From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: **
X-Spam-Status: No, score=2.1 required=5.0 tests=BAYES_05,INVALID_DATE,
	MSGID_SHORT,REPLYTO_WITHOUT_TO_CC autolearn=no autolearn_force=no
	version=3.4.4
Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!cmcl2!rutgers!rochester!cornell!uw-beaver!ssc-vax!bcsaic!ted
From: ted@bcsaic.UUCP (Ted Jardine)
Newsgroups: comp.lang.ada
Subject: Re: Adathink
Message-ID: <1844@bcsaic.UUCP>
Date: Tue, 11-Aug-87 13:55:59 EDT
Article-I.D.: bcsaic.1844
Posted: Tue Aug 11 13:55:59 1987
Date-Received: Fri, 14-Aug-87 04:27:24 EDT
References: <870806210856.08p@VLSI.JPL.NASA.GOV>
Reply-To: ted@bcsaic.UUCP (Ted Jardine)
Distribution: na
Organization: Boeing Computer Services AI Center, Seattle
Keywords: Ada, AI
Summary: Why is an Ada-type language essential for AI delivery?
List-Id: <comp.lang.ada>

In article <870806210856.08p@VLSI.JPL.NASA.GOV> larry@VLSI.JPL.NASA.GOV writes:
>As for as self-modifying code is concerned, I think those who've done "AI-type"
>products find that very little code of this sort is needed or desirable.  This
>is one reason why compilable LisP is such a useful tool.  Self-referencing can
>also be done in other ways than self-modifying code.
>
>The relation between AI languages and Ada/C/etc. seems to be that LisP/
>Smalltalk/etc. are very effective for exploratory programming and prototyping,
>but when a system is to be delivered it's best to translate it to a compiled
>language first.
>
>As for control structures, this is a subject I'd like to see some advice on
>too. ...

As one who has been developing AI applications for the past several years, I am
very interested in the topic expressed here.  There is a class of AI system that
does not involve self-modification.  Its members are primarily advisory or
consultant programs on very well defined problem domains.  There are probably
lots of instances where such systems might be useful, but they don't address
complex problems.  The majority of AI applications require that the system
learn, both in the sense of modifying its knowledge-base (or data base) and in
the sense of modifying its problem solving capabilities.  Learning in the first
sense doesn't require self-modifying code.  Learning in the second sense may
require self-modifying code, but at the very least will surely require that new
control structures and control techniques be capable of being created by the
system.

The set of control structures provided in Ada, C, etc. have been shown to be
sufficient for the construction of algorithmic solutions to problems under a
set of assumptions.  These assumptions are seldom, if ever, stated explicitly,
but I believe that among them is at least the exclusive use of the Von Neuman
architecture machine.  I know of no proof that the 'structured programming'
set of control structures is not sufficient, but it seems very likely that
there is more to the universe than just these.  Why do people believe that
translation to a compiled language, or a language such as Ada is essential
for a 'delivered' system?  I am honestly interested in other people's views
and experiences, even though my expectations lead me to believe that such
translation is far less important than the creation of a solution to a complex
problem.  Please put your nickels on the table, I think we can all benefit.

TJ {With Amazing Grace} The Piper
(aka Ted Jardine)  CFI-ASME/I
Usenet:  ...uw-beaver!ssc-vax!bcsaic!ted
CSNet:   ted@boeing.com
-- 
TJ {With Amazing Grace} The Piper
(aka Ted Jardine)  CFI-ASME/I
Usenet:  ...uw-beaver!ssc-vax!bcsaic!ted
CSNet:   ted@boeing.com