| From: | Robert Hayden <rahayden@*****.WEEG.UIOWA.EDU> |
|---|---|
| Subject: | 20th Century Expert Systems |
| Date: | Tue, 27 Oct 92 07:51:09 CET |
Ok, let's all gather around the campfire and I'll tell you all a little story.
There are two types of expert systems I know something about, one of which
I have dabbled a little bit in.
At Mankato State University in Mankato, MN (USA), they have an IBM AS/400
with a 4GL system running. 4GL is short for "fourth generation language"
which is a type of programming language. The system at MSU would take
english input and spit out cobol code (which, in itself is not very
elegant). Buy being able to tell the computer in everyday english what
you wanted it to do, you were able to DRASTICALLY reduce programming time.
You might spend 20 minutes setting your parameters and then the computer
would chug away for about 5 and spit back 20,000 lines of code (with
documentation no less :). I only used it 3 times and never did anything
elegant, but i did watch a grad student friend of mine write some pretty
hefty stuff on it.
Another expert system that I know about was being designed by a group of
grad students at some southern california techie university. The system
was CAD oriented and they were calling it ECAD (for Expert Computer Aided
Design). In any case, the purpose of the program was to design model
airplanes and then to use a connected CAM (computer aided manufacturing)
system to actually produce the parts to the aircraft.
What the students would do was to input a series of requirements and the
computer would then output the blueprints and control the machinery to
tool out the parts.
For example (and I'm being grosely simplistic):
Payload Size: 20lbs
Engine Thrust: 300 lbs
Engine Wt.: 40 lbs
Fuel Wt.: 10 lbs/gal
Fuel Consumption: .3 gal/minute @ full power
Distance Required to Fly Payload: 10 miles
Max Wt: 120 lbs
The computer would then design the aircraft, inputting all of these
variable and all of it's knowledge on aerodynamics. Then it would spit
out a design that it thinks would carry a 20 lb payload 10 miles and weigh
less than 120 lbs. Then, given a command, would actually cut pieces like
the wings and things to make sure they were completely accurate.
In any case, the results the students were getting were >85% accurate and
many times the failures could be blamed on things the computer was not
programmed to take into consideration (such as weather).
The beauty of the system is that it was very easy to change the parameters
and get a new design. For example, they might change the payload size
from 20 to 30 lbs, and the computer would respond with a larger wing area
to provide more lift at the same time it sweeps the wings back for better
aerodynamics.
That is how expert systems work in 1992.
------------------------
Now, how the expert systems I proposed work:
The parameters are entered through the computer into the firmware. The
firmware uses the data in the datasoft to create a solution and then tests
it in the Knowsoft (runs a simulation). It takes the data from the
simulation and modifies the data in the program and then runs it again,
ultimately arriving at what it thinks is a solution (then makes some
checks vs the computer's INT rating).
This is pretty much directly analogous to 20th century computers.
>> Robert Hayden | rahayden@*****.weeg.uiowa.edu | aq650@****.INS.CWRU.Edu <<
>>>-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-<<<
>> I'm currently writing a new opera: <<
>> It's called The Divorce of Figaro <<
