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:

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

EricMH,

Randomness 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 us.

What I have yet to see is application to real world data. For example, apply these concepts to bacterial populations, or to comparisons of mammalian genomes. Show us how CSI is measured, and show us how it is able to differentiate between naturally occurring mutations and designed mutations. Until this is done, it is all meaningless.

keiths, to Eric:

A further problem is that Dembski misapplies specification, as I explained here. He also assumes “a dictionary of 100,000 basic concepts” while giving no justification for that particular number.

There’s some egregious equivocation in this thread, beginning with the OP.

Dembski, joined later by Marks and Ewert, has published three different definitions of specified complexity. He has stated a “conservation law” only for CSI as originally defined — the log-improbability, log(1/p), of an event with a detachable specification. In his second and third definitions, CSI is a log

ratioof nonconstant quantities. His semiformal arguments about conservation of [complex specified] information (log-improbability of an event with a detachable specification), given in his bookNo Free Lunch: Why Specified Complexity Cannot Be Purchased without Intelligence(2002), do not apply to the later forms of specified complexity. And formal results that have been derived for the later forms of specified complexity do not apply to the original form. Thus EricMH is grossly wrong when he makes sweeping claims about “Dembski’s ID theory.”Dembski said nothing about conservation of the second “CSI” that he defined (2005). That “CSI” was recently dubbed

semiotic specified complexityby George Montañez (who, unlike EricMH, attends to essential mathematical details).In 2009 (?), Dembski and Marks restated the Law of Conservation of Information (LCI), with “information” referring to active information instead of complex specified information (specified complexity). They gave no indication at all that Dembski had previously claimed that his original form of specified complexity was conserved. In the years since, they have said nothing whatsoever about Dembski’s original LCI. That is, they have never claimed that Dembski came up with two laws, one for specified complexity, and the other for active information. They simply abandoned Dembski’s original claim that specified complexity was conserved, and transferred his “conservation of information” rhetoric to active information.

Ewert, Dembski, and Marks introduced algorithmic specified complexity (ASC, not CSI) in 2012. It is the only one of Dembski’s three forms of specified complexity that they address in their book,

Introduction to Evolutionary Informatics(where the “information” in “conservation of information” is strictly active information). Rather than say that ASC is conserved, they say that that high ASC is improbable. Furthermore, they bill ASC as a measure of meaningful information. (In my most recent post, I proved thatASC is not conservedin the sense that algorithmic mutual information is conserved. I also demonstrated, with a striking pictorial example, that it is ludicrous to regard ASC as a measure of meaning.)None of EricMH’s “Things like CSI” is like Dembski’s original CSI — the only CSI that he ever said was conserved. One of them, the Martin Löf test for randomness, is not like any of the forms of specified complexity. The third form, ASC, is somewhat like classical mutual information and algorithmic mutual information for the simple reason that its formal expression is a mixture of expressions from the two. Observing the formal similarity, and leaving it at that, is utterly vacuous. More interesting — I first noticed this about five years ago, but have never figured out what to make of it — is the observation that ASC,

-log2 p(x) – K(x|c),

where x and c are binary strings, is a generalization of randomness deficiency. That is, set c to the empty string, and ASC

isthe randomness deficiency of x relative to probability distribution p over binary strings. This is not good news for ID theory, because it leads to an ugly question:Why should a generalized measure of randomness deficiency be regarded as a measure of meaning?Ewert, Dembski, and Marks have not raised the question, let alone attempted to answer it. (To repeat, I made it abundantly clear in my most recent post that ASC is not a measure of meaningful information.)George Montañez has generalized the “high ASC is rare” theorem of Ewert, Dembski, and Marks, obtaining a theorem that applies not only to ASC, but also to semiotic specified complexity (and Szostak’s “functional information”). Montañez has failed to resist the urge to call his theorem a “conservation of information” theorem. (He does not explain why he calls it a “conservation of information theorem” — he just does it.) However,

Montañez’s theorem does not apply to Dembski’s original form of specified complexity.EricMH wants us to make something of the “conservation of information” theorem for algorithmic mutual information, so I repeat that I proved, in the most recent of my opening posts at TSZ, non-conservation of algorithmic specified information.

Finally, I should mention that EricMH is not the only one to use “CSI” equivocally in this thread. Some participants who are ostensibly opposed to ID have been aiding and abetting EricMH in his revision of the history of ID.

In the preceding comment, I accidentally entered “algorithmic specified information” where I meant “algorithmic specified complexity.”

Tom:

At UD in 2009, I criticized Dembski’s use of the phrase “Law of Conservation of Information”. Montañez defended it for some odd reasons.

Here’s the exchange:

keiths:

Montañez:

keiths:

Montañez:

keiths:

keiths,1. All three of the “conservation of information” theorems Dembski and Marks had at the time (in “Life’s Conservation Law”) follow easily by Markov’s inequality. But Dembski and Marks have never mentioned Markov’s inequality. I suspect that they did not want to reveal the little trick that they were playing.

2. With obfuscation of Markov’s inequality,

you can get

with substitutions and Taking random variable to be the probability of success of a randomly selected “search,” George would regard the obfuscated Markov’s inequality as saying that active information is conserved in the search for a search. However, we might as well take it as saying that high active information is improbable in a random search for a search. The “high specified complexity is improbable” theorem is quite similar in form to the “high active information is improbable” theorem. Indeed, George applies Markov’s inequality in the proof. My guess is that his reason for calling his “high specified complexity is improbable” theorem a “conservation of information” theorem is that it has the look and the feel of a “high active information is improbable” theorem that has already been called a “conservation of information” theorem.

3. I suspect that almost all physicists would scoff at calling Markov’s inequality, with random variable taking a probability as its value, conservation of information — particularly when there is already a principle of conservation of information (an equality) in quantum mechanics (never mentioned by IDists). I suspect that Dembski and Marks knew better than to make it clear what they were doing.

4. Joe Felsenstein gets the credit for recognizing Markov’s inequality.

Tom,

I agree. Calling it “conservation of information” is inaccurate and gives it an unwarranted air of profundity. It’s pure marketing.

He may be following a precedent, but it’s a precedent that he defended (for questionable reasons including “it has a nice ring to it”) when Dembski and Marks first proposed it.

Yes, that’s one of the innovations of CSI to include context to account for such things. Randomness deficiency is a form of CSI, but CSI itself is a broader concept.

For example, one could have an incompressible bitstring (i.e. l(X) <= K(X)), so no randomness deficiency, that yet has high CSI. If the human mind were a halting oracle, then much of what it creates would be this sort of incompressible CSI.

Well, then you have a Fields Medal to win. Much mathematical work has been based on the premise that all heads is in some sense objectively less random than equal heads and tails, such that all heads is good reason to suspect a non random source, if such alternative hypotheses are available.

This is false. We can never be less capable than algorithms, only slower.

The rest of your argument is based on whether humans are perfect halting oracles like Turing described, but I’ve already explained this is not necessary for my argument. All that is necessary is that the human mind set is not computable, and there is a whole lot of uncomputable space between what algorithms can do and what perfect halting oracles can do. In fact, since there is so much (infinite!) space, it would seem more appropriate to place the burden of proof on those claiming the human mind is reducible to a Turing machine.

The specification can be changed by the stochastic process or stay the same, the conservation of information still applies. I don’t understand the problem you think this poses for Dembski’s COI. Maybe if you can include a very clear example of the problem in your explanation I’ll get it.

I think we are talking cross ways. My only point in that lengthy paragraph is to explain why it is not enough to say “improbable.” Specification is an essential part of the argument.

The point is not whether humans operate exactly like a hypothetical halting oracle, but whether such a model is necessary to simulate the behaviors that the human mind exhibits.

Just like no one is going to program a physics simulation with a literal Turing machine, but it is still true that a Turing complete language is adequate to simulate anything we program.

Partly correct. While it is true there are objective ways to formalize inherently meaningful patterns, that is not necessary for the CSI argument. All that is necessary is the knowledge base used for specification is independent of the process generating the event under analysis.

CSI only guarantees true positives. So, it cannot say X is random. All it can say is that X is not random.

Not so much specify, but assume, the expected distribution. For example, Kolmogorov defined compressibility as non-random because most bitstrings are incompressible. This, of course, assumes a uniform distribution over bitstrings, or something equivalent. Every concept I refer to like this does something similar. They assume there is nothing a priori special about the prior distribution that will favor some particular outcome, and so if these special events show up they at least disqualify the “everything is equally likely” null hypothesis. So, the null hypothesis and test go hand in hand with these concepts.

Sorry, we’ll have to agree to disagree and end it here. We are just going in circles at this point.

You are wrong. Montañez recent paper shows how it’s all related. Essentially, just normalize the specifications that add up to more than one so you get a well behaved probability, and you get the improbability result, which in turn can easily be translated into the form of Dembski’s original COI argument in his NFLT book.

You are constantly harping on minor and inconsequential issues. As for your non-conservation article, I explained to you in that thread why your proof doesn’t apply to ASC, because you are not transforming the random variable with your function, as you should if you want it to apply to ASC.

It’d be great if you actually came up with something that’s fundamentally wrong with ID theory. You are an extremely talented mathematician and it’d be very interesting to see you come up with such a result. Otherwise, it seems you are more interested in making a lot of noise about nothing, and you can probably spend your time more productively elsewhere.

And with that I believe I’ve done my due diligence in replying to everyone’s comments. I didn’t really get much out of this interaction, just many recycled old arguments and poorly understood ID theory, and it’s taken up 2h of my scant free time. So, I’ll be taking a hiatus from this site, too.

EricMH,

It may be satisfying for you to grab your ball and go home, declaring that your opponents don’t know what they’re talking about and aren’t worth your time. But it isn’t true, and it just makes you look bratty.

EricMH,

Youmay be going in circles, but I’m raising a point which you have yet to address: namely, Dembski’s 2005 revision of CSI so that it takes “Darwinian and other material mechanisms into account”. That’s an enormous concession that erases any potential usefulness that CSI might otherwise have had.I’ve also pointed out that Dembski’s use of specification is invalid:

No response from you. I understand. What could you possibly say?

According to the abstract of Montañez’s paper,

What Montañez calls semiotic specified complexity does

notappear inNo Free Lunch(2002). He correctly cites “Specification: The Pattern That Signifies Intelligence” (2005). That was where Dembski first made specification a matter of degree.In

No Free Lunch, an event categorically does or does not have a detachable specification. There is not, as in semiotic specified complexity and algorithmic specified complexity, a numerical measure of descriptive complexity. So the normalization you describe makes no sense at all in the context ofNo Free Lunch.When Judge Jones ruled that ID was not science, in Kitzmiller v. Dover Area School District (2005), the most sensational claim of ID was that design detection was underwritten by Dembski’s Law of Conservation of [Complex Specified] Information — a law of nature, comparable to the Second Law of Thermodynamics. William Dembski, then touted as the “Isaac Newton of information theory,” abandoned his putative law in, IIRC, December 2008, when he and Marks released a preprint of “Life’s Conservation Law.”

Eric,

In

No Free Lunch, Dembski attempted to demonstrate that the flagellum is designed.He failed. Do you agree, and do you understand why?

In biology, specification actually makes certain configurations

moreprobable, not less. This is again because of viability (and in a broader sense, natural selection for fitness) coupled with the fact that progeny will generally be genetically and morphologically very close to its ancestors – that is the way biology works.Natural selection is a very strong filter that biases the probability distribution hugely towards certain specific and non-random outcomes. Unless you build that somehow into your mathematics, you aren’t talking about biology.

Intuitively that premise appears valid. A series of only heads is far less robust to errors than a sequence of equal heads and tails. Flip one tail and it is game over for the all-heads outcome, whereas a string of tails can still be offset by a string of heads later in the flipping to still result in an equal proportion at the conclusion of the experiment.

Your question is, when is an outcome non-random enough to warrant the inference to a different source than blind coin flipping? If there is a filter that removes the tails before you even get to see them, you will end up with an all-heads outcome even if the flips themselves have equiprobable outcomes. It is this combination that permits complex specified outcomes. I don’t see why we would need to conclude that such a process constitutes ‘design’, i.e. conscious intent.

All that this ID mathbabble does is validate Darwin’s conclusion that natural selection can be the explanation for complex biological entities, circumventing the barrier of extremely low probabiities that exist under an assumption of equi-probability. You don’t need to convince us, we have understood this for over 150 years.

If there is anything I hate in these discussions it is the use of the word ‘random’ without further qualification. Random can mean a lot of very different things – stochastic, equiprobable, without aim or intent, haphazard, and so on.

In what sense do you use the word above?

Since evolution clearly is non-random (‘random’ in the sense of having equiprobable outcomes), and nobody suggests that it is otherwise, I don’t see the relevance of CSI to the evolution debate. You are trying to falsify something that nobody claims.

What is wrong with it is that ID demonstrates neither intelligence nor design. All it does is rule out processes with equi-probable outcomes (processes hat nobody proposes for evolution). You then inject this with a dose of unproven dualistic philosophy to draw sweeping and unwarranted conclusions.

As such, ‘ID’ is a complete misnomer, and all these maths have litte to no relevance to biology or evolution.

Why don’t you try instead to model the proposed combinatorial process of heritable variation and natural selection in challenging environments? If you do that, and you can demonstrate that CSI cannot be generated by such a process, you might have a chance to convince people.

OK. Suppose that we had a specification that included all of the most fit genotypes, and they were a small enough fraction of all possible genotypes that you designated the set as having as CSI. And somebody came along and put forward a theorem that showed that evolutionary processes could not get you into that set if you started outside of it. And showed this in general, not just for some particular case.

Now this would be a Big Deal. A

verybig deal. It would basically stop population genetics modeling in its tracks. A hundred years of population genetics theory, gone in a single day.Now note what it requires you to do. You have to have a set, and compare where the population’s genotypes are before and after evolutionary processes work. So the specification used

.must be the same in those two generationsDembski’s Law of Conservation of Complex Specified Information sort-of claimed to do that in

No Free Lunch2002. I say, “sort of” since for all the extensive explaining of what CSI was, he never actually discussed this part of the argument. And if you look carefully, you will see that in sketching the proof of the LCCSI, Dembski has a specification before that is different from the specification after. So whatever it does accomplish, that Law does not keep the specification the same. To contradict population genetics theory one has to keep them the same, and have some law showing that to be in the specified set, you have to already have been in it the generation before.Does EricMH agree with this argument? I hope that he is still reading, because he called for me to make an explanation. And also, this is his thread.

Let me add that, to show that Specified Information is not conserved, we can use a simpler example, which will be found in my 2007 article (Google with terms “Dembski” and “Felsenstein” to find it).

We have a digital image, an array of 10,100 0’s and 1’s, which looks like a flower (it was in fact made from a photo of a flower). And we take a permutation of the integers 1 through 10,100 and we scramble it. We know the permutation, so we can in principle unscramble it any tme we want. The information is conserved, since we can reverse the permutation at will and the image will return, unchanged.

But if we use the specification “looks like a flower” the original image has it, and the scrambled image doesn’t. So although information is conserved

Specified Information isn’t. The amount of it can either increase or decrease (depending on which direction we’re going), while all that time, the Shannon information is conserved.Again, I hope that EricMH is convinced by this example that the conservation of information does not apply to Specified Information.

Thanks Dr. Felsenstein, your explanations are a bit clearer.

For your first example, we’d need to define it a bit more. The important thing is whether the specification is independent of the evolutionary process. If it is, then whether the specification is the same or not is immaterial. Winston’s improbability of ASC still holds.

Your second example is the same as Tom English’s example in his non conservation of ASC post. I pointed out there the problem is he doesn’t appropriately modify the random variable. The transformation must not only be applied to the particular instance, but also the random variable the instance is drawn from, if you want to apply the argument to ASC. Once you transform the random variable, then Winston’s proof holds.

It’s not very satisfying, I’m speaking out of frustration. I’d love to see a good refutation of ID theory, instead of constant misunderstandings, strawmen, completely irrelevant points, etc. I see here at TSZ, at PS, at UD, in articles, and so on. And, the more I learn the basics of fields like information theory, statistics, computer science the more irrelevant the counter arguments become and I see even the ‘counter arguments’ actually support ID theory.

So many very intelligent, highly educated people claim ID is completely bogus. Why, then, is it so hard to formulate an articulate refutation of the theory?

But, it looks like I got some good (perhaps) responses after I complained, so I’ll revisit these comments in the coming days.

I’m not very aware of any of the above, but why do you think his COI for active information doesn’t apply? Why do you claim Winston’s and Montanez’s conservation laws are irrelevant?

EricMH:

Consider the possibility that you’re misunderstanding the counterarguments or failing to recognize the flaws that critics are identifying.

In the other thread, for example, you just praised a paper of Winston Ewert’s:

I identifiedmore than 20 substantive errors in that “neat” paper. You need to be a bit more skeptical of ID and ID-related claims.Not sure if I am following along, but this seems straightforward enough:

Joe said that his specification was “has high fitness”. Since the evolutionary process involves natural selection, the specification does not seem independent to me: natural selection results in an increased representation of genomes that fit the specification.

I’d hope we can set ASC and its conservation aside. I have large doubts that ASC means anything that is relevant to evolution or adaptation. I hope soon to put up a post (at PT) raising my objections to ASC.

In establishing that SI is not conserved, I was thinking of SI as used by William Dembski from 2002 on, in his Law of Conservation of Complex Specified Information, to argue that evolutionary forces cannot accumulate SI in the genome. You have declared that his argument has never been refuted. So I was giving some simple examples showing that he did not succeed in showing that SI cannot be put into the genome by natural selection.

Do you agree with me about that, or not?. It will do no good to redirect attention to later arguments of Dembski and Marks, as the original LCCSI argument was one of those that you declared to be unrefuted. Or have you had second thoughts about that?Note also that Functional Information arguments, which design advocates like to invoke, are using the original SI, not an ASC argument.

Again, let’s set ASC aside, I hope soon to point out its utter irrelevance to refuting that natural selection can bring about substantial amounts of adaptive evolution soon. Maybe that argument of mine will be a failure, maybe not. Let’s get settled first whether you see the force of the arguments against use of ordinary CSI to establish limits on adaptation.

In the original Dembski 2002 argument, where he uses the Law of Conservation of Complex Specified Information, he uses a mapping from

this generation back to the previous one so define a region then that corresponds to the specification now. He argues, straightforwardly, that this constructed specification in the previous generation is a region of just as low probability as the current specification. He’s right about that. But to use that mapping in contructing the previous generation’s specification, one has to use knowledge of the evolutionary processes.

So that’s where the violation of Dembski’s condition, a violation by Dembski himself, occurred.

If Dembski had taken a region (genomes of high enough fitness) and kept it the same in both generations, and somehow shown that if you ended up in it, you had to start in it, that would be a huge problem for evolutionary biology. I would not raise the issue of independence of that specification, because I would then still be stymied by the inability of evolutionary processes to get the genome better-adapted. (In that hypothetical case).

The 2002 year argument???

Don’t you think science has progressed since then, Joe? Or, are you living in the past?

I’m pretty sure that Dembski has fine-tuned his argument as he should, just like any progressive scientist has to do to remain respected… Einstein had fine-tuned and even abandon some of his theories…

It seems obvious to me that you expect Dembski’s current scientific views to affect his past views, just like in quantum mechanics, where retrocausality shows that our future actions affect past events…

This expectation seems reasonable…at least to me… 😉

Thanks for the explanation. To make sure I understand properly: the mapping is a transformation of some bitstring (a genome) from generation t to a modified bitstring at t+1, right? And Dembski argues that the set of genomes at t has a specification that is as unlikely as that at t+1 because there exists a mapping from one to another. SInce he needs to know the reverse mapping he violates the condition of independence.

That part I don’t understand, because that could never be. The whole point of the evolutionary process is to push any genome towards the region of high fitness. In Eric’s previous OP I got the impression that he was not denying that populations could evolve to higher fitness, but rather looking for the place where the information from Intelligence entered the process. He concluded that it was hardcoded in the evolutionary algorithm (Eric, correct me if I misunderstood). Of course it is; his evolutionary algorithm contained the information that allele 1 has higher fitness than 0. In the real world, populations receive that information by feedback from the environment.

Certainly always a possibility, just that does not appear to be the case, in my own humble judgment of my judgment 🙂

But, at the very least it is objectively clear there is no articulate, concise refutation of the core ID claim: that CSI is conserved.

We have long rambling articles by Shallit, Erik, and Wein, that end up being really difficult to make heads or tails of, and when I do finally pin down a supposed refutation it is either a strawman, an irrelevant disagreement, a big bag of insults and condescension, or even supports the ID position.

We have Dr. English and Dr. Felsenstein claiming they have such a refutation, which appears fairly convoluted, and also seems to be the same concept: that you can apply a function to an event and get ASC all over the map. But, as I have pointed out a couple times, you just get a new random variable and the improbability of ASC continues to apply.

Then we have “refutations” that you’ve offered which seem to amount to you disagreeing with how ID theory is applied to biology, which, first of all, is irrelevant to whether the mathematical portion of ID theory is correct, and second seems to be a lot of speculation and personal opinion. Now, it all may be correct, but it is definitely not clearly and articulately stated, nor does it come across as very definitive or even relevant.

Over at PS we have Swamidass insisting he can “disprove ID” which just amounts to him proclaiming “I doubt it” and “you should trust my judgment because I’m an information theory expert.” Least convincing of the lot.

The only cogent response I’ve seen is by Devine, which again is just his personal opinion to utilize algorithmic information more than Dembski does, and is not really a refutation of anything. At least he provides some helpful food for thought and essentially agrees with Dembski.

So, overall I have seen nothing that qualifies as a ‘refutation of ID’, and I believe I have examined all the ‘refutations’ that are available online.

On the other hand, as I’ve also stated multiple times, we have well known, mainstream conservation of information theorems that all the critics seem to ignore, and they act like Dembski is proposing something entirely unheard of.

Thus, the critics seem much less than honest in their criticism of the theory. Instead it appears they have an agenda to discredit ID by any means possible, instead of give the theory a fair hearing.

If the specification is not independent, then the problem is ill posed. Lack of independence means it is an invalid specification. What is the issue?

No, as I responded to Corneel just above, you are calling something else specified information other than what Dembski et. al. are referring to. So, your example is ill posed.

If you can give a specific reference to this argument I can look it up, but currently I do not know what you are talking about. ASC is a form of CSI, and if you want to discard ASC, then I don’t really know what you are referring to.

Anyways, my hopes have been dashed, and I don’t really see any substantive responses. I will have to devote my limited time elsewhere.

Sure. Page 148 of

No Free Lunchwhere Dembski writes:That page is also where you will find the argument about going from a generation back to its antecedents, constructing different specifications of equal strength

I hope that this does not mean that, after calling for a specific reference, you are departing without waiting for it. That would dash

myhopes.But not your (correct I believe) prophecies at Panda’s Thumb.

Yeah, same here.