[1087] in Public-Access_Computer_Systems_Forum

home help back first fref pref prev next nref lref last post

Re: Expert systems for training

daemon@ATHENA.MIT.EDU (Ernest Perez)
Fri Aug 28 15:37:32 1992

Date:         Fri, 28 Aug 1992 14:16:16 CDT
Reply-To: Public-Access Computer Systems Forum <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
From: Ernest Perez <demoep@mail.psi.net>
To: Multiple recipients of list PACS-L <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
In-Reply-To:  <9208272215.AA29075@worldlink.worldlink.com>

----------------------------Original message----------------------------
>Hi all. Does any of you know about any expert system done for training in anyth
>ing? Any such system in public domain, (if so, where)?.
>I aks this because i`ll soon start work on a system for the training of employe
>es of the National Tax administration of Colombia, and I`ll like to get an exam
>ple of such a system (on pC`s).
>Any related info will be welcome, and highly appreciated. THANKS,

Jorge,

AI Corp. sells "1st-Class HT," an excellent expert system generator, with
advanced hypertext and macro/shelling abilities.  1st Class is also
unique, in that it uses the "I3" algorithm (Inductive Iteration...or
something along that line).  This enables the generation of an expert
system network or rule table from supplied examples.

The author/user defines outcomes, and significant factors.  You then
either fill in examples of variables and the outcomes.  When you have
sufficient entered to give a good sample of case outcomes, you simply
command the system to build the rule table.  ***You can also build the
system upon past experience examples, imported from several standard
formats.*** These include dBase, Lotus, comma-delimited ASCII, etc.
The 1st Class rule building interface looks very similar to a
spreadsheet, and is quite simple for editing, changing examples, etc.

Rule bases I've built and seen built take a few seconds to 15 or 20
seconds for the construction of the expert system.

The system allows:
  * custom screen design & layout
  * custom queries for entering of variables by end user
  * custom online help for any entry or part of the system progress
  * hypertext explanations, background, auxiliary information, reasoning
to explain conclusions, etc.  Each hypertext screen can also have a
network of links.
  * shelling to external programs, passing variables via memory or via
disk files
  * free expert system runtime, no licensing of the distributed expert
systems
  * access to and direct read/write to dBase relational tables, for
lookups or calculation
  * printing a rule tree of the generated system, for debugging/analysis
  * generating generic C code and (Pascal or Basic code, I've forgotten
which), allowing you to transport the expert function into your own
applications.

      ...and much other good stuff.

The I3 algorithm allows for FAST building of systems.  The work is
essentially in having experts defining factors and outcomes and
constructing examples.  You must then get knowledgeable folk to
analyze the generated system, trace out the rule tree (to see if it's
really valid), etc.  Interesting thing about the rule tree is that it
will always ask the minimum number of questions needed.  An answer
may eliminate need for later questions.  It asks questions in the most
effective order, to result in least number of questions.  E.g., you may
have a rule tree built on 14 determining factors; but a sequence of
answers may result in asking of only 4 questions.

It's a NICE system!  The only problem is the steep price...$2495.  That does
give you a good bit of power though...free runtimes of whatever you
develop, ability to import past experience, interface both ways to other
applications, customized expert system screens and operational
appearance, the hypertext/text display ability, and incredibly
productive system-building power.

They have nice brochure packages, and some nice demos.  They also have a
tutorial package for about $40 (credited to purchase).  The tutorial
walks you through some system creation.  It also contains a limited set
of the full program.  There is some unspecified limitation on the number
of determining factors, of possible outcomes, or of user-defined example
cases. You can build some *very* nice small systems, though.  I've not
hit the limits on the tutorial version in any of the small samples I built.
(I also have the full package now)


Versions available for PC, LAN, VAX (transportable from PC)

AICorp, Inc.
138 Technology Drive
Waltham MA 02254-9748
Tel: 800-667-2400 or 617-891-6500
Fax: 617-893-8919

home help back first fref pref prev next nref lref last post