From: Marty Fouts <mathematician@usenet.nospam.fogey.com>
Newsgroups: talk.origins
Subject: Re: Dembski's Intelligent Design Hypothesis
Date: 1 Oct 1999 02:52:48 -0400
Organization: The University of Ediacara
Lines: 83
Message-ID: <wkg0zv36vw.fsf@usenet.nospam.fogey.com>
References: <37DAE85D.EAA0F817@erols.com> <37F03609.C3510BB@mpmd.com> <199909280601.XAA15302@cx33978-a.dt1.sdca.home.com> <37F28B4E.87B33630@mpmd.com> <199910010537.WAA51258@cx33978-a.dt1.sdca.home.com>
x-spamblock: the return address is valid as is
x-no-archive: yes
X-Newsreader: Gnus v5.6.45/Emacs 20.3
X-Trace: nntp1.ba.best.com 938761060 200 irk@206.184.162.220


[everything deleted in order to hijack thread.]

It would seem to me that there is a simple test of Dembski's thesis
that follows.  (With apologies to Turing.)

The _Intelligent Design_  double blind filter test.

1) Assume two sources of bit streams,  one stochastic, the other
   presenting examples of 'designed' streams.

2) Impose a requirement that either of the two sources would, upon
   query, produce a fixed length bit sequence possibly from a cache of
   already prepared sequences.  The length of the sequence would be a
   parameter but would remain constant during any one run of the
   test.

3) Allow the investigator to choose a suitable length for the
   sequences to be tested.  Call the length of a test sequence L.

4) Allow the investigator to choose a suitable number of samples to be
   included in the test.  Call the number of samples N.

5) Interpose between the investigator and the two sources a 'sampler',
   which would use a random mechanism for selecting which of the two
   sources to query for sequences.

The test then, would consist of the following protocol:

  A) The investigator begins by informing the sampler of the number of
     sequences required, N, and the length specified for each
     sequence, L.

  B) The sampler then informs each of the sources of the investigators
     choice, and waits until each tells the sampler that it is ready.
     The sources are arbitrarily labeled 0 and 1.

  C) The sampler then repeats, N times, the following protocol:

        i) start a timer.
       ii) randomly choose a source, remembering it.
      iii) request a sequence from the selected source.
       iv) wait for the sequence from the source.
        v) wait an additional length of time so that the total time
           for steps i through v is constant, within a small error bar.
       vi) present the sample to the investigator.
      vii) wait for the investigator to indicate that the sampler may
           continue.

  D)  The investigator then provides the sampler with an N bit
      sequence, each bit indicating whether the pattern is thought to
      be undesigned (= 0) or designed (= 1) [alternatively, the
      investigator could return 1 bit for each sample as part of
      acknowledging receipt of the sample.]

  E)  The sampler compares the sequence returned by the investigator
      to the sequence it remembered from its selection process and
      reports the correlation.[1]

The test would be even more interesting if, rather than 2, there are S
sources, an unknown number of which produce random sequences, while
the remainder use a variety of different designs to generate the
sequences.

I would suspect that one of the quantum noise random generator chips
would be a good source for the random string.  I would suggest *not*
using a pseudo-random number generator.

Prediction:  This experiment will show the strongest correlation to
simple designs.  The more complex the design, the smaller the
correlation.

No prediction is made about the sign of the correlation.

Marty

[1] You may pick your own among a plethora of statistical methods for
    evaluation the correlation here.  ;-)

-- 

				- 30 -


[Reproduced here by permission of the author per message of 
19991001 -- WRE]