CSI, GAs, etc.

From: Paul Nelson (pnelson2@ix.netcom.com)
Date: Mon Oct 02 2000 - 11:54:06 EDT

  • Next message: FMAJ1019@aol.com: "Re: CSI, GAs, etc."

    Hi Wes,

    If you'd like to do an article on this topic for
    Origins & Design, jump right in. (Contact me
    off-list about length, et cetera.)

    I've got to cut out of this discussion because of
    upcoming lecture commitments. A few comments,
    however.

    When I spoke at the University of Colorado a couple
    of weeks ago, a bright undergraduate came up after
    the talk and said, "Dr. Nelson, you've just GOT to
    go on the net and play Conway's 'Game of Life' --
    that will answer all the questions you have about
    natural selection!" I listened as this young man
    described the remarkable, organismal-appearing
    patterns that arise from what he called "a few
    simple rules."

    Interesting, I replied. But then there's Conway.
    Right?

    The undergraduate was silent for a moment, and looked
    down at his feet. So I went on:

    All evolutionary algorithms that we know have at least
    one author, or intelligent designer. In the case of
    the Game of Life, for instance, that would be Conway.
    In many (all?) cases, the authors work hard writing
    code, and debugging that code, to ensure that their
    programs run and actually produce results.

    It's instructive to look at the history of evolutionary
    computation (EC) and genetic algorithms (GAs). Wagner
    and Altenberg (1996, p. 968) observe:

         Among the earliest experiments in evolutionary
         computation, Friedberg (1959) attempted to
         evolve functioning computer programs by mutating
         and selecting the code, but found that the
         mutations effectively randomized the behavior
         of the programs, and adaptive evolution was
         impossible. There is no way to improve the
         performance of a conventional computer program
         by randomly altering letters in the source
         code. It became understood that the mutation/
         selection process is not universally effective
         in producing adaptation if favorable mutations
         cannot be produced....In contrast to Friedberg's
         results, Koza (1992) succeeded in evolving
         computer programs that perform well on complex
         tasks (such as prediction of protein structure
         or random number generation) by recombining
         branches of parse trees for the programs.
         Ray (1992) succeeded in designing computer
         programs that exhibit evolution as an
         emergent property by careful design of the
         data structures.

    Note: "by careful design of the data structures."
    Wagner and Altenberg (1996, p. 968) continue:

         Hence, the Darwinian solution of optimization
         problems is possible if and only if the problem
         is "coded" in a way that makes the mutation-
         recombination-selection procedure an effective
         one.

    All known EC and GAs implicate at least one intelligent
    designer. Thus they cannot possibly provide sound
    counterexamples to the claim that CSI requires a
    designer, whether immediately or remotely.

    [Side comment. Unlike Wes, I find the issue of the
    designer's "distance" from CSI to be a non-starter.
    The central ID argument is that CSI requires a designer.
    If that's established, then we can look at the evidence
    to see (if possible) at what point, or by what means,
    the designer acted. Opinions differ in the ID
    community on this, as should be expected for any
    difficult empirical question.]
      
    Now, as a mechanism, natural selection can do nothing
    without the prior existence of organisms. As Wagner
    and Altenberg (1996, pp. 967-968) express the point:

         The "representation problem" is how to code a
         problem such that random variation and selection
         can lead to a solution. The representation
         problem underlies the issue of whether selection,
         mutation, and/or recombination can produce
         adaptation.

         For biology, the "representation problem" has
         some unsettling implications. If, as evolutionary
         biology asserts, all adaptations are the result
         of mutation and selection, organisms have to be
         evolvable. But once one calls into question the
         inevitability of organisms being evolvable, one
         can ask, how and why did an evolvable genome
         originate in the first place?

    Minimal complexity considerations (see, e.g., Hutchison
    et al. 1999) suggest that entities capable of heritable
    variation require a lot of specified complexity (CSI).
    But of course, we've already discovered that by our
    own design of GAs, et cetera. You're going to have to
    work very hard, and debug a lot of code, before the
    programs run. I can't think of a poorer analogue for
    the action of a blind watchmaker (Dawkins 1986)
    than the deliberate human design of EC and GAs.

    Back to work. One last comment, for those who
    want me to do CSI calculations for them:

    Sorry, life is short. I've already provided citations
    (Scherer 1983, 1996; Rust 1992) where such calculations
    have been done to my satisfaction. Pardon me if this
    sounds ad hominem, but I suspect it isn't CSI estimates,
    really, that you want.

    Paul Nelson
    Senior Fellow
    The Discovery Institute
    www.discovery.org/crsc

    Dawkins, Richard. 1986. _The Blind Watchmaker_.
       New York: W.W. Norton.

    Hutchison, Clyde A. 1999. Global Transposon Mutagenesis
       and a Minimal Mycoplasma Genome. _Science_ 286:2165-
       2169.

    Rust, Peter. 1992. How Has Life and Its Diversity Been
       Produced? _Perspectives on Science & Christian Faith_
       44:80-94.

    Scherer, S. 1983. Basic Functional States in the Evolution
       of Light-driven Cyclic Electron Transport. _Journal
       of Theoretical Biology_ 104:289-299.

    Scherer, S. 1996. _Entstehung der Photosynthese: Grenzen
       molekularer Evolution bei Bakterien? Berlin: Pascal
       Verlag, 96 pp.

    Wagner, Gunther and Altenberg, Lee. 1996. Complex
       Adaptations and the Evolution of Evolvability.
       _Evolution_ 50:967-976.

     



    This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 11:53:18 EDT