From: Wesley R. Elsberry <welsberr@orca.tamu.edu>
Newsgroups: talk.origins,comp.ai.genetic
Date: 1997/02/28
Message-ID: <5F5BLK$LC1@NEWS.TAMU.EDU>#1/1
Subject: Dembski & the TSP


William Dembski gave a talk last Saturday at the "Naturalism,
Theism, and the Scientific Enterprise" conference on intelligent
design in general, and the results of his information theoretic
analyses in particular. As expected, he claimed that functions,
algorithms, chance, and natural laws were incapable of generating
"complex specified information", whether considered seprately or
in combination.

I would like to thank everyone who sent me or posted possible
questions for Dembski. I collated my own points of difference
plus those into a handout which I had available that morning.
Several of the points brought up in these newsgroups were also
brought up in the discussion. Bill Jefferys asked about the
proof that Dembski offered concerning gradualness of information
acquisition under selection. I brought up genetic algorithms
as an empirical disproof.

In general, Dembski's response was to claim that his analysis
was both general and sound, and therefore his questioners could
not be correct.

In discussion of my presentation the day before, Dembski had
indicated that GAs find solutions to difficult problems because
an intelligent programmer had made them. By reference to Dembski's
own analysis, I took this to be a claim that the programmer had
somehow infused information about the particular solution that a
GA finds, and thus tainted the process.

I asked Dembski about a specific case, a 500 city Traveling Salesman
Problem. I asked where, precisely, the information about the
particular tour found was introduced by the programmer. I eliminated
the GA population handling routines, since those can be found in
general toolkits applicable to many problems, not just 500 city
TSPs. I eliminated the evaluation function, since by Dembski's
own analysis, no function can produce information.

Dembski responded that he objected to the notion that CSI can be
gotten on the cheap or for free, and said that the information
produced by a GA isn't untainted, since intelligence had to
design the hardware, the operating system, the GA program
itself, and the problem specification.

I don't believe the response during the discussion to have been
responsive to the point. Dembski's stated claim is that functions,
algorithms, chance, and natural laws cannot *ever* produce CSI.
Given an example of what by Dembski's standards is clearly CSI
resulting from the application of a GA, what I'm told is that
only algorithms which are made by an intelligence can produce
CSI. This is a wholly different claim, and much weaker.

The conference web site is

http://www.dla.utexas.edu/depts/philosophy/faculty/koons/ntse/ntse.html

--
Wesley R. Elsberry, 6070 Sea Isle, Galveston TX 77554. Central Neural System
BBS, 409-737-5222, 1:385/385, ANNs, GAs, Alife, AI, evolution, and more.
Student in Wildlife & Fisheries Sciences. http://www.rtis.com/nat/user/elsberry
"shakespeare and i are frequently coarse" - archy

[Other on-line articles]

[ID-Commentaries]