Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!gatech!EU.net!news.sprintlink.net!noc.netcom.net!netcom.com!vrotney
From: vrotney@netcom.com (William Paul Vrotney)
Subject: Re: Student - what's AI?
In-Reply-To: gab@triode.apana.org.au's message of 27 May 1995 03:53:46 GMT
Message-ID: <vrotneyD9891L.AME@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3q67oa$l4h@triode.apana.org.au>
Date: Sat, 27 May 1995 08:03:21 GMT
Lines: 62
Sender: vrotney@netcom12.netcom.com

> In article <3q67oa$l4h@triode.apana.org.au> gab@triode.apana.org.au (The Gab) writes:
> 
>    I have just been shown prolog at school and told that that's what AI is. 
>    Is that really all it is? Feed the computer a whole lot of facts and then 
>    cross-reference the info back?
> 
>    If so, what's all this about ai being really cool and make it possible 
>    for computers to do all sorts of great stuff.

No, Prolog is not AI.

First, AI is "synthesized intelligence" as opposed to what we know as
natural intelligence.  Which leaves the definition of just "intelligence".
And of course that is simple:

         "Intelligence is what intelligence does."

You may think this is a flippant answer to your question, but these days the
definition of AI is still considered by most to be intractable.

The thing that makes AI "really cool" is actually trying to define it and
implement something using AI.  Their exists a growing body of AI techniques
consisting of algorithms and data representations.  The collection of
literature on these techniques along with other musings represents the
science of AI.  Taking a problem currently unsolved or solved in such a way
that the results are considered unintelligent and implementing a program to
produce what most would consider intelligent results using your own brew of
AI or this body of AI techniques is the "really cool" adventure.  For
example think about a program that can exhibit common sense, something that
most programs these days don't do very well.  Imagine a video game where the
characters exhibit their own intelligent behavior as opposed to a fixed
scenario type of behavior.  The science of AI is wide open for some young
Einstein.  You don't need a lot of money to participate in AI.  You can do
it today on an inexpensive home computer.

One of the important results from AI research is the realization that to do
AI you need to encode "a whole lot of facts" as you say.  This is surely a
problem and one of the things that causes many people to give up.  But this
may just be because we don't know how to implement AI very well yet.
Perhaps the answer here is in building a program that is able to collect all
those necessary facts automatically by itself.  Or perhaps the answer is
discovering the science of specific or general bodies of knowledge and
encoding that, much like we can encode physics equations to exhibit the
knowledge of physics.  Or perhaps a combination of many techniques.  No one
currently knows the answers to these questions, which are intriguing.

Beware.  There are a number of AI workers these days who refuse to use the
term "AI".  But don't let them discourage you.  Many of these poor people
have be forced into thinking that AI is too much hype or should be aligned
with something political or economic as opposed to a science.  There have
been a lot of failures in AI, as there is in any new science.  There have
been and still are a lot of AI charlatans as there is in any new science.
Many of the "give up AI people" will ask you what is the difference between
an AI technique and just an ordinary Computer Science technique.  Don't be
disappointed if you can't answer, it's a trick question based on the
intractability of the definition of intelligence.  Just because a term is
intractable is not a good reason to stop using it.  For example, where would
we be today if our predecessors gave up on using the term "infinity".

-- 

William P. Vrotney - vrotney@netcom.com
