Wrongly Inferred Design
When claiming to having scientific evidence of Intelligent Design (ID), one of the arguments most often cited by ID proponents is that of Dr. William Dembski, based on a method that he calls the Design Inference. Not only is this method in itself claimed to provide support for the ID position, but it is also said to provide the probability theoretical underpinnings for the Irreducible Complexity argument of Dr. Michael Behe. Thus, the Design Inference plays a very important role in the ID movement’s claims to scientific respectability. If the Design Inference is fatally flawed–as this article will show it is–then the position of the ID movement is seriously weakened.
For Dembski’s most detailed description of the Design Inference, the reader should refer to his monograph The Design Inference: Eliminating Chance Through Small Probabilities , hereafter referred to as TDI. This is an expensive and somewhat technical volume, for both of which reasons it may be inaccessible to the lay reader. However a briefer description of the Design Inference can be found online in Dembski’s article “The Explanatory Filter” .
There have been a number of previous critiques of the Design Inference, but I feel these have overlooked some significant points. Furthermore, there has been a great deal of confusion about the precise nature of the Design Inference, which stems primarily from the high level of equivocation and obfuscation in Dembski’s work. This article will attempt to clarify the nature of the Design Inference and demonstrate the following fundamental flaws:
1. Dembski has failed to solve the problem of specification. 2. Dembski has failed to make clear the precise nature of the Design Inference. Once clarified, it will be seen to be either trivially simple or invalid. 3. Dembski has never provided any data to support his claim that the Design Inference has successfully detected design in biological structures.
1. The Problem of Specification
A major part of the Design Inference is “…eliminating chance through small probabilities” (to quote the subtitle of TDI). The tool by which Dembski tries to achieve this aim is his Law of Small Probabilities, which states that “…specified events of small probability do not occur by chance” (TDI, p 5). The issue of specification is crucial to this method. A specification made before observing the outcome of an event is equivalent to the “rejection region” of conventional statistical tests. Where Dembski departs from conventional statistics is in claiming to have a method of formulating a valid specification after observing the event. If this claim were well-founded, then Dembski would have made a revolutionary contribution to statistical theory.
The problem with formulating a specification after observing the outcome of an event is that one can almost invariably find a small-probability rejection region which includes the observed outcome, by deliberately matching the rejection region to the outcome, and thereby reject the chance hypothesis in question. Such a contrived rejection region is called a “fabrication” by Dembski. To quote Dembski: “Specifications are the non-ad hoc patterns that can legitimately be used to eliminate chance and warrant a design inference. Fabrications are the ad hoc patterns that cannot legitimately be used to eliminate chance” (TDI, p 13).
Dembski claims to have a reliable method for distinguishing between specifications and fabrications, which he describes at some length in TDI. However, he gives us no theoretical justification for this method. Indeed, in a reply to criticism from Eells, he claims that no such justification is needed, because he is merely “…offering a rational reconstruction of a common human activity” . This seems to be a reference to a rather inscrutable passage at the top of page 147 of TDI. But why should we accept that this is a reconstruction of how we actually think when we intuitively infer design? Surely a reconstruction must be based on empirical evidence, and Dembski offers us none. With neither a theoretical justification nor any empirical evidence, all we have is Dembski’s intuition, and he fails even to support his intuition with any fully worked-out examples of successful applications of the Design Inference. (In the Caputo case, his most detailed example, he doesn’t establish a probability bound, so we don’t know if the calculated probability is small enough to infer design.) To show that Dembski’s intuition is wrong, I will present two counterexamples, in which patterns satisfying his criteria for being specifications actually turn out to be fabrications.
My first counterexample is based on one of Dembski’s own examples, the Caputo case. Dembski wishes to test the chance hypothesis that the ballots were selected at random (with equal probabilities of Democrat or Republican being selected), and proposes the following specification: “The Democrats got the top ballot line at least 40 times.” In TDI (pp 166-167) he claims that this pattern fulfills all his requirements for being a specification. However my contention is that this pattern is in reality a fabrication, because it has been selected to fit the observed event. To see this more clearly, let us suppose–contrary to fact–that the 41 ballots in question had been headed by Democrats and Republicans in strict alternation, or that all of them had been headed by a Republican candidate. In place of the historical sequence DDDDDDDDDDDDDDDDDDDDDDRDDDDDDDDDDDDDDDDDD we might have seen RDRDRDRDRDRDRDRDRDRDRDRDRDRDRDRDRDRDRDRDR or RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRR.
While these hypothetical outcomes would not have favoured Caputo’s party, as the historical one did, they should still have led an astute observer to suspect that the draws were not random (i.e. were not in accordance with the chance hypothesis). We can speculate on reasons why Caputo might have engineered such sequences: in the first case, perhaps he deliberately alternated the ballots out of a misguided desire to be seen as unbiased; in the second case, perhaps he was secretly in the pay of the Republican party. But such speculations are unnecessary. The fact is that, if one of these outcomes had been observed, we would still have had grounds to conduct a test of the same chance hypothesis, but we would have selected a different specification. Hence, Dembski’s specification is actually an ad hoc fabrication selected to fit the observed outcome. A genuine specification should have included all outcomes that were as noteworthy (in some sense) as the one actually observed. By excluding some such outcomes, Dembski has made the specification too narrow, and thus calculated too small a probability.
For my second counterexample, let’s say that I’m sent a computer text file which contains just one English word, say “design.” I wish to test the chance hypothesis that the characters in the text file were generated at random. To be precise, I’ll take as my chance hypothesis: “randomly selected characters were added to the file one at a time until an end-of-line character occurred.” I propose the specification: “any English word.” This pattern meets Dembski’s three criteria for a valid specification, namely:
* CINDE. As my “side information,” I take the contents of an English dictionary. (For simplicity, let’s not worry about unlisted variations in word endings, such as plurals.) My selection of a dictionary has no effect on the probability of receiving the word “design,” given the chance hypothesis, so the conditional independence criterion is met. Note that Dembski’s formulation of this criterion is confusing. It is stated formally as “P(E|H&J) =
P(E|H) for any information J generated by I” (TDI, p 152). But, if the chance hypothesis (H) uniquely defines a probability distribution, as it does in all Dembski’s examples and must do if we are to calculate a probability, then additional side information (I) cannot possibly alter the probability of an outcome. From Demsbki’s examples, however, particularly the Caputo case (TDI, pp 166-167), it appears that the CINDE criterion could be more clearly stated as the requirement that the side information include only information known independently of the outcome of the event. Using Dembski’s words, I can say that my side information meets the requirement, because the contents of my English dictionary “can be known and exhibited irrespective of” my receipt of any text file (TDI, p 167).
* TRACT. In formal terms, Dembski states this criterion as “phi(D|I) < lambda” (TDI, p 152). But, since he gives us no procedure for selecting a complexity measure phi or a limit lambda, it’s unclear how we should apply the criterion. In less formal terms, Dembski states the criterion as “…a matter of confirming that the problem of formulating D [the specification] on the basis of I [the side information] is tractable for S [a subject]” (TDI, p 146). It’s not entirely clear what this means, but at least one subject (me) had no difficulty formulating the specification, so I assume that the criterion is met. (This criterion seems to be redundant. Dembski’s method already requires us to formulate a specification. If the problem of formulating a specification is intractable, then we obviously cannot apply the method. Why does this need to be stated as an additional criterion? To quote from Fitelson, Stephens and Sober : “Because tractability depends on your choice of language and computational procedures, we think that TRACT has no evidential significance at all.”)
* DELIM. The word “design” will be in our English dictionary, so the delimiter condition is satisfied.
So it seems that my proposed specification is, according to Dembski’s criteria, a valid one. However, in reality it is nothing more than a fabrication, because I chose my side information (and hence my specification) in an ad hoc fashion to fit the pattern that I noticed in the text file, namely the pattern of an English word. Had I received the sequence “dessein,” I could have selected the specification “any French word.” Had I received the sequence “jogfsfodf,” I could have selected the specification “any English word with the letters shifted to the next in the alphabet” (shifting the letters back again gives the word “inference”). And so on. Each of these possible sequences, and many others besides, could be considered just as noteworthy as “design,” so our specification should be wide enough to include all of them, and we should calculate the probability of any of these sequences occurring. Dembski’s method causes us to calculate far too low a probability in this case, which means we’re liable to reject the chance hypothesis too easily.
The fundamental flaw in Dembski’s specification method is this. While the CINDE criterion prevents us from selecting side information containing information about the outcome of the event, it doesn’t prevent us from making our selection based on the outcome of the event. Since we choose our side information in the light of the outcome, it is not independent of the outcome, and therefore neither is the specification.
Before leaving the subject of specification, I’d like to consider how the concept might be applied to the phenomenon in which ID proponents are really interested–the origin of biological structures. Although Dembski has failed to give an adequate account of specification, I nevertheless claim that “the origin of intelligent life” is a specified event, because, without it, there would be no-one here to formulate the specification. Thus, this cannot be an ad hoc fabrication just concocted to fit the observed event. It is an essential prerequisite of the observed event. It also follows that any event which is an essential prerequisite of intelligent life (such as “the origin of life”) is also a specified event. The problem for the ID proponent, however, is that such specifications are probably too vague to be of any practical use, as we cannot imagine all the different forms that life might take.
Consider for example the common Creationist argument that this or that biological molecule could not have occurred by chance. Let’s take hemoglobin as an example. While the typical creationist would probably attempt to calculate the probability of the origin of hemoglobin in the specific form that we see it today, a proponent of Dembski’s methods might recognise the need for a wider specification, and might choose as his specification “any molecule which would perform the same function as hemoglobin.” Perhaps it might even be possible to place reasonable bounds on the range of possible forms that such a molecule might take, enabling a probability to be calculated. However, this would still not be an adequate specification unless we could establish that this function is essential for intelligent life. Otherwise, we would simply be fitting the specification in an ad hoc way to the particular form of life that we observe.
Associated with the problem of specification is the problem of establishing a probability bound, i.e. how “small” does a probability have to be before we can reject the chance hypothesis? I claim that Dembski’s approach to probability bounds is also fundamentally flawed. I don’t propose to discuss this issue in detail here, since, without an adequate specification, the problem of establishing a probability bound does not even arise. Suffice to say that Dembski’s failure to establish a probability bound in the Caputo case is, in my opinion, indicative of the problems with his method. In the case of the origin of life, however, since this is an essential prerequisite to our being here as observers, I consider Dembski’s universal probability bound (1/2 x 1/10^150) to be reasonable, indeed conservative, providing we assume that there are no other universes besides the one in which we exist (or only a small number of such universes). Dembski considers the possible relevance of other universes, but dismisses it as “the inflationary fallacy” (TDI, pp 214-217). While he’s right to dismiss the relevance of other universes in the general case, such as Mother Teresa turning out to have been an ax murderer (Dembski’s example, not mine!), he cannot do so for the case of the origin of life, because of the well-known “selection effect.” Dembski himself describes and accepts the relevance of the selection effect with regard to the number of planets on which life could potentially have originated (TDI, pp 182-183), and exactly the same argument applies with regard to the number of universes in which life could potentially have originated. I’m not saying that other universes actually exist. That is a moot point. But, if they do exist, they are certainly relevant to the probability of the origin of life.
2. What is the Design Inference?
As far as I have been able to determine, the Design Inference amounts to no more than the following: once we’ve eliminated all regularity and chance hypotheses that could explain a phenomenon, we must conclude that the phenomenon was the result of design, and the way that we eliminate chance hypotheses is by using the Law of Small Probabilities. Unfortunately, this simple idea is couched in a lot of confusing and equivocal language, making it difficult to understand exactly what Dembski means. As a result, there are a number of substantially different interpretations of the Design Inference in circulation, among both supporters and critics of Dembski. I have attempted to obtain clarification of these issues from Dembski, but none has been forthcoming. I therefore cannot be sure that my interpretation of Dembski is correct, but I believe the interpretation I’ve arrived at is the most charitable one, indeed the only one that makes any sense of the Design Inference.
Note. Dembski initially refers to his procedure for detecting design as the Explanatory Filter and to the underlying logical argument as the Design Inference. At times, however, he refers to the procedure as the Design Inference. For simplicity, I therefore treat the two terms as synonymous, and deliberately use the term Design Inference in both senses.
2.1 The Design Inference is Eliminative
The Design Inference infers design by eliminating chance and regularity as explanations. Furthermore, it’s not sufficient to eliminate just one chance explanation. The formal logic of the Design Inference, on pages 50-51 of TDI, makes clear that we must consider and reject “…all the relevant chance hypotheses that could be responsible for E [the observed event]…”
I believe that some readers of Dembski have missed this vital point, and mistakenly come to the conclusion that only one chance hypothesis need be considered. This may be because all but one of Dembski’s examples entertain only a single chance hypothesis (and that one occurs only in a footnote), or it may be because of several misleading passages, such as the following:
â€œBy contrast, a successful design inference sweeps the field clear of chance hypotheses. The design inference, in inferring design, eliminates chance entirely, whereas statistical hypothesis testing, in eliminating one chance hypothesis, opens the door to others.â€ [TDI, p 7]
This passage seems to have it the wrong way around. It is not inferring design that eliminates chance entirely; it is eliminating chance entirely that allows us to infer design.
Because this is such a vital point, and yet has been widely misunderstood, let me quote some more passages which show that the Design Inference requires all chance hypotheses to be considered and rejected before design can be inferred:
â€œThe safecracking example presents the simplest case of a design inference. Here there is only one possible chance hypothesis to consider, and when it is eliminated, any appeal to chance is effectively dead. Design inferences in which multiple chance hypotheses have to be considered and then eliminated arise as well. We might, for instance, imagine explaining the occurrence of a hundred heads in a row from a coin that is either fair or weighted in favor of heads with probability of heads 0.75. To eliminate chance and infer design in this example we would have to eliminate two chance hypotheses, one where the probability of heads is 0.5 (i.e., the coin is fair) and the other where the probability of heads is 0.75. To do this we would have to make sure that for both probability distributions a hundred heads in a row is an SP [small probability] event, and then show that this event is also specified. In case still more chance hypotheses are operating, design follows only if each of these additional chance hypotheses gets eliminated as well, which means that the event has to be an SP event with respect to all the relevant chance hypotheses and in each case be specified as well.â€ [TDI, p 44, footnote]
â€œSince an event has to have small probability to eliminate chance, and since the design inference infers design by eliminating all relevant chance hypotheses, SP(E;H) has to be satisfied for all H in [curly H].â€ [TDI, p 52]
â€œGiven a chance hypothesis H that could conceivably explain E, S [a subject] is obliged to retain H as a live possibility until a positive warrant is found for rejecting it.â€ [TDI, p 220]
Note that Dembski refers above to “…eliminating all relevant chance hypotheses…” (my emphasis). He uses the word “relevant” several times in this context, but never elaborates on what he means by it. However, the final passage quoted above indicates that we must eliminate any chance hypothesis we can think of that could conceivably explain the observed event. It should be clear from this that any inference of design is only as secure as our ability to think of the relevant chance hypotheses. The true explanation for the event may be a chance hypothesis of which we are unaware, and in that case we are liable to erroneously arrive at a conclusion of design. Dembski, however, with assertions such as “The sp/SP events of the Explanatory Filter exclude chance decisively…” (TDI, p 41), gives the impression that such errors cannot occur. Furthermore, elsewhere he makes this claim explicit:
â€œI argue that the explantory [sic] filter is a reliable criterion for detecting design. Alternatively, I argue that the Explanatory Filter successfully avoids false positives. Thus whenever the Explanatory Filter attributes design, it does so correctly.â€ 
This claim is quite clearly untenable. Since it predates TDI, and is not made explicitly in TDI, perhaps we should not take it too seriously. Indeed, even a close associate of Dembski, with whom I raised the issue, was surprised by it. However, if Dembski no longer wishes to stand by this claim, he should clarify the issue by making a clear retraction.
2.2 Chance Hypotheses
Unfortunately, Dembski never defines the vital term “chance hypothesis.” This omission becomes particularly problematical when we try to apply the Design Inference to biological structures. I propose that “Darwinian evolution” is a relevant chance hypothesis for the origin of biological structures. But many readers of Dembski, both critics and supporters, have disagreed, on the grounds that Darwinian evolution is a combination of chance and regularity, not just chance. Dembski himself has not clearly stated his position on this. But it seems to me that the position of those who claim it is not a chance hypothesis is untenable, for the following three reasons:
* If it isn’t a relevant chance hypothesis, and therefore doesn’t have to be considered and rejected before we can infer design, then the Design Inference is clearly nonsensical. How can we conclude design through an eliminative procedure without eliminating the natural explanation which is most widely held by the scientific community?
* All chance events (except maybe quantum events) are subject to the effect of non-probabilistic natural laws, and therefore involve an element of regularity. For example, the roll of a die is subject to mechanical forces. Since all chance hypotheses involve an element of regularity, it makes no sense to distinguish chance hypotheses from chance-and-regularity hypotheses.
* Dembski himself has written elsewhere:
â€œDoes nature exhibit actual specified complexity? This is the million dollar question. Michael Behe’s notion of irreducible complexity is purported to be a case of actual specified complexity and to be exhibited in real biochemical systems (cf. his book Darwin’s Black Box). If such systems are, as Behe claims, highly improbable and thus genuinely complex with respect to the Darwinian mechanism of mutation and natural selection and if they are specified in virtue of their highly specific function (Behe looks to such systems as the bacterial flagellum), then a door is reopened for design in science that has been closed for well over a century. Does nature exhibit actual specified complexity? The jury is still out.â€ 
This makes it pretty clear that Dembski himself considers the “Darwinian mechanism” to be a chance hypothesis that must be rejected in order to infer design.
I therefore assume that “Darwinian evolution” is a relevant chance hypothesis for the origin of biological structures.
Furthermore, evolutionary biologists do not claim that their understanding of evolution is perfect. There are many uncertainties and unknowns. So any attempt to rule out “Darwinian evolution” must consider all possible variations of the evolutionary process. Thus, in practice “Darwinian evolution” is not just a single chance hypothesis, but a whole family of chance hypotheses, and the Design Inference requires us to reject all of them before inferring design.
Dembski tells us that there are two categories of explanation that must be rejected before we can infer design: regularity and chance. But what’s the difference between a regularity explanation and a chance explanation? According to Dembski a regularity explanation is one according to which “…E is highly probable…” (TDI, p 38). What’s more, even deterministic events can be treated as highly probable ones:
â€œIt is convenient to think of all such regularities as probabilistic, assimilating the nonprobabilistic case to the probabilistic case in which probabilities collapse to 0 and 1.â€ [TDI, p 38]
Since Dembski never establishes any reason why highly probable events should be treated differently from the events of “intermediate probability” which correspond to chance explanations, or gives any method for establishing a boundary value to separate the two categories, there is no theoretical reason why we should not treat regularities as a type of chance hypothesis; they just happen to have higher probabilities.
2.4 Design vs Intelligent Agency
Dembski has caused considerable confusion through his peculiar use of the terms “design” and “intelligent agency.” He defines “design” as “the set-theoretic complement of the disjunction regularity-or-chance” (TDI, p 36), i.e. it’s what is left over after eliminating regularity and chance explanations. But he also specifically denies that an inference of “design” necessarily entails intelligent agency:
â€œThus, even though a design inference is frequently the first step toward identifying an intelligent agent, design as inferred from the design inference does not logically entail an intelligent agent. The design that emerges from the design inference must not be conflated with intelligent agency.â€ [TDI, p 9]
On the other hand, he never gives any additional criterion for distinguishing intelligent agency from mere “design,” and he goes on to write:
â€œYet in practice, to infer design is not simply to eliminate regularity and chance, but to detect the activity of an intelligent agent.â€ [TDI, p 62]
I will not attempt to reconcile these apparently contradictory statements. Instead I will simply assume that the latter statement accurately reflects Dembski’s position, and will ignore the former statement. In other words, I will assume Dembski is claiming that rejection of chance and regularity does allow us to infer the involvement of an intelligent agent. This then allows us to consider the terms “design” and “intelligent agency” to be synonymous (and I assume they are both synonymous with the more widely used term “intelligent design”). If this assumption is not correct, then Dembski needs to clearly state what additional criterion is necessary to distinguish intelligent agency from mere “design,” or else recognise that the Design Inference cannot detect intelligent agency.
The question then is whether Dembski is justified in claiming that eliminating regularities and chance allows us to infer design/intelligent agency. Well, it seems to me that this is trivially true. As explained above, the Design Inference is eliminative. When we infer design, we are effectively saying: “We’ve eliminated all the non-design explanations we can think of, and we assume that there are no more non-design explanations that we can’t think of.” It follows (providing our assumption is correct) that the true explanation must involve design. Because of this eliminative approach, we conveniently have no need to define design/intelligent agency. It can mean whatever we want it to mean, as long as it excludes all the regularity and chance explanations that we’ve explicitly rejected. This is useful, because, as far as I can tell, proponents of Intelligent Design have not clearly defined what they mean by the term.
2.5 Specified Complexity and CSI
Another factor which has contributed to confusion over the Design Inference has been Dembski’s use of the terms “specified complexity” and “CSI” (Complex Specified Information). Althought TDI explores the the relationship between probability, complexity and information, it doesn’t use the terms specified complexity or CSI. For explanations of these terms, the reader must refer to other articles by Dembski  . It appears, however, that these two terms are eqivalent, and that their introduction involves no change in the method of the Design Inference. All that’s happened is that Dembski has transformed the calculated probabilities, and the probability bounds with which they’re compared, by applying a simple function: taking the logarithm to base 2 and then negating, i.e. specified complexity or CSI = -log2(P), where P is the probability of a specified event or a probability bound. Since the same transformation has been applied to both sides of an inequality, there is no net effect. If this interpretation is not correct, then Dembski needs to clearly state how his method changes with the use of these terms, and why. The point is significant, because, although these terms apparently add nothing to Dembski’s method, they do serve to obfuscate the issues in two important ways.
First, the Design Inference requires the probability of a specified event to be calculated (and shown to be small) for each relevant chance hypothesis. However, when Dembski uses the term specified complexity (or CSI), he generally refers simply to the specified complexity of an event or phenomenon, ignoring the fact that the specified complexity (which is a function of probability) must be defined with respect to a particular chance hypothesis. Thus, he obscures the fact that, if there is more than one relevant chance hypothesis, then the event or phenomenon has more than one measure of specified complexity. Furthermore, if we know that an object is designed (let’s say we watched it being made), does it make any sense to say it has specified complexity? If we know the true explanation, and it is design, then there are no relevant chance hypotheses with respect to which we can calculate the specified complexity!
Second, Dembski’s use of the terms specified complexity, information and CSI is misleading, because he conflates his own use of these terms with those of other writers, without ever establishing that both he and they are using them in the same sense. For example:
â€œAs Davies puts it: â€˜Living organisms are mysterious not for their complexity per se, but for their tightly specified complexityâ€™ (p. 112).â€ 
â€œManfred Eigen, for instance, writes, â€˜Our task is to find an algorithm, a natural law that leads to the origin of information,â€™ where by â€˜informationâ€™ I understand him to mean specified complexity.â€ 
â€œCSI is what all the fuss over information has been about in recent years, not just in biology, but in science generally. It is CSI that for Manfred Eigen constitutes the great mystery of biology, and one he hopes eventually to unravel in terms of algorithms and natural laws. It is CSI that for cosmologists underlies the fine-tuning of the universe, and which the various anthropic principles attempt to understand (cf. Barrow and Tipler, 1986). It is CSI that David Bohm’s quantum potentials are extracting when they scour the microworld for what Bohm calls â€˜active informationâ€™ (cf. Bohm, 1993, pp. 35-38). It is CSI that enables Maxwell’s demon to outsmart a thermodynamic system tending towards thermal equilibrium (cf. Landauer, 1991, p. 26). It is CSI on which David Chalmers hopes to base a comprehensive theory of human consciousness (cf. Chalmers, 1996, ch. 8). It is CSI that within the Kolmogorov-Chaitin theory of algorithmic information takes the form of highly compressible, non-random strings of digits (cf. Kolmogorov, 1965; Chaitin, 1966).â€ 
â€œEvolutionary biology has steadfastly resisted attributing CSI to intelligent causation. Although Manfred Eigen recognizes that the central problem of evolutionary biology is the origin of CSI, he has no thought of attributing CSI to intelligent causation. According to Eigen natural causes are adequate to explain the origin of CSI. The only question for Eigen is which natural causes explain the origin of CSI. The logically prior question of whether natural causes are even in-principle capable of explaining the origin of CSI he ignores.â€ 
In this way, Dembski gives the impression that the existence of specified complexity or CSI in nature is already widely accepted. But, if these other writers are using the terms to mean different things, then this impression is a false one. While I’m not familiar with the work of these other writers, I’m highly doubtful that they are using these terms with the same meaning as Dembski. In any case, the onus is on Dembski to show that the meanings are equivalent before using the terms in this way. (I doubt whether the term CSI is actually used at all by the cited writers. I suspect Dembski is just giving his own interpretation of their work.)
2.6 Summary of Section 2
* The precise nature of the Design Inference is unclear from Dembski’s writings.
* By the most charitable interpretation of the Design Inference (the only one that seems to make any sense), it amounts to no more than the simple statement that, once we’ve eliminated all the non-design hypotheses that we can think of, we should conclude that the phenomenon was the result of design, and the way that we eliminate chance hypotheses is by using the Law of Small Probabilities.
* The Design Inference cannot rule out false positives, as has been claimed by Dembski. After making an inference of design, the possibility always remains that we have overlooked the true non-design explanation.
* Dembski’s introduction of the terms specified complexity and CSI has done nothing but add to the confusion surrounding the Design Inference.
I suspect that Dembski will respond to these charges (if he responds at all) by saying that I’ve misunderstood him. If this turns out to be the case, then I must point out that such misunderstandings are rife among Dembski’s supporters as well as his opponents. Even his closest associates don’t seem to understand the Design Inference, and Dembski himself refuses to answer questions about it. Thus, the blame for any misunderstanding must rest squarely on Dembski’s shoulders. If my interpretation of the Design Inference is incorrect, then it’s time for Dembski to state clearly and publicly exactly what the method of the Design Inference is.
3. Do the Calculation
In a passage quoted above, Dembski wrote: â€œDoes nature exhibit actual specified complexity? The jury is still out.” Elsewhere, however, he has claimed that specified complexity has been detected in biochemical systems, and that therefore design has been found in nature:
“So there exists a reliable criterion for detecting design strictly from observational features of the world. This criterion belongs to probability and complexity theory, not to metaphysics and theology. And although it cannot achieve logical demonstration, it does achieve a statistical justification so compelling as to demand assent. This criterion is relevant to biology. When applied to the complex, information-rich structures of biology, it detects design. In particular, we can say with the weight of science behind us that the complexity-specification criterion shows Michael Behe’s irreducibly complex biochemical systems to be designed.” 
An almost identical passage appears in Dembski’s book “Intelligent Design” .
In TDI (p 228), Dembski stresses:
“There is a calculation to be performed. Do the calculation. Take the numbers seriously. See if the underlying probabilities really are small enough to yield design.”
“I stress again, Do the probability calculation!”
So the question is, where’s the probability calculation that justifies the claim made by Dembski above? Well, I’ve searched high and low, as have other critics, and I have been unable to find any such calculation. No such calculation appears in TDI or in any other work of Dembski’s that I’ve seen. From the quote above, one might assume that the calculation can be found in Behe’s work. However, although Dembski writes that “the CSI of a flagellum far exceeds 500 bits” [8, p 178], he provides no calculation to support that claim, and there is no such calculation in Behe’s book, Darwin’s Black Box . Indeed, Behe’s probabilistic claims seem to be based on intuition and not calculation:
“Systems requiring several parts to function that need not be well-matched, we can call “simple interactive” systems (designated ‘SI’). Ones that require well-matched components are irreducibly complex (‘IC’). The line dividing SI and IC systems is not sharp, because assignment to one or the other category is based on probabilistic factors which often are hard to calculate and generally have to be intuitively estimated based on always-incomplete background knowledge. Moreover, no law of physics automatically rules out the chance origin of even the most intricate IC system. As complexity increases, however, the odds become so abysmally low that we reject chance as an explanation (Dembski 1998).”  [The reference is to TDI.]
In case I’d overlooked some relevant calculation, I wrote to Dembski asking for a citation. Although he replied to my email, he failed to cite any calculation, and simply repeated the flagellum claim quoted above. A similar request sent by Wesley Elsberry has (as of the date of writing) gone unanswered.
Dembski’s failure to provide empirical data in support of his claim not only shows that claim to be hollow, but also prevents us from arriving at a clearer understanding of the method of the Design Inference. If Dembski presented a detailed working of an application of the Design Inference to a biological structure, that should help resolve the various uncertainties mentioned above.
Supporters of Dembski have sometimes claimed that this or that calculation is an application of the Design Inference. However, given the uncertainties about what an application of the Design Inference entails, it is impossible to accept such claims coming from any source other than Dembski himself.
So I call on Dembski either to cite a calculation which substantiates his claim, or else to retract the claim.
If Dembski could genuinely show that the probability of any sort of life evolving in this universe was below his universal probability bound, then I believe he would be justified in claiming that life did not evolve, providing we don’t resort to a multiplicity of universes. But this idea is hardly a new one, and we didn’t need all the panoply of TDI to tell us this. As Dembski himself points out (TDI, pp 55-62), this sort of argument is already widely accepted by people on all sides of the debate, including Dawkins, an arch-opponent of ID. The problem for ID proponents is that no-one has yet been able to perform a valid probability calculation of this sort, and it seems unlikely that anyone will be able to do so in the foreseeable future. All Dembski has achieved with his Design Inference has been to muddy the waters, and to give a false aura of respectability to the sort of bogus probability calculations that creationists have been trotting out for years.
 Dembski, W.A., The Design Inference: Eliminating Chance Through Small Probabilities,CambridgeUniversity Press, 1998.
 Dembski, W.A., â€œThe Explanatory Filter: A three-part filter for understanding how to separate and identify cause from intelligent design,” an excerpt from a paper presented at the 1996 Mere Creation conference, originally titled “Redesigning Science,” http://www.arn.org/docs/dembski/wd_explfilter.htm.
 Dembski, W.A., “How Not to Analyze Design”, http://www.leaderu.com/offices/dembski/docs/bd-analyze.html.
 Fitelson, Stephens and Sober, “How Not To Detect Design,” Philosophy of Science 66 (3), 1999,http://philosophy.wisc.edu/sober/dembski.pdf.
 Dembski, W.A., “Explaining Specified Complexity,” Metaviews 139, 1999 https://www.metanexus.net/Magazine/tabid/68/id/3066/Default.aspx.
 Dembski, W.A., “Intelligent Design as a Theory of Information,” 1998, http://www.arn.org/docs/dembski/wd_idtheory.htm.
 Dembski, W.A., “Science and Design,” First Things 86, 1998, http://www.arn.org/ftissues/ft9810/dembski.html.
 Dembski, W.A., Intelligent Design: The Bridge Between Science & Theology, InterVarsity Press, 1999.
 Behe, M.J., Darwin‘s Black Box, Simon & Schuster, 1998.
 Behe, M.J., â€œSelf-Organization and Irreducibly Complex Systems: A Reply to Shanks and Joplin,” Philosophy of Science 67 (1), 2000, http://www.arn.org/docs/behe/mb_selforganizationreplytoshanksandjoplin.htm.