new distribution (call it Q) to generate Y. But, since the improbability of ASC result applies to all random variables, then it will apply to this new transformed random variable as well. Hence, ASC is conserved under function transformation of the original random variable, whether the function is deterministic or stochastic.

    Hopefully this is enough detail, BruceS.

    0
  30. BruceS: My text search of the book did not find any hits for the phrase “mind boggling”. Or any kind of boggling for that matter.

    Well, did they manage to invent a self replicating machine? Has anyone? No. It is beyond our current engineering skill, hence it has boggled our collective minds.

    The Von Neumann architecture, which all our personal computers use, was invented in an attempt to create a self replicating machine. In fact, much of our computer science theory (e.g. Turing) was originated due to naturalists trying to derive a naturalistic origin of life and mind. They’ve all failed, but at least now we have Microsoft’s Clippy.

    0
  31. EricMH,

    Hi Eric. Question that’s been asked a few times, but never answered. If Darwinism / Materialism is so bad for science, why haven’t you IDists come up with something better?

    0
  32. EricMH:

    CSI measures whether an event X is best explained by a chance hypothesis C, or some specification S.Consequently, it is measured by the difference of two parts: complexity and specification.Complexity is the negative log of the event’s probability according to a chance hypothesis: -log P(X|C).Specification is the number of descriptions (given some description language common to intelligent agents) that are at least as concise as the description of the event: S(X).

    So, CSI(X) = -log P(X|C) – S(X).If this value is greater than 1, then we infer X is most likely the product of an intelligent agent.

    We can then imagine a set of various Xs, some with positive CSI and some without.

    When are you going to stop imagining a set of various Xs and actually measure some real ones? I would be interested in knowing the CSI of, say, a bacterial flagellum, a pencil factory, and the Emperor’s new clothes.

    Can you do that? If not, why not?

    0
  33. faded_Glory, to Eric:

    When are you going to stop imagining a set of various Xs and actually measure some real ones? I would be interested in knowing the CSI of, say, a bacterial flagellum, a pencil factory, and the Emperor’s new clothes.

    Can you do that? If not, why not?

    I posted this earlier in the thread:

    Eric,

    In No Free Lunch, Dembski attempted to demonstrate that the flagellum is designed.

    He failed. Do you agree, and do you understand why?

    No response from Eric.

    0
  34. keiths, to Eric:

    As Joe has told you repeatedly, he is not talking about ASC. He is talking about Specified Information, and he has supplied you with specific references.

    It’s time to stop dodging and address his actual argument, even if it’s not the argument you wish he were making.

    Eric:

    The same argument applies to CSI, although in this case CSI needs to be normalized via Dr. Montañez concept of a kardis.

    No, it doesn’t. Dembski didn’t normalize it, and it isn’t needed. CSI is just any quantity of specified information greater than 500 bits, as defined by Dembski.

    Is there a specific problem you see in my explanation?

    Yes. The (repeated) problem is that you are dodging Joe’s argument, which doesn’t invoke (or depend on) ASC or Montañez or “normalization by kardis”. Why not respond straightforwardly to Joe’s actual argument as he actually expressed it? Obfuscation is neither necessary nor helpful.

    0
  35. keiths: The (repeated) problem is that you are dodging Joe’s argument, which doesn’t invoke (or depend on) ASC or Montañ

    I don’t think there is a problem in principle with taking a mathematical argument and transferring it to a logically equivalent argument based in different domain, and then refuting the argument as expressed in that different domain.

    But you do have to show that the argument and its refutation are logically equivalent across domains.

    In this case, my idea was that perhaps Eric was claiming Joe’s argument against CSI could be transferred to the same argument against ASC, and then refuted there, which would also serve to refute Joe’s argument against CSI, assuming all the logically equivalences I mentioned were made explicit and then justified.

    My concern was that Eric did not show the work needed to justify a process like the above.

    What do you think of the details of Eric’s extended response to me in terms of meeting those objectives for inter-domain arguments? You have worked much more with Id math ideas than I have.

    0
  36. EricMH: Hopefully this is enough detail, BruceS.

    Thanks for taking the time to do this long response on Joe’s argument. I will work through and post if I have further concerns.

    On self-replicating: I agree engineering the ideas is hard for current technology. But using those current difficulties as a kind of gaps argument is not a justification for some philosophical version of immaterialism.

    (FWIW, the only gaps argument that has any philosophical merit are the various permutations of Chalmers Zombie arguments against physicalism, although they fail as well, I think).

    0
  37. Some comments on Eric’s response to my argument about the nonexistence of a Law of Conservation of Complex Specified Information.

    EricMH: Alright, I’m assuming the person making the argument is familiar with standard ID literature, but I’ve been neglecting the rest of the audience. My apologies.

    First, a brief overview of CSI. I’ll assume you are generally familiar with the concept, so won’t go into too much detail.

    CSI measures whether an event X is best explained by a chance hypothesis C, or some specification S.

    Stop right there. The specification does not “explain” the event X. In the case I used the specification was “like a flower”. It was not a theory of how the flower came to be, or how the image of a flower came to be.

    Consequently, it is measured by the difference of two parts: complexity and specification.

    Same issue. I’d say instead that SI(X) is a measure of whether the event X and all events equally or better-specified is improbably small.

    Complexity is the negative log of the event’s probability according to a chance hypothesis: -log P(X|C).

    No, the negative log of the probability of all events that are that well-specified or more.

    Specification is the number of descriptions (given some description language common to intelligent agents) that are at least as concise as the description of the event: S(X).

    Not in the development by William Dembski, in either his 1996 book The Design Inference: Eliminating Chance through Small Probabilities or his 2002 book No Free Lunch: Why Specified Complexity Cannot Be Purchased Without Intelligence. In both of them Dembski mentions Algorithmic Complexity arguments, but then sets them aside as not quite the right way to use information to evaluate evolution.

    Instead Dembski (on page 148 of NFL) says that

    The specification of organisms can be crashed out in any number of ways. Arno Wouters cashes it out globally in terms of the viability of whole organisms. Michael Behe cashes it out in terms of minimal function of biochemical systems. Even the staunch Darwinist Richard Dawkins will admit that life is specified functionally, cashing out the functionality of organisms in terms of reproduction of genes. Thus Dawkins (1987, p. 9) will write: ‘Complicated things have some quality, specifiable in advance, that is highly unlikely to have been acquired by random chance alone. In the case of living things, the quality that is specified in advance is . . . the ability to propagate genes in reproduction.”

    All three of these examples are components of fitness, or traits positively correlated with fitness. In later writings, Dembski says that Algorithmic Specified Complexity is one of the kinds of specification consistent with his argument. But in 2002 it is not a part of Dembski’s argument.

    [Holloway] So, CSI(X) = -log P(X|C) – S(X). If this value is greater than 1, then we infer X is most likely the product of an intelligent agent.

    We can then imagine a set of various Xs, some with positive CSI and some without.

    Presumably such as my examples of the digital image of a flower and the scrambled version of that image.

    [Holloway] The question is if we take some X such that CSI(X)0 [Typo: I think a > is missing here, JF], what does this imply about f? Does f give us CSI for free, or does the transformation require that f itself contain CSI?

    Dembski conservation of information argues that for this transformation to take place f must itself contain CSI. There is never a way to get CSI for free.If CSI(X)0 [same Typo, JF], then this implies CSI(f) > 0. Further, he argues this is a tight requirement, so that CSI(f) >= CSI(f(X)) – CSI(X). Finally, he argues that randomizing f doesn’t get around this problem.
    While F as a random variable may range over some f in F with positive CSI, the random variable F itself must contain positive CSI if we are to expect positive CSI by randomly sampling and applying f(X). In other words, it must be the case that CSI(F) >= E[CSI(f(X))] .

    Dr. Felsenstein’s counter argument is that we can take some randomly generated function f and apply it to an X such that CSI(X) > CSI(f(X)). This doesn’t invalidate Dembski’s argument.

    However, Dr. Felsenstein then claims we can invert f to get f-1, such that f-1(f(X))=X, so we get the situation that CSI(f(X)) = a] <= 2^-a.

    No, my example, was a case where we apply a randomly generated permutation f, and get an image that no longer has CSI. I did not use ASC arguments or the expression Holloway gives.

    In my example, the function f self-evidently did not contain any information about the flowerness of the image. Nor would the inverse of that function. And I am not “claiming” that we can invert the function. We can. It is dead easy. If the function takes bit i in the image and maps it into bit j, then its inverse takes bit j and maps it into bit i. And so on.

    And then we’re off into Kolmogorov-land, a fine place to be under other circumstances, but not part of Dembski’s 2002 argument.

    [Holloway] The key element that makes it work is the prefix-free Kolmogorov complexity, which gives us the Kraft-McMillan inequality.

    … which is about as relevant to Dembski’s argument as MacMillan books or Kraft Cracker-Barrel Cheese.

    I omit the rest of Holloway’s argument with me and with Tom English as irrelevant to my example. When this part of the argument is over, I hope to publish at Panda’s Thumb a more extended argument as to the irrelevancy of ASC arguments to rejecting natural selection in favor of design. It will involve arguing that shortness of description has nothing to do with any component of fitness.

    It is important to look into Holloway’s argument about the validity of Dembski’s argument, because Holloway is going around announcing to the world that Dembski’s basic argument is not refuted. Actually it has been demolished.

    0
  38. A silly typo. I wrote:

    Same issue. I’d say instead that SI(X) is a measure of whether the event X and all events equally or better-specified is improbably small.

    when of course it should have been:

    Same issue. I’d say instead that SI(X) is a measure of whether the probability of the event X and all events equally or better-specified is improbably small.

    Smallness of the event not being the issue.

    0
  39. Joe Felsenstein: Some comments on Eric’s response to my argument

    Joe: Thanks for taking the time to respond to Eric’s points. I am not sure, however, if you responded to the argument he is making. In particular, I am not concerned with whether he is accurately representing some version of Dembski’s CSI. I am trying to understand the argument as presented solely in his post.

    As I said in my post to Keith, I read Eric as attempting to refute your argument by transferring it in a logically equivalent way to a different math domain and refuting the version in that domain. In particular, I see these three excepts from his post as key to the structure of the argument

    Quoting Eric:

    Dr. Felsenstein’s counter argument is that we can take some randomly generated function f and apply it to an X such that CSI(X) > CSI(f(X)).
    […]
    The key element that makes it work is the prefix-free Kolmogorov complexity, which gives us the Kraft-McMillan inequality. […]
    […]
    Dr. Felsenstein’s sort of counter argument does not succeed if Dr. English’s argument is flawed.

    I don’t think the first statement is correct, as your counter-argument is based on a single specified counter-example, rather than a process of random selection. But setting that issue aside, I read the other statements as saying that if your argument involves random selection of f, then your argument depends on a result from Kolmogorov Complexity theory, and refuting Dr. English’s counter to the KC used by ASC also refutes the argument Eric attributes to you.

    I’ll ask Eric separately for his feedback on that construal of the structure of his argument.

    I don’t have the math skills to evaluate Eric’s approach or the KC math on which it depends. But I think I understand why he believes KC and hence ASC is relevant to responding to your argument.

    On your upcoming post on ASC and fitness: I will look forward to it. I do see it, however, as addressing the usefulness of ASC for doing science, not as addressing the pure mathematics of Eric’s arguments. I’ve always thought that biological usefulness was much more important than the mathematics on its own; that’s my rationalism versus empiricism spiel.

    0
  40. EricMH: Hopefully this is enough detail, BruceS

    Eric: Is my construal of the structure of your argument fair (see my preceding post).

    What do you think of my concern that you characterize Joe’s f as being randomly selected whereas he is focusing on a single, carefully constructed counter example? (And not, eg, the properties of some distribution of f’s).

    0
  41. Can someone explain to me what the relevance of all this is, when nobody has ever measured or calculated the CSI of anything of biological importance?

    What good is a measurement that never leaves the realm of hypotheticals? What conclusions can we ever hope to draw from a thing like that?

    0
  42. Richardthughes: If Darwinism / Materialism is so bad for science, why haven’t you IDists come up with something better?

    Dr. Ewert’s dependency graph of life is pretty good. Denton has shown there is a Platonic structure to species. Also, the concepts of ID are in current use. Dr. English has highlighted Milosavljević’s work with the algorithmic significance metric, which is very similar to ASC. The biggest ID result is the ENCODE project showing most of the genome is functional, contrary to the prediction of Darwinism.

    It doesn’t seem accurate to say that ID has not come up with something better. Additionally, what has Darwinism/Materialism provided that is of value? Everything valuable in modern biology depends on computational and information theory constructs, which are the result of an ID consistent paradigm, not a D/M paradigm.

    0
  43. Joe Felsenstein: In my example, the function f self-evidently did not contain any information about the flowerness of the image. Nor would the inverse of that function. And I am not “claiming” that we can invert the function. We can. It is dead easy. If the function takes bit i in the image and maps it into bit j, then its inverse takes bit j and maps it into bit i. And so on.

    Well, we seem to be talking past each other. You are essentially saying the same thing I did, and I pointed out the information about high CSI segments is contained in the inversion, which is pretty obvious. But, if you think you’ve refuted Dembski’s argument still, not sure what else I can say. More power to you.

    At this point, we’ve covered the same ground a couple times. I’ve explained to you why your argument doesn’t work. You continue to insist it does. I guess we can go our separate ways now.

    0
  44. BruceS: Eric: Is my construal of the structure of your argument fair (see my preceding post).

    What do you think of my concern that you characterize Joe’s f as being randomly selected whereas he is focusing on a single, carefully constructed counter example? (And not, eg, the properties of some distribution of f’s).

    Yes, you uniquely seem to be a pretty fair minded commentator on this site, and once again you fairly outline my argument. My thanks to you.

    Dr. Felsenstein’s starting point (flower image) is carefully constructed. The random scrambling he mentions is the f being chosen at random.

    0
  45. EricMH: Dr. Ewert’s dependency graph of life is pretty good. (…) The biggest ID result is the ENCODE project showing most of the genome is functional, contrary to the prediction of Darwinism.

    Wow, you have absolutely no idea what you’re talking about. You’re just spewing nonsense. Holy crap.

    0
  46. EricMH,

    EricMH: The biggest ID result is the ENCODE project showing most of the genome is functional, contrary to the prediction of Darwinism.

    Yeah, I’m going to pile on here and rubbish that sentence too. Junk DNA is not a prediction of ‘Darwinism’, and ENCODE has not falsified it.

    0
  47. BruceS: I don’t think the first statement is correct, as your counter-argument is based on a single specified counter-example, rather than a process of random selection.

    The example is a typical image from the set of specified (like-a-flower) images. The permutation f is not carefully constructed, but is just the first permutation I could make with the pseudorandom number generator I had. Almost any other permutation would have done the same — make the image no longer satisfy the specification like-a-flower. All those permutations are each easily inverted to go back from the scrambled image.

    But setting that issue aside, I read the other statements as saying that if your argument involves random selection of f, then your argument depends on a result from Kolmogorov Complexity theory, and refuting Dr. English’s counter to the KC used by ASC also refutes the argument Eric attributes to you.

    I am not sure which of us, me or Eric, made those statements about my argument depending on a result from KC theory. I did not, as far as I know. I cannot see how anything in KC is relevant to refuting my counterexample.

    I don’t have the math skills to evaluate Eric’s approach or the KC math on which it depends. But I think I understand why he believes KC and hence ASC is relevant to responding to your argument.

    I am glad to hear that you understand that. I certainly have no clue about why he can argue that.

    On your upcoming post on ASC and fitness: I will look forward to it. I do see it, however, as addressing the usefulness of ASC for doing science, not as addressing the pure mathematics of Eric’s arguments. I’ve always thought that biological usefulness was much more important than the mathematics on its own; that’s my rationalism versus empiricism spiel.

    I have worked my whole career on the use of theory in biology. Immediate usefulness is important, but not essential. For example, the Hardy-Weinberg calculations in population genetics are useful, but they also show that during the process of reproduction with random mating (and no other evolutionary processes), gene frequency does not change. Knowing that is helpful in evaluating claims that genetic variation will rapidly disappear from a population.

    0
  48. faded_Glory:
    Can someone explain to me what the relevance of all this is, when nobody has ever measured or calculated the CSI of anything of biological importance?

    What good is a measurement that never leaves the realm of hypotheticals? What conclusions can we ever hope to draw from a thing like that?

    Specified information (and even use of “complexity” in the discussion) was not invented by William Dembski et al., though they certainly helped people notice it. It was invented by Leslie Orgel (he of RNA World fame). A closely related concept, Functional Information, was invented by Jack Szostak and Robert Hazen. (A less clear version was also discussed by me in 1978).

    They were trying to make an information-theoretic calculation of how much biological information there was in well-adapted systems. Hazen, Carothers, and Szostak used it in analyzing data on adaptations of RNA molecules in in-vitro experiments involving replicating them.

    Basically, in Dembski’s argument, a calculation that there is CSI is an information-tinged stand-in for just saying that the genotype has achieved so high a level of adaptation that a monkey typing a genome on a 4-letter typewriter could never do as well even once in the history of the universe. All the stuff involving taking logarithms is informationish window-dressing.

    0

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.