Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!amdcad!ames!ll-xn!husc6!endor!crawford
From: crawford@endor.harvard.edu (Alexander Crawford)
Newsgroups: comp.ai
Subject: Re: The future of AI.... (nothing about flawed minds)
Message-ID: <3111@husc6.UUCP>
Date: Wed, 4-Nov-87 23:58:00 EST
Article-I.D.: husc6.3111
Posted: Wed Nov  4 23:58:00 1987
Date-Received: Sat, 7-Nov-87 18:34:59 EST
References: <6667@ut-ngp.UUCP>
Sender: news@husc6.UUCP
Reply-To: crawford@endor.UUCP (Alexander Crawford)
Organization: Aiken Computation Lab Harvard, Cambridge, MA
Lines: 22
Keywords: future ai machine learning natural language
Summary: Machine learning and natural language processing

The first impact from AI on software in general will be natural
language interfaces.  Various problems need to be solved, such as how
to map English commands completely onto a particular application's set
of commands COMPLETELY.  (As Barbara Grosz says, if it can be said, it
can be said in all ways, e.g. "Give me the production report",
"Report", "How's production doing?".)  Once this is completed for a
large portion of applications, it will become a severe disadvantage in
the marketplace NOT to offer a natural-language interface.

Coupled with a NLI, machine-learning will allow applications to
improve in different ways as they are used:
	-Interfaces can be customized easily, automatically, for
	 different users.
	-Complex tasks can be learned automatically by having the
	 application examine what the human operator does normally.
	-Search of problem spaces for solutions can be eliminated and
	 replaced by knowledge.  (This is called "chunking".  See
	 MACHINE LEARNING II, Michalski et al. Chapter 10:
	 "The Chunking of Goal Hierarchies: A Generalized Model of
	 Practice" by Rosenbloom and Newell.)

-Alec (crawford@endor.UUCP)