brantly.henr@XEROX.COM (01/22/87)
Does anyone have any experience with a package called "Expert Opinion" by Mind-Soft that runs on the ST? It's advertised as "the first general-purpose expert system creation package for the Atari ST". Retails at 99.95 but I've seen it mail order as low as $75. Both forward & backward chaining are said to be supported. Anybody have a personal review on this one to share? Thanks, Dennis [Dennis Brantly, Xerox Corp., Rochester, NY., (716-671-0160)] [Let it snow, let it snow, let it snow!]
rostain-alain@YALE.ARPA.UUCP (01/28/87)
I did review Expert Opinion for Derek Electronics in New Haven.
I think they had it on sale for just $60 at the time (october?)
Anyway, before you consider buying, here's my review:
The packaging of EXPERT OPINION, a "productivity" software product from Antic,
claims it is "a powerful AI tool for building expert systems." Fortunately
for the authors, the user's manual is more honest. The $Trouble-shooting$
section points out the serious limitations of the program and advises the
user to conform his/her "design to the system's strengths, and steer clear
of its weaknesses." EXPERT OPINION is not a powerful AI tool, nor will it
help users be more productive. It is, at best, a fun toy.
If you plan to write a complex
and productive expert system, don't buy this product (or any PC
AI tool for that manner). On the other hand, if
what you're looking for is an introduction to an aspect of Artificial Intelligence, as of
1970 or so, and a tool that will help you explore some of the problems
with expert systems, then EXPERT OPINION may be for you.
The program is actually 2 programs. The first, BASE, is for creating,
modifying and deleting knowledge bases. A knowledge base in this system
is just a set of rules of the form "If premise1 (and premise2 and ...) then
assertion1 (and assertion2 and ...)" where premises and assertions are
just facts of the form (subject verb object (optional)). For some unknown
reason the program forces you to link assertions with "or" instead of "and,"
which is counterintuitive and logically incorrect since in execution,
all assertions are concluded when all the premises match.
Another problem with BASE is the ASSISTANCE feature, which is supposed to help
the user create new rules by suggesting all possible new rules given all
existing rules. Since the number of possible new rules is huge, it
just isn't practical for the user to go through these in order to create
new rules. Despite these two minor problems, BASE is generally well
written and easy to use.
The second program is called EXPERT (or something close to this).
This program is the "inference engine" for the system. It basically
allows you to create, modify, and delete facts from the system, and run
a knowledge base on existing facts. The engine can be run in 3 modes:
DEDUCE, VERIFY, and EXPERTISE. In simple terms, DEDUCE takes all
existing facts and runs the rules on them until no more facts are found;
VERIFY asks for an inputted desired fact and tries to prove it
given the rules and preexisting facts; and EXPERTISE is just a silly
variation of DEDUCE. The most noticable fault in this part of the system
is the FACTS? command, whose purpose is to display to the user all
existing facts. What this command does is display each fact, one at a time,
for about a second on the screen. So if you have 30 facts you may have to sit
through 1/2 a minute of ticker tape facts! Obviously this poorly
written command was not authored by the same person who wrote most of
the rest of the interface.
One big question I have is "why did the authors choose to divide the
system into 2 parts?" In order to effectively build an expert system, a
user must constantly be switching between writing rules (BASE) and testing
then out (EXPERT). Dividing these 2 tasks into two programs makes
EXPERT OPINION difficult to use. This was a poor design decision.
To sum up my review, EXPERT OPINION is a toy pretending to be
an expert system tool. Among it's documented serious theoretical limitations listed
are 2 major ones. There is no concept of negation (i.e. the user can't
assert that something is FALSE) and there are no variables (i.e. I cant have
a rule like "IF X IS A PERSON, THEN X HAS 2 ARMS AND 2 LEGS" and input the fact
"JOHN IS A PERSON" and expect the system to figure out that John has
2 arms and a leg.) Given these constraints, it is impractical
to write an expert system using EXPERT OPINION. It is just too simplistic.
Potential buyers should be aware of this when considering this product
for purchase.
Alain Rostain
Yale University
-------