Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!europa.chnt.gtegsc.com!howland.reston.ans.net!ix.netcom.com!netcom.com!olea
From: olea@netcom.com (Michael Olea)
Subject: Re: Who needs AI when there is OO?
Message-ID: <oleaDAA1xB.L02@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3rrnnq$k2@pheidippides.axion.bt.co.uk>
Date: Fri, 16 Jun 1995 17:58:22 GMT
Lines: 34
Sender: olea@netcom12.netcom.com

Michael Cox <cox_m@gfms.bt.co.uk> writes:

>I work in developing network management systems for supporting the repair
>and maintenance of telecommunications networks.

>Object-orientated modelling of networks entities and the interactions between 
>them, operations on them etc in C++ is seen as the panacea to solve all
>network management problems.

>We do have one rule-based system that associates related alarm messages from
>the network which my boss says he is 'uncomfortable with'.

>Can anybody suggest any justifications for proposing a rule-based solution over
>or as part of an object model based one?

>What about AI tools in general is OO a panacea or can AI tools add value that 
>cannot be realised with OO?

>- Michael Cox.


	OO is packaging; AI, at least in the form of expert systems,
is technology.  OO no more makes rule-based systems obsolete then it
makes bubble-sort obsolete; rather, it provides an often convenient
means of implementing systems.  So the question of how best to
associate related alarm messages is not "with objects or with rules",
but on the basis of a tractable model.  Are hueristics required or
is there enough information (and compute time) for a deterministic
calculation?  Presumably hueristics are required since a rule-based
system is in place.  Is there probabilistic information available?
Conditional probabilities, etc?  Certainly use OO, but use it to
implement an appropriate inference system (e.g. Bayesian Network,
Decision Tree, whatever).

