         Artificial Intelligence

             Whil Hentzen
 Greater Cincinnati IBM PC User's Group

The first problem of Artificial
Intelligence (AI) is that everyone
has his own definition.  Herb Simon
and John McCarthy from the East
Coast, Edward Feighenbaum and Nils
Nilsson on the West Coast, we in
America's heartland - no two
definitions are alike, and all have
been misunderstood by the media as
they try to capture the latest
buzzwords for the 80's.

Instead of trying to set a definition
in concrete, lets use this:
Artificial Intelligence is a process
by which a device is made able to
perform tasks which, when they are
performed by humans, are said to
require some thought.

The reason for the hedging in that
statement is that it usually happens
that as soon as a machine can do some
task, that task is declared not to
have needed intelligence to be
performed in the first place.

Let's look at some of the
applications of AI.

GAMES

Computers that play chess (and win!)
use AI techniques.  The first idea is
to create a tree that branches out
into all the possible moves that the
two players can make.  However, as
the game develops, the number of
branches of a chess tree become so
p#large that even a Cray works
overtime.  The next step is to assign
values to moves - a high value to a
move that takes the other queen, a
somewhat lower value to a move that
opens your king to mate in two moves.
Thus, whole sets of branches can be
eliminated because of the low value
calculated.  The final step is to
create strategies, plans,
counterplans and all the other things
that humans do.

THEOREM PROVING

This is basically checking up on the
work of mathematicians and other
folks of that breed.  As mathematics
get more complicated (e.g. Fermat's
Last Theorem or the Four Color
Problem), a computer with a
human-like mind is handy to double
check your work.  The English
language analogy would be a spelling
checker that made sure you used
"your/you're" and "effect/affect" in
the right places.

PREDICATE CALCULUS

This is applying Boolean Logic to
ideas instead of what BL usually is
applied to.

For example, imagine the diagnostic
procedures doctors go through to
determine a patients' illness.  Now
imagine if you had some symptoms that
were related to a rare disease...a
computer with the ability to sort
through all these Rules wouldn't
care, because it had a perfect memory
and rather fast search time.  A
doctor would probably have a tougher
ptime.

These Rules could be hooked up with
programs and called Expert Systems.
There are useful expert systems
around, however - several of the
famous ones are Mycin (a medical
diagnostic expert system), and
Prospector, a mineral deposits
locator that recently found a massive
Molybdenum strike in Montana.

PATTERN RECOGNITION

How to give ears and eyes to a
machine.  Let's talk about vision.

A camera takes a picture and breaks
it down into PIXELS.  Sufficient
resolution for an 8.5 x 11 drawing
might be 250 dots per inch.  This
means, for black and white only, that
picture has 11.5 million pixels to
process.  That takes a while.  You
and I (and your 12 year old son) can
look at a piece of paper and
determine if it is an insurance form
or a centerfold rather quickly. The
computer, however, must analyze each
and every pizel.

One way of speeding it up is to break
the picture up into regions of light
and dark, of edges and places, of
shapes and objects.  Then it compares
these to images it already has stored
in memory, and makes "guess" if they
are pretty close.  Naturally, if the
picture is in color, the number of
pixels that must be processed
increases tremendously.

Try doing that on an abacus!
============================                                                                                                                            