The Design Inference

From: Richard Wein (rwein@lineone.net)
Date: Sun May 14 2000 - 11:35:31 EDT

  • Next message: Stephen E. Jones: "Re: Scientists Complete Map of Second Human Chromosome, etc."

    Until now, I've based my dismissal of Dembski's design argument on the
    articles I've found on the web. I've read all of Dembski's non-theological
    articles that I've been able to find, and several reviews of his book, "The
    Design Inference". However, I wanted to read the book itself, and have it
    had on request from the public library for many weeks. Now, at last, I've
    received it and read it. (No, the wait wasn't because of it's
    popularity--quite the opposite! The book is virtually unknown here in
    Britain, and the library had to acquire it through the
    inter-library loan system.)

    I'd like to summarize my views on the book, and hopefully provoke some
    discussion. (By the way, I studied statistics at university, to BSc level,
    but that was a long, long time ago, so I would consider myself a
    well-informed layman on the subject of statistics.)

    The Law of Small Probability
    ---------------------------------------
    Most of the book is devoted to establishing this law, which says--put
    simply--that "specified events of small probability do not occur by chance".

    It seems to me that this law introduces two novelties into statistical
    theory:
    (a) It allows a specification to be established retrospectively, i.e. after
    the event in question has occurred.
    (b) It provides a method for setting a small probability bound, below which
    we can justifiably say that events do not occur.
    Let me say that these are novelties as far as I'm concerned, but I can't say
    with any confidence that they haven't already been addressed in the
    statistics literature.

    Now, I don't propose to discuss the LSP in detail. Such a discussion would
    be pretty technical and probably not of interest to many readers here. If
    Dembski has succeeded in putting this law on a firm theoretical basis, then
    I think he will have made a significant contribution to statistics. However,
    I'm rather doubtful about whether he has done so. Several of his inferences
    seem dubious to me. But I don't feel competent to make a definite
    pronouncement on the matter. I'd like to wait and see what the statistics
    community has to say on the matter. Does anyone here know what the reaction
    of the statistics community has been?

    Anyway, regardless of whether this law does indeed have a firm theoretical
    foundation, I'm quite willing to accept it as a practical methodology. The
    law *seems* reasonable, and it's one which we all intuitively apply. It's
    our intuition of such a law that enables us to cry foul when we see someone
    deal a perfect Bridge hand (each player receives 13 cards of one suit). It's
    our intuition of such a law that leads us to conclude that Nicholas Caputo
    rigged the ballots (in Dembski's favourite example).

    Dembski wants to establish this law because he hopes to use it to to prove
    that life could not occur by chance. Well, I have no problem with that.
    Dawkins implicitly uses such a law when he argues, in The Blind Watchmaker,
    that "We can accept a certain amount of luck in our explanations, but not
    too much."

    In developing his LSP, Dembski is doing science (or perhaps, more
    accurately, mathematics). Whether it is good or bad science remains to be
    seen (as far as I'm concerned). However, when he moves on from the LSP to
    the Explanatory Filter, Dembski jumps from the arena of science into the
    quagmire of pseudo-science.

    The Explanatory Filter
    ------------------------------
    Dembski's Explanatory Filter (EF) says that, once you've eliminated
    regularity and chance as possible explanations of an event, you must
    conclude that the event is the result of design. So what's wrong with this?

    Well, first of all, Dembski is equivocal about what he means by "design". He
    initially defines design to be the "set-theoretic complement of the
    disjunction regularity-or-chance", or, in other words: "To attribute an
    event to design is to say that it cannot reasonably referred to either
    regularity or chance" (p. 36). By this definition, the EF is tautological,
    but Dembski promises that he will later provide us with a means of
    determining which cases of "design" can be attributed to "intelligent
    agency". Or is he going to attribute *all* cases of design to intelligent
    agency? This is where Dembski starts to equivocate.

    i) On page 36, he writes: "The principal advantage of characterizing design
    as the complement of regularity and chance is that it avoids committing
    itself to a doctrine of intelligent agency. In practice, when we eliminate
    regularity and chance, we typically do end up with an intelligent agent.
    Thus, in practice, to infer design is typically to end up with a "designer"
    in the classical sense." Dembski's use of the word "typically" strongly
    imples that not all cases of design can be attributed to intelligent agency,
    i.e. that design does not necessarily imply intelligent agency.

    ii) In Section 2.4, "From Design to Agency" (starting p. 62), Dembski
    returns to this issue and attempts to establish a connection between design
    and (intelligent) agency. I'm not going to address the issue of whether he
    succeeds in doing so. All I'm interested in for now is whether he claims to
    establish a *necessary* connection, i.e. that *all* cases of design can be
    attributed to agency. The answer is that he does. In the final paragraph of
    the section, he summarizes: "It's now clear why the Explanatory Filter is so
    well suited for recognizing intelligent agency: for the Explanatory Filter
    to infer design coincides with how we recognize intelligent agency
    generally." And again: "It follows that the filter formalizes what we have
    been doing right along when we recognize intelligent agents. The Explanatory
    Filter pinpoints how we recognize intelligent agency" (p. 66).

    iii) In case anyone should try to reconcile the contradiction of (i) and
    (ii) above by claiming that "typically" should be read as something like "to
    all intents and purposes", let me point out that Dembski actually gives an
    example of a situation where the EF (according to Dembski) indicates design
    but we where we cannot (according to Dembski) infer an intelligent agency.
    The example is on page 226, and I'll give details if anyone is interested.
    But I think it's sufficient to note Dembski's conclusion: "Thus, even though
    in practice inferring design is the first step in identifying an intelligent
    agent, taken by itself design does not require that such an agent be
    posited. The notion of design that emerges from the design inference must
    not be confused with intelligent agency." (Note that the terms "design
    inference" and "Explanatory Filter" appear to be synonyomous. One might have
    assumed that DI = EF + the mysterious extra criterion that allows us to
    distinguish between simple design and intelligent agency, but the last
    sentence quoted shows that this cannot be the case.)

    So despite, the claim to the contrary on page 66, it seems that the EF on
    its own is not sufficient to identify intelligent agency. In that case, what
    additional information is required? Dembski continues (p. 227): "When the
    design inference infers design, its primary effect is to limit our
    explanatory options. Only secondarily does it help identify a cause. To
    investigate a cause we need to investigate the particulars of the situation
    where design was inferred. Simply put, we need more details. In the Caputo
    case, for instance, it seems clear enough what the causal story is, namely,
    that Caputo cheated. In the probabilistically isomorphic case of Alice and
    Bob, however, we may have to live without a causal explanation..." So, in
    order to attribute the Caputo case to design, we need to know the causal
    design story (he cheated). But the whole purpose of the design inference was
    to give us a way of identifying design *without* knowing the causal story.
    Dembski has just shot himself in the foot!

    Having seen Dembski demolish the whole raison d'etre of his own EF, it
    hardly seems worth discussing it any further. But I'd like to clear up
    another point of confusion, namely the distinction between Dembski's
    "regularity" and "chance" categories.

    It seems rather confusing to use the name "regularity" in opposition to
    "chance", since even chance events exhibit regularities. What is a
    probability distribution if not a regularity? When we look further, we see
    that the events Dembski assigns to the regularity category are those which
    "will (almost) always happen". In other words, those with a high probability
    (a probability of 1 or nearly 1). In fact, Dembski later refers to them as
    "highly probable" (HP) events. Dembski also refers to chance events as
    events of "intermediate probability" (IP). So why draw a distinction between
    HP and IP events? After all, the boundary between them is undefined (Dembski
    never gives a boundary probability), and both categories of events are going
    to ultimately suffer the same fate (be dismissed as not due to design). I
    can see no theoretical reason for distinguishing between them, only a
    practical one: when considering the nature of a particular event, we can
    rule out design immediately if its probability is sufficiently high--there's
    no need to waste time worrying about whether it's specified or not. From a
    logical point of view, however, we might just as well lump these two
    categories together. And, if we do that, what should we call the new
    category? Well, we could call it "regularity", since, as I've already said,
    even chance events have regularities. But this seems to presuppose that the
    remaining category (design) can't also exhibit regularities, which seems to
    me like an unwarranted assumption. In fact, the only sensible name that I
    can think of is "not designed"!

    So, it seems that if the Explanatory Filter says anything at all, it amounts
    to the following: once we've eliminated all the possible "not designed"
    explanations, we must conclude that the event is due to design. In other
    words, it's a tautology!

    The Inflationary Fallacy
    -------------------------------
    Although I said I wasn't going to discuss Dembski's Law of Small Probability
    in detail, I'd like to address one issue related to it, not because it's an

    important one, but just because it interests me.

    In justifying his "universal probability bound", Dembski argues that he
    doesn't need to allow for the probabilistic resources of the multiple
    universes which, according to some physicists, result from inflationary
    big-bang cosmology or quantum mechanics. He writes: "...there is something
    deeply unsatisfying about positing these entities simply because chance
    requires it" (p. 215). It's rather parochial of Dembski to assume that
    physicists have proposed these theories just to undermine probabilistic
    arguments of the sort he wants to make. I'm sure that's not the case. And,
    if these other universes really do exist, then we must face up to the
    implications.

    While not accepting the possibility that such universes exist, Dembski
    attempts to argue that, if they did, they would make the concept of chance
    unintelligible: "But with unlimited probabilistic resources, we lose any
    rational basis for eliminating chance". Leaving aside the question of
    whether these multiple universe theories necessarily entail an *infinity* of
    universes, this is an interesting point, and I think it betrays Dembski's
    agenda. There would indeed, no longer be any rational basis for rejecting a
    chance explanation of the origin of intelligent life (Dembski's aim). No
    matter how infinitesimally small the probability of life might be, it would
    occur in an infinitesimal proportion of universes, and we wouldn't be
    surprised to find ourselves in such a universe, because, we couldn't
    possibly find ourselves in any of the others.

    However, the same argument does not apply when we consider other chance
    events which are not vital to our very existence. To take Dembski's example:
    "Despite a life of unremitting charity and self-sacrifice, did Mother Teresa
    in her last days experience a neurological accident that caused her to
    become an ax murderer?" Well, if we assume that such an event had
    infinitesimal but non-zero probability, then, yes, there will be universes
    where that happened. But there's no particular reason why we should find
    ourselves in one of those universes. Therefore we have every right to be
    surprised, nay astounded, if our own Mother Theresa was revealed to be an ax
    murderer, and to reject the chance hypothesis. It follows that there will be
    some *some* universes in which the chance hypothesis will be wrongly
    rejected, but the probability of that happening in *our* universe is
    infinitesimal.

    In short, I think Dembski is wrong to exclude multiple universes in
    principle. However, I for one would also find it deeply unsatisfying if the
    naturalistic explanation for life had to resort to multiple universes,
    unless the arguments for those multiple universes were unimpeachable (of
    which I'm yet to be convinced).

    Is There Design in Nature?
    -------------------------------------
    Interestingly, Dembski doesn't address this question in TDI, although it's
    clear that that's what the book is leading up to, and some of his supporters
    claim that it does indeed do so. For example, in "Scientists Find Evidence
    of God" (http://www.arn.org/docs/insight499.htm) by Stephen Goode we find:

    "Dembski recently published his own addition to the ever-growing Intelligent
    Design Movement, a closely argued book that he calls The Design Inference,
    in which Dembski (whose impressive list of degrees led one friend to
    describe him as "the perpetual student") brings to bear his knowledge of
    symbolic logic and mathematics to argue in favor of design in nature."

    I suspect that many rank-and-file creationists are laying out their
    hard-earned cash for this book in the expectation of finding an argument for
    ID in nature. If so, they're wasting their money, because what they're
    actually getting is mostly a technical treatise on statistics, which, valid
    or not, is going to be of interest to very few people. By the way, here in
    Britain the book costs £40, about 4 times the cost of a typical popular
    science book. That's quite a lot of money to waste if it isn't what you
    wanted!

    Anyway, given that Dembski doesn't attempt to apply the Explanatory Filter
    to nature in this book, does he do it anywhere else? Well, I haven't been
    able to find an application of the Explanatory Filter as such, but I've
    found some on-line articles in which Dembski uses some related concepts
    named "actual specified complexity" and "complex specified information"
    (CSI). As far as I can tell, these two terms are synonymous with each other,
    and a phenomenon is considered to possess these attributes if it results
    from a specified event of small probability.

    So what does Dembski say about actual specified complexity? Well, nothing
    very definitive:

    "Does 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." Explaining
    Specified Complexity
    (http://www3.cls.org/~welsberr/evobio/evc/ae/dembski_wa/19990913_explaining_
    csi.html).

    Unfortunately, it seems that some of Dembski's followers haven't heard that
    the jury is still out:

    "Drawing upon design-detection techniques in such diverse fields as forensic
    science, artificial intelligence, cryptography, SETI, intellectual property
    law, random number generation, and so forth, William Dembski argues that
    specified complexity is a sufficient condition of inferring design and that,
    moreover, such specified complexity is evident in nature." William Lane
    Craig, (http://www.leaderu.com/offices/dembski/docs/bd-dibook.html).

    Just one more point. Dembski use specified complexity as a measure of
    improbability. But, as he himself says: "...probabilities are always
    assigned in relation to chance hypotheses". So it's misleading to refer to
    specified complexity as a singular measure. A phenomenon has a separate
    measure of specified complexity for each of the possible chance hypotheses
    that could explain it.

    Well, that will do for now. Comments would be welcomed!

    Richard Wein (Tich)
    "The most formidable weapon against errors of every kind is reason. I have
    never used any other, and I trust I never shall." -- Thomas Paine, "Age of
    Reason"



    This archive was generated by hypermail 2b29 : Sun May 14 2000 - 11:33:56 EDT