Dembski's universal small probability bound

From: Wesley R. Elsberry (welsberr@inia.cls.org)
Date: Sat Jun 17 2000 - 20:06:10 EDT

  • Next message: Paul Nelson: "Re: Question Regarding "Redesigning Science""

    Richard Wein wrote:

    RW>Wesley, since you've raised this subject again, I'd like to
    RW>reiterate my view that CSI, as Dembski defines it, has not
    RW>been shown to exist in nature.

    RW>Since we last discussed the subject, I've had it confirmed
    RW>to me that Dembski, in his book "Intelligent Design",
    RW>defines the test for CSI as:

    RW> -log2 P(E) > 500

    RW>where E is a specified event. (I haven't read the book
    RW>myself, but this is basically the same as the definition
    RW>Dembski gives in an on-line article.)

    I just got a copy of this book, and I would agree that much of
    the work can be seen in Dembski's on-line articles, modulo a
    few editing changes. (Expect more on that later.) But that
    is not the test for CSI. At best, that is a description of
    Dembski's universal small probability bound. CSI is tested as
    the conjunction of complexity and specification.

    RW>This means that the test for CSI is the same as the test
    RW>for design as defined in his earlier book "The Design
    RW>Inference", namely:

    RW> P(E) < 1/2 X 10^150

    Not quite. TDI's formulation of the universal small
    probability bound is better put as

      P(E|H) < 1/2 X 10^-150

    See pages 203-209 of TDI.

    The conditional probability is, IMO, significant.

    RW>But no-one to my knowledge has ever succeeded in showing
    RW>that the probability of a specified event in nature is this
    RW>small.

    A 100-city TSP solution meets the criterion of the universal
    small probability bound. If by "nature" Richard is referring
    to biological events exclusive of human action, then we at
    least have Dembski's assurance in "ID" that the bacterial
    flagellum exceeds this value. But I would agree that no
    detailed probability calculation demonstrates this.

    RW>I've seen some calculations by IDers which claim to
    RW>calculate the probability, for example, of a given protein
    RW>forming from a number of amino acids. But they assume that
    RW>the amino acids are selected as i.i.d. (identical
    RW>independently distributed) random variables. Of course,
    RW>this assumption is totally unrealistic, because evolution
    RW>by random mutation and natural selection would not produce
    RW>such a probability distribution (the amino acids are *not*
    RW>independent).

    This is where I see the conditional probability used in TDI
    come in. We compute the complexity of an event on the basis
    of how improbable the event is if it were due to chance.
    This does not mean that we actually believe it to be due to
    chance, or that we consider the chance hypothesis sufficient
    to the task. But like other forms of statistical inference,
    chance gives us a null hypothesis to perhaps reject.

    RW>So the demand we should be making of Dembski is not "show
    RW>that natural selection can't produce CSI", but "show that
    RW>CSI actually exists in nature".

    I think that CSI as a property is not that remarkable. But I
    agree that someone who urges others to "do the calculation"
    might break loose with a few examples that show *all* the
    work, just to get everybody started. Preferably, one or more
    completely worked examples for each of the possible
    termination nodes in Dembski's Explanatory Filter would be
    provided. At least one set of examples should be taken from
    biological science.

    Wesley



    This archive was generated by hypermail 2b29 : Mon Jun 19 2000 - 19:59:39 EDT