Loading ...
Sorry, an error occurred while loading the content.
 

ID-Commentary: "Why Evolutionary Algorithms Cannot Generate Specified Complexity"

Expand Messages
  • Morgan Grey
    I appologize for the long delay in posting this, but the last couple of weeks have been rather hectic. This is my second commentary on ID-texts. This week, I m
    Message 1 of 1 , Oct 2, 2001
      I appologize for the long delay in posting this, but
      the last couple of weeks have been rather hectic.

      This is my second commentary on ID-texts. This week,
      I'm commenting on Dembski's "Why Evolutionary
      Algorithms Cannot Generate Specified Complexity",
      Dembski's second post on why evolutionary algorithms
      can't explain specified complexity, from
      <http://www.metanexus.org/archives/message_fs.asp?&listtype=Magazine&ARCHIVEID=3080>
      (also online at
      <http://www.geocities.com/evolutionsteori/3080.html>).

      My last such post can be found at
      <http://www.geocities.com/evolutionsteori/IDC-001.html>.

      Again, note that the original Metaview contains some
      =20's and a few =9's, which I have removed. And again,
      readers are invited to check for themselves if I have
      correctly conveyed the original message by doing so.

      -------------------------------------------------------
      Metaviews 152: "Specified Complexity" by William
      Dembski
      grassie@... William Grassie
      Metaviews 152. 1999/11/1. Approximately 3575 words.

      BG> Below is another posting from William Dembski at
      BG> Baylor University in Waco, TX. Dembski
      BG> continues his discussion of evolutionary
      BG> algorithms (see Meta 139) and presents a
      BG> mathematical argument for why such algorithms
      BG> cannot generate specified complexity

      Dembski does no such thing. The points of critique
      launched in this essay are limited to:

      1) Evolutionary algorithms always solve their
      problems, setting the probability of success at 1, and
      the complexity therefore at 0.

      2) Evolutionary algorithms get their "specified
      complexity" from the fitness functions, and thus have
      not *created* it.

      None of these are supported by any kind of
      "mathematics", unless one considers any essay with
      numbers in it to be "mathematical".

      BG> as asserted by Richard Dawkins.

      Dembski here continues the practice I also critiqued
      in my previous installment of ID-Commentary: Namely,
      to only criticize Dawkins' "misleading" Weasel
      program, instead of dealing with *real* problems
      solved by evolutionary algorithms, as asked by critics
      of Dembski. This is especially suspect, since "Why
      Evolutionary Algorithms Cannot Generate Specified
      Complexity" (and its companion-piece "Explaining
      Specified Complexity") is being presented as an *in
      principle*-refutation of the possibility of
      evolutionary algorithms producing "specified
      complexity".

      When speaking to the general public, who only know
      Dawkins' Weasel program, this tactic might work very
      well, but leave more informed skeptics wondering why
      Dembski keeps avoiding the *real* challenges, if his
      "explanatory filter" really is capable of doing what
      has been attributed to it.

      BG> A number of equations are presented in the
      BG> appendix.
      BG>
      BG> Dembski concludes that "all the specified
      BG> complexity we get out of an evolutionary algorithm

      BG> has first to be put into the construction of the
      BG> evolutionary algorithm and into the fitness
      BG> function that guides the algorithm. Evolutionary
      BG> algorithms therefore do not generate or create
      BG> specified complexity, but merely harness already
      BG> existing specified complexity." I am not sure I
      BG> follow the entire argument,

      Bill Grassie's confusion is understandable, since
      Dembski has a wonderful ability to cloak everything he
      says in a highly techincal and intimidating babble.
      Therefore, most of my comments will deal with what
      Dembski is actually *saying*, trying to "translate"
      his impressive-sounding lingo, showing that it often
      covers simple and uncontroversial statements.

      BG> but I am certainly reminded of my first
      BG> programming course as a freshman in college in
      BG> 1975, when I clocked 70 hours one week in the lab
      BG> trying to code a quick sort algorithm. Some more
      BG> teleological interventions would have helped.
      BG>
      BG> I will entertain responses on the Metaviews list
      BG> and try to run some compilation in a week or so.
      BG> If you want immediate gratification conversation,
      BG> check out the Reiterations List at for a higher
      BG> volume, lightly moderated discussion.
      BG>
      BG> -- Billy Grassie
      BG>
      WAD> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
      =-=-=-=-=-
      WAD> =-=-= From: bill@... (William A.
      WAD> Dembski) Subject: Specified Complexity
      WAD>
      WAD> WHY EVOLUTIONARY ALGORITHMS CANNOT GENERATE
      WAD> SPECIFIED COMPLEXITY by William A. Dembski
      WAD>
      WAD> In my last piece for META, I asserted that
      WAD> evolutionary algorithms cannot generate specified

      WAD> complexity and motivated this assertion by
      WAD> pointing to the failure of Richard Dawkins's
      well-
      WAD> known METHINKS IT IS LIKE A WEASEL example to
      WAD> generate specified complexity. My point was that
      WAD> Dawkins's evolutionary algorithm converged on
      WAD> METHINKS IT IS LIKE A WEASEL with probability
      WAD> one, and therefore reduced the complexity of
      WAD> generating this sequence to zero. With reference
      WAD> to specified complexity, complexity and
      WAD> probability are inverse notions: High complexity
      WAD> presupposes many live possibilities and
      WAD> correspondingly assigns low probability to anyone

      WAD> of these possibilities. Thus, while it's true
      WAD> that shaking out random scrabble pieces would
      WAD> render METHINKS IT IS LIKE A WEASEL highly
      WAD> improbable (and therefore complex), Dawkins's
      WAD> evolutionary algorithm renders that sequence
      WAD> certain and thereby removes its complexity.
      WAD>
      WAD> Basically, the problem here is one of setting the

      WAD> relevant probabilistic context. Within a random-
      WAD> scrabble-shaking-scenario, this sequence is
      WAD> complex and specified, but within Dawkins's
      WAD> evolutionary algorithm it is no longer complex
      WAD> (though it remains specified). I therefore
      WAD> concluded my last piece by saying that just as
      WAD> Darwinian evolution only delivers the
      WAD> **appearance** of design (an assertion all
      WAD> Darwinists perforce accept), so too it only
      WAD> delivers the **appearance** of specified
      WAD> complexity.

      Dembski forgets the other half of his conlusion: That
      his actual/appearant split of "specified complexity"
      makes it considerably more difficult to determine
      whether life indeed *is* an instance of specified
      complexity:

      "Does nature exhibit actual specified
      complexity? The jury is still out." (Dembski,
      in Meta #3066)

      [...]
      WAD> In general terms, the problem of generating
      WAD> specified complexity via an evolutionary
      WAD> algorithm can be conceived as follows. We are
      WAD> given a phase space of possible solutions to a
      WAD> problem and a fitness function over that phase
      WAD> space. Our task is to optimize this fitness
      WAD> function by finding a point in the phase space
      WAD> that attains a certain level of fitness. Think of

      WAD> it this way: The phase space is a vast plane, the

      WAD> fitness function is a vast hollowed-out mountain-
      WAD> range over the plane (complete with low-lying
      WAD> foothills and incredibly high peaks). The task of

      WAD> an evolutionary algorithm is by moving around in
      WAD> the plane to get to some point under the
      mountain-
      WAD> range where it attains at least a certain height
      WAD> (e.g., 10,000 feet). The collection of all such
      WAD> places on the plane where the mountain range
      WAD> attains at least that height (here 10,000 feet)
      WAD> we will call the **target**. Thus the job of the
      WAD> evolutionary algorithm is by navigating the phase

      WAD> space to find its way into the target (see
      WAD> Appendix 1 below).

      What Dembski here calls the "phase space" is already
      known to readers of Dawkins as "genetic space":

      "Imagine a museum with gallaries stretching
      towards the horizon in every direction, and as
      far as the eye can see upwards and downwards
      as well. Preserved in the museum is every kind
      of animal form that has ever existed, and
      every kind that could be imagined. Each animal
      is housed next door to those it most
      resembles. Each dimension in the museum -that
      is, each dimension along which a gallary
      extends- corresponds to one dimension in which
      the animals vary. For example, as you walk
      north along a particular gallary you notice a
      progressive lengthening of the horns of the
      speciments in the cabinets. Turn round and
      walk south and the horns shorten. Turn and
      walk east and that horns stay the same but
      something else changes, say the teeth get
      sharper. Walk west and the teeth grow blunter.
      Since horn length and teeth sharpness are only
      two out of thousands of ways in which animals
      can vary, the gallaries must criss-cross one
      another in many-dimensional space, not just
      the ordinary three-dimensional space that we,
      with our limited minds, are capable of
      visualizing." (Dawkins, R., 1996, "Climbing
      Mount Improbable", pp. 200)

      In the case of Dawkins' weasel program, the "phase
      space" is 28-dimensional (since there are 28 positions
      that can vary), where each dimension is 27 characters
      long (since there are 26 letters + space). In the
      weasel program, the algorithm can move any numbers of
      characters, but is restricted to moving a certain
      number of dimensions at a time (kinda' like the tower
      in chess, which can only move either back-forth or
      left-right, but can move any number of spaces).

      WAD> Now, the phase space (which we are picturing as a

      WAD> giant plane) usually comes with some additional
      WAD> topological structure, typically given by a
      WAD> metric or distance function (see Appendix 2).
      WAD> This topological structure tells us how points in

      WAD> the phase space are related geometrically to
      WAD> nearby points.

      The concept Dembski is trying to convey is that known
      to the biological community as a "fitness landscape",
      where increasing altitude stands for increasing
      fitness, as defined in terms of reproductive sucess.

      In the case of Dawkins' weasel program, the fitness
      landscape is a 29-dimensional cone, placed "over" the
      28-dimensional "chessboard" (a.k.a. "phase space"). On
      the space directly under the top of the cone, the
      target sequence ("methinks it is like a weasel") is
      written, while the spaces around it are labelled with
      sequences very close to it (e.g. "yethinks it is like
      a weasel" and "methinks it is like a geasel").

      WAD> Also, even though the phase space is huge, it
      WAD> tends to be finite (strictly finite for problems
      WAD> represented on computer and topologically finite,

      WAD> or what topologists call "compact," in general).

      This is quite uncontroversial. The number of possible
      28-letter sequences, each position with 27 possible
      outcomes (28^27 ~ 10^39) *is* "huge, [but] finite".

      WAD> Moreover, such spaces typically come with a
      WAD> uniform probability that is adapted to the
      WAD> topology of the phase space (see Appendix 3).

      With respect to Dawkins' weasel program, this pretty
      much means that the very first sequence has no more
      probability coming up "jhdhonfybyyeev nzyvqqtiilke"
      than "xgyhsnszciuhanomqtwlpgwaaumu". I know of no
      algorithms, where this is not the case. Dembski's
      reason for mentioning this is unclear.

      WAD> Basically this means that if you get out your
      WAD> tape measure and measure off a three by five foot

      WAD> area in one part of the phase space, the uniform
      WAD> probability will assign it the same probability
      WAD> as a three by five foot area in another portion
      WAD> of the phase space. All the spaces to which I've
      WAD> seen evolutionary algorithms applied do indeed
      WAD> satisfy these two conditions of having a finite
      WAD> topological structure (i.e., they are compact)
      WAD> and possessing a uniform probability. Moreover,
      WAD> this uniform probability is what typically gets
      WAD> used to estimate the complexity/improbability of
      WAD> the target (i.e., the area of the phase space
      WAD> under the mountain range where it attains a
      WAD> certain requisite level -- e.g., 10,000 feet).
      WAD>
      WAD> For instance, in Dawkins's
      WAD> METHINKS*IT*IS*LIKE*A*WEASEL example, the phase
      WAD> space consists of strings of upper case Roman
      WAD> letters and spaces (represented by asterisks) of
      WAD> length 28. A uniform probability on this space
      WAD> assigns equal probability to each of these
      WAD> sequences -- approximately 1 in 10^40. It's this
      WAD> improbability that corresponds to the complexity
      WAD> of the target sequence and with respect to which
      WAD> this target sequence constitutes an instance of
      WAD> specified complexity.

      Again, Dembski is being very unclear about what
      *exactly* he means by "specified complexity". Judging
      from the above, one would think that the "complexity"
      (i.e. "propability") of a certain event should be
      calculated only with respect to a single chance
      hypothesis. But in TDI (pp. 50) Dembski says that his
      explanatory filter needs to "sweep the field clear of"
      *all* chance hypotheses.

      This wouldn't be much of a problem, since "specified
      complexity" is never even mentioned in TDI. But since
      Dembski is constantly referring to specified
      complexity as a characteristic feature of "design", as
      well as to TDI as his "scholarly argument" for his
      ideas, this is unlikely to be anything *but* a
      problem.

      WAD> In general, given a phase space with a target
      WAD> sitting under those places where the mountain
      WAD> range attains at least a certain level (e.g.,
      WAD> 10,000 feet), the (uniform) probability of
      WAD> randomly choosing a point from the phase space
      WAD> and landing in the target will be very small. In
      WAD> Dawkins's example, the target equals the
      WAD> character string METHINKS*IT*IS*LIKE*A*WEASEL and

      WAD> the improbability is 1 in 10^40. For non-toy
      WAD> examples the improbability is typically much less

      WAD> than my universal probability bound of 1 in
      WAD> 10^150 that I justify in The Design Inference
      WAD> (Cambridge, 1998; cf. section 6.5). Indeed, if
      WAD> the probability of the target were not small, a
      WAD> random search through the phase space would
      WAD> suffice to find a point in the target, and there
      WAD> would be no need to construct an evolutionary
      WAD> algorithm to find it.

      Again, few people would disagree that "methinks it is
      like a weasel" is too long to find just by randomly
      selecting letters and spaces. Indeed, Dawkins himself
      concluded that it "would be a long time coming" before
      this would produce the target sequence ("The Blind
      Watchmaker", pp. 47).

      WAD> We therefore suppose that the target is just a
      WAD> tiny portion of the whole phase space; or, in
      WAD> slightly more technical language, the (uniform)
      WAD> probability of the target in relation to the
      WAD> phase space as a whole is exceedingly small.
      WAD> What's more, the target, in virtue of its
      WAD> explicit identification, is specified (certainly
      WAD> this is the case in Dawkins's example where the
      WAD> target includes but one point and coincides with
      WAD> the character string
      WAD> METHINKS*IT*IS*LIKE*A*WEASEL). Thus it would seem

      WAD> that to find a point in the target would be to
      WAD> generate specified complexity.

      But just as Morris and Whitcomb thinks that
      radiometric datings only show the "appearent age" of
      the Earth, so Dembski believes that the solution
      produced is only "appearant specified complexity".

      WAD> But let's look deeper. Consider an evolutionary
      WAD> algorithm that does in fact find the target. An
      WAD> evolutionary algorithm can be conceived as a
      WAD> stochastic process that moves around the phase
      WAD> space some finite number of times (see Appendix
      WAD> 4). Let's call the evolutionary algorithm E. The
      WAD> evolutionary algorithm starts at some point E(0)
      WAD> in the phase space (usually chosen at random).
      WAD> Then it moves to E(1). Then to E(2). Then to E
      WAD> (3). Etc. For E successfully to find the target
      WAD> (i.e., to find a point under the mountain range
      WAD> where it attains at least a certain level --
      WAD> e.g., 10,000 feet) then means that within a
      WAD> manageable number of steps n, E is very likely to

      WAD> land in the target -- i.e., some one of E(0), E
      WAD> (1), ..., E(n) is likely to land in the target
      WAD> (see Appendix 5). Simply put, the algorithm E has

      WAD> to get us into the target with high probability
      WAD> and in a relatively short number of steps. In the

      WAD> Dawkins example, E(n) rapidly converged to
      WAD> METHINKS*IT*IS*LIKE*A*WEASEL for n around 40.
      WAD>
      WAD> An evolutionary algorithm needs to be contrasted
      WAD> with pure random sampling. Pure random sampling
      WAD> treats the phase space as a giant urn from which
      WAD> we draw items at random according to the uniform
      WAD> probability. In that case, a random sample from M

      WAD> of size k will contain a point in the target with

      WAD> probability better than 1/2 provided that k is
      WAD> around the reciprocal of the (uniform)
      WAD> probability of the target. Since we are assuming
      WAD> that the probability of the target is less than
      WAD> my universal probability bound of 1 in 10^150
      WAD> given earlier, it follows that k will need to be
      WAD> at least 10^150. This number is enormous and far
      WAD> exceeds the number of computations conceivable
      WAD> for any traditional computer. Moreover, it
      WAD> doesn't seem that quantum computation is going to

      WAD> render this number tractable either since the
      WAD> points in phase space need to be made explicit in

      WAD> any random sampling scheme (implying decoherence
      WAD> and thus preventing us from exploiting quantum
      WAD> superposition).

      Since all of the above is the case, both with respect
      to Dawkins' weasel program, as well as all examples of
      evolutionary algorithms that I am aware of, I am
      puzzled as to why Dembski finds it relevant to
      mention.

      WAD> Let's now return to the evolutionary algorithm E.

      WAD> We're going to allow ourselves a certain number
      WAD> of steps, call it m, for E to land in the target.

      WAD> Clearly m is going to have to be much less than
      WAD> 10^150 if we're going to program E on a computer
      WAD> and have any hope of E landing in the target.
      WAD> With m fixed, we can determine the probability
      WAD> that E will land in any subset of phase space in
      WAD> m steps (see Appendix 6). For instance, in the
      WAD> Dawkins example, for m = 100 and the target
      WAD> sequence METHINKS*IT*IS*LIKE*A*WEASEL and E the
      WAD> cumulative selection algorithm Dawkins
      WAD> constructed, the probability of E attaining the
      WAD> target in m = 100 steps is approximately 1.
      WAD>
      WAD> What this means is that even though with respect
      WAD> to the uniform probability on the phase space the

      WAD> target has exceedingly small probability, the
      WAD> probability for the evolutionary algorithm E to
      WAD> get into the target in m steps is no longer
      WAD> small. And since complexity and improbability are

      WAD> for the purposes of specified complexity parallel

      WAD> notions, this means that even though the target
      WAD> is complex and specified with respect to the
      WAD> uniform probability on the phase space, it
      WAD> remains specified but is no longer complex with
      WAD> respect to the probability induced by
      WAD> evolutionary algorithm E.

      Now Demsbki seems to have returned to claiming that
      complexity needs to be calculated with regard to *all*
      relevant chance hypotheses (as opposed to just the
      "uniform probability").

      While few would disagree that life is complex with
      regard to the chance hypothesis of it being assembled
      by throwing random molecules together, it is quite
      another matter if it is complex with regard to it
      having come about through the actualization of
      heritable modifications, exclusion of certain
      modifications through differental reproductive
      success, and specified through the conditions of the
      environment (i.e. natural selection). In fact, whether
      this is so is the very point in question, and IDers
      are just assuming their conclusion when they claim
      that life contains "specified complexity".

      WAD> Does this mean that the evolutionary algorithm
      WAD> has in fact generated complex specified
      WAD> information, but that in referring to a loss of
      WAD> complexity with respect to E I'm simply engaging
      WAD> in some fancy redefinitions to avoid this
      WAD> conclusion? I don't think so. Remember that we
      WAD> are interested in the **generation** of specified

      WAD> complexity and not in its reshuffling.

      This seems to be a complete non sequitur. Dembski
      hasn't shown that the "specified complextiy" has been
      "reshuffled", and his "reminding us of it" seems
      obscure. Indeed, Dembski doesn't even think that there
      is any specified complexity to be "reshuffled" to
      begin with! According to his argument, the sequence
      produced by Dawkins' weasel program doesn't contain
      specified complexity because it is produced by
      Dawkins' weasel program.

      And contrary to Dembski's assertions, he *is* "simply
      engaging in some fancy redefinitions", if only with
      respect to claims that life contains "specified
      complexity".

      WAD> To see what's at stake here, we need to be clear
      WAD> about a restriction that needs to be placed on E
      WAD> if it is to count as a genuine evolutionary
      WAD> algorithm (i.e., a legitimate correlative of the
      WAD> Darwinian mutation-selection mechanism). It is
      WAD> not, for instance, legitimate for E to be able to

      WAD> survey the mountain range (i.e., fitness
      WAD> landscape), see where in the phase space it
      WAD> attains a global maximum, and then head in that
      WAD> direction. That would be teleology. No, E must be

      WAD> able to navigate its way to the target either by
      WAD> randomly choosing points from the phase space or
      WAD> by using those as starting points and then
      WAD> selecting other points in the phase space based
      WAD> **solely** on the topology of the phase space and

      WAD> without recourse to the fitness function, except
      WAD> to evaluate the fitness function at individual
      WAD> points of the phase space already traversed by E.

      WAD> In other words, E must move around the phase
      WAD> space only on the basis of its topology and the
      WAD> elevation of the fitness function at points in
      WAD> the phase space already traversed by E.

      Of course not! Again, since this doesn't apply to any
      of the evolutionary algorithms that Dembski is
      supposed to deal with, I am at a loss, trying to
      understand why Dembski considers it relevant to
      mention.

      [...]
      WAD> Certainly this means that the evolutionary
      WAD> algorithm E is highly constrained in the use it
      WAD> can make of the fitness function. But there's
      WAD> more. It means that the success of E in hitting
      WAD> the target depends crucially on the structure of
      WAD> the fitness function.

      Finally, Dembski seems to have arrived at his major
      criticism of evolutionary algorithms as producers of
      specified complexity: They don't produce specified
      complexity, but gets it from the fitness function.

      WAD> If, for instance, the fitness function is totally

      WAD> flat and close to the ground whenever it is
      WAD> outside the target, then it fails to discriminate

      WAD> between points outside the target and so cannot
      WAD> be any help guiding an evolutionary algorithm
      WAD> into the target. For such a fitness function, the

      WAD> probability of the evolutionary algorithm landing

      WAD> in the target is no better than the probability
      WAD> of pure random sampling landing in the target,
      WAD> which as we know is inadequate to get us there
      WAD> (see Appendix 7).
      WAD>
      WAD> But the problem is even worse. It follows by a
      WAD> combinatorial argument that for any partition of
      WAD> the phase space into pieces none of which has
      WAD> probability more than the probability of the
      WAD> target (which by assumption is less than 1 in
      WAD> 10^150), for the vast majority of these partition

      WAD> elements the probability of the evolutionary
      WAD> algorithm E entering them is going to be no
      WAD> better than pure random sampling. It follows that

      WAD> the vast majority of fitness functions on the
      WAD> phase space that coincide with our original
      WAD> fitness function on the target but reshuffle the
      WAD> function on the partition elements outside the
      WAD> target will not land the evolutionary algorithm
      WAD> in the target (this result is essentially a
      WAD> corollary of the No Free Lunch theorems by
      WAD> Wolpert and Macready).

      As I also pointed out, last week, Dembski's (mis)use
      of Wolpert and Macready's "No Free Lunch theorems" is
      bordering on the intellectually dishonest. According
      to Wesley, "NFL isn't about essential capacity of an
      algorithm to produce a solution; it is about
      comparative efficiency of algorithms in producing
      solutions." (see my last Commentary)

      WAD> Simply put, the vast majority of fitness
      WAD> functions will not guide E into the target even
      WAD> if they coincide with our original fitness
      WAD> function on the target (see Appendix 8).

      In order to put Dembski's objection into perspective,
      allow me to use a specific example: An evolutionary
      biologist might claim that a certain rodent can evolve
      longer teeth, if having longer teeth confers a
      reproductive advantage: Mutations for longer teeth
      appear and are selected for, increasing the specified
      complexity of the genome of the offspring of the
      rodent (if only with respect to the "uniform
      probability").

      Dembski's hypothetical response to this would be that,
      Yes, natural selection indeed *can* enlarge the teeth
      of rodents, thereby increasing the specified
      complexity (with respect to the "uniform probability")
      of its genome. But since it depends on longer teeth
      confering a reproductive advantage, this specified
      complexity hasn't really been created, only
      "reshuffled".

      The creative act of the Intelligent Designer would in
      this case be to determine that having longer teeth
      would cause the rodent in question to have more
      offspring.

      One wonders if this is the same Dembski who wrote that
      "design ... located in natural laws ... becomes an
      empty metaphor":

      "But as soon as design is located in natural
      laws, design becomes an empty metaphor. I know
      what it is for a watch to be designed. I only
      know what it is for the *process* of making a
      watch to be designed in the derivative sense
      that I know what it is for a watch to be
      designed. Locating design in natural laws has
      the effect of reversing this ordinary logic
      and thereby vitiating design. If I can't
      ascertain that a thing is designed, I can't
      ascertain that that the process giving rise to
      the thing is designed. Unless we can infer an
      intelligent agent from the structure, dynamics
      and function of *things*, we are not going to
      infer such an agent from the *processes* that
      agent supposedly used to bring about those
      things. If imputing design to things is
      problematic, then imputing design to the
      processes that gave rise to those things is
      doubly problematic." (Dembski, W.A.,
      1999, "Intelligent Design: The Bridge Between
      Science and Theology", pp. 78, original
      emphasis)

      This internal inconsistency on Dembski's part
      notwithstanding, his objection suffers from serious
      problems.

      First of all, it is clear that the role of natural
      selection in the production of specified complexity
      still looms large enough to call into question the
      sweeping claims made about Dembski's explanatory
      filter having reinstated God within science. Even if
      the objection of "Why Evolutionary Algorithms Cannot
      Generate Specified Complexity" and the claim that life
      contained specified complexity were to be taken at
      face value, Dembski's explanatory filter would at most
      allow for some sort of deist-god, creating the
      universe with the physical constants and mechanisms
      that would make it possible for life to evolve by
      purely natural processes. This might be what Dembski
      means by "designed", but I doubt that the evangical
      Christian community buying his books or in other ways
      supporting the ID movement financially will like what
      Dembski is saying.

      Second, and more importantly, Dembski's objection is
      difficult (if not impossible!) to test. At the moment,
      we have no idea what causes the natural constants to
      be the way they are, and thus, any hope of putting
      fitness functions into Dembski's explanatory filter is
      a far way into the future. And if those doesn't happen
      to be designed either, Dembski can just claim that
      whatever caused *them* to be that way must be
      designed.

      Indeed, there is a problem of infinite regress here.
      Whenever the source of whatever feature in question is
      offered, Dembski can just lean back and ask "Well, how
      did *that* come about?" And since one must always
      produce a new source to satisfy him, Dembski can
      continue playing this game for as long as he wants (or
      until his opponents grow tired of playing with him).

      The third problem with Dembski's objection flows
      naturally from the second. Earlier this year, Dembski
      claimed that "[i]f it could be shown that biological
      systems like the bacterial flagellum that are
      wonderfully complex, elegant, and integrated could
      have been formed by a gradual Darwinian process (which
      by definition is non-telic), then intelligent design
      would be falsified on the general grounds that one
      doesn't invoke intelligent causes when purely natural
      causes will do."
      (<http://www.discovery.org/viewDB/index.php3?program=CRSC%20Responses&command=view&id=584>)

      It could be argued that an unknown Designer, for
      whatever reasons, created certain features with an
      "appearant evolvability", and that this possible
      falsification would only falsify the design of
      biological structures anyway (leaving Ross' "fine
      tuning argument" safe and sound).

      But by assigning specified complexity to the fitness
      landscape, Dembski has effectively destroyed any hope
      of ID ever being falsifiable (at least with respect to
      "wonderfully complex, elegant, and integrated"
      "biological systems" being "formed by a gradual
      Darwinian process").

      Whenever natural selection (or, as in this case,
      evolutionary algorithms) is observed producing any of
      the things that, according to Dembski, was directly
      and supernaturally designed, he can just claim that it
      was all "hardwired" in the forces of nature.

      [...]
      WAD> I have omitted many details. I have also omitted
      WAD> some complications which to my mind make the
      WAD> problem of generating specified complexity via
      WAD> evolutionary algorithms even more problematic (in

      WAD> nature, for instance, the fitness function will
      WAD> not stay fixed but vary over time).

      Dembski needs to show why this is "problematic" (as
      oppposed to "easy" or "indifferent").

      WAD> Some of the details are treated in chapter 6 of
      WAD> my recently released Intelligent Design: The
      WAD> Bridge Between Science & Theology (InterVarsity).

      As already reported, "Intelligent Design" contains
      little, if any, new material. It merely repeats
      Dembski's assertion that any information "created" by
      any un-intelligent processes must be contained in the
      process from the start.

      WAD> A full treatment will have to await a book I'm
      WAD> currently writing (Redesigning Science: Why
      WAD> Specified Complexity Is a Reliable Empirical
      WAD> Marker of Actual Design).

      According to Nelson, this book was renamed to "No Free
      Lunch":

      From
      <http://www.calvin.edu/archive/evolution/200006/0123.html>:
      -------------------------------------------------------
      From: Paul Nelson (pnelson2@...)
      Date: Mon Jun 19 2000 - 21:22:58 EDT
      [...]
      Bill Dembski sent me the following note, saying he
      had tried to e-mail Wesley but the mail bounced
      back as undeliverable. Anyway, here's Bill's
      reply to Wesley's question about Bill's new book,
      "Redesigning Science":


      > It's about two-thirds completed. The working title
      now
      > is _No Free Lunch_. I'm in touch with a publisher.
      > I expect it will be out late 2001.
      >
      > --Bill

      [...]

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

      According to Amazon.com, it will be published November
      this year:

      From
      <http://www.amazon.com/exec/obidos/ASIN/0742512975/qid=1002053488/sr=2-1/ref=sr_8_5_1/103-7752700-2613463>:
      -------------------------------------------------------
      No Free Lunch : Why Specified Complexity Cannot Be
      Purchased Without Intelligence
      by William A. Dembski

      List Price: $35.00
      Our Price: $35.00

      This item will be published in November 2001. You may
      order it now and we will ship it to you when it
      arrives.

      Hardcover - 336 pages (November 2001)
      Rowman & Littlefield; ISBN: 0742512975
      [...]
      -------------------------------------------------------

      It will be interesting to see if *this* book contains
      the "in principle refutation" that so far have been
      lacking...

      WAD> But I want to make these preliminary results
      WAD> available because the misconception that one can
      WAD> purchase specified complexity on the cheap is
      WAD> widespread and ill-conceived.
      WAD>
      WAD> The only known generator of specified complexity
      WAD> that we know is intelligence.

      Given Dembski's stringent criteria for what can be
      considered to be "specified complexity", it is
      questionable if even the actions of intelligent
      entities can be considered to be manifestations of
      "specified complexity".

      WAD> Sans intelligence, a process that yields
      WAD> specified complexity merely converts already
      WAD> existing specified complexity.
      WAD> We are seeing a similar phenomenon with
      WAD> inflationary cosmologies, which attempt to wash
      WAD> out cosmological fine-tuning but invariably seem
      WAD> to smuggle it back in. Smuggling in specified
      WAD> complexity is not the same as **generating**
      WAD> specified complexity. I challenge the biological
      WAD> community to take these results seriously, and
      WAD> reevaluate how it understands the generation of
      WAD> specified complexity.
      [...]
      META> Permission is granted to reproduce this e-mail
      META> and distribute it without restriction with the
      META> inclusion of the following credit line: [This is

      META> another posting from the Meta-List . Copyright
      META> 1997, 1998, 1999. William Grassie.]
      -------------------------------------------------------




      =====
      Morgan

      "Evolution is to the social sciences as statues are to
      birds: a convenient platform upon which to deposit badly
      digested ideas." (Steve Jones, 2000, "Darwin's Ghost", pp.
      xxvii)

      __________________________________________________
      Do You Yahoo!?
      Listen to your Yahoo! Mail messages from any phone.
      http://phone.yahoo.com
    Your message has been successfully submitted and would be delivered to recipients shortly.