Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!ndcheg!uceng!dmocsny
From: dmocsny@uceng.UC.EDU (daniel mocsny)
Newsgroups: comp.ai
Subject: Re: Question on Chinese Room Argument
Summary: Artifacts, Complexity Engineering
Message-ID: <847@uceng.UC.EDU>
Date: 9 Apr 89 15:23:16 GMT
References: <10992@bcsaic.UUCP> <16872@cup.portal.com>
Organization: Univ. of Cincinnati, College of Engg.
Lines: 47

In article <16872@cup.portal.com>, dan-hankins@cup.portal.com 
(Daniel B Hankins) debates the nature of "artifact" with
gilbert@cs.glasgow.ac.uk (Gilbert Cockton).

Gilbert Cockton writes:
> >Given a well-understood task, computer programs will out-perform humans. 
> >Given a poorly understood task, they will look almost as silly as the
> >author of the abortive program.

Daniel Hankins ripostes with examples of man-made systems that can
solve poorly-understood problems.

The discussion reminds me of Stephen Wolfram's discussion in his
paper "Approaches to Complexity Engineering," which appeared in
Physica D, 1985. In this paper, Wolfram contrasted traditional
engineering design with the emerging art of complexity engineering.
In a traditional artifact, the engineer proceeds from a detailed
logical description of every part of a system and all its behaviors.
The system, if successful, does what it is designed to do---nothing
more and nothing less. The parts of the system interact with each
other in tightly constrained, often linear, ways. Motions are usually
periodic and synchronous. Failure in one part of the system often
causes catastrophic failure of the entire system. The system can
usually tolerate only a limited degree of environmental change.

A complex system, on the other hand, consists of a large collection of
individually simple parts, each having only a limited repertoire of
possible behaviors. Each part interacts with its neighbors according
to a fairly short list of typically nonlinear transition rules. The
system as a whole exhibits enormously complex emergent behaviors,
which the "designer" usually cannot predict in detail (since the
system is computationally irreducible). The complex system can
potentially exhibit other desired properties---e.g., robustness and
adaptiveness. The trick in complexity engineering, of course, is to
select the transition rules that yield the desired emergent behaviors.

We are only just beginning to learn how to do this. If we succeed,
then we may be able to take a huge chunk out of the "Logical
Specification Problem." I.e., our limited ability to comprehend and
manipulate lengthy logical specifications greatly restricts the
complexity of our traditionally-engineered artifacts. The logical
specification for a complex system of the type in the above paragraph
is quite short in comparison to the behavior obtained. This is more
in keeping with our ability to design things.

Dan Mocsny
dmocsny@uceng.uc.edu