Conference heron::euro_swas_ai

408.0. "EXCELLENT LETTER !!" by MR4DEC::ROACH (TANSTAAFL !) Mon Feb 17 1992 21:12

                  I N T E R O F F I C E   M E M O R A N D U M

                                        Doc. No:  007935
                                        Date:     17-Feb-1992 01:23pm EST
                                        From:     PAPAGEORGE
                                                  PAPAGEORGE@LMOADM@AIDEV@MRGATE@NRGATE@NRO
                                        Dept:      
                                        Tel No:    

TO:  PAT ROACH@LMO


Subject: EXCELLENT LETTER !!                                         

From:	GUESS::GILBERT      "Ken Gilbert"   17-FEB-1992 10:10:30.33
To:	@k$dis:aitc_staff
CC:	
Subj:	FYI - Letter to the editor I sent out today...

Dear Mr. Card,

   I am  not  a  regular  reader  of  IEEE Software, but a colleague of mine
presented me  with  an  article  that appeared in your September 1991 issue;
"What's  So  Bad  About  Rule-based  Programming"  by  Xiaofeng  Li  of  NEC
America, and I felt compelled to comment.

   With  all  respect  to  Dr. Li,  although  he makes strong statements and
draws  pointed  conclusions  about rule programming, his conclusions are not
supported with equally strong evidence. My experience with rule  programming
has  been in sharp contrast to Mr. Li's. I've worked with rules for the past
10 years, 4 as a software engineer working on Digital's configuration expert
system (XCON),  and the last 6 working on expert system development tools. I
have empathy for Mr. Li. All the issues he raises are real,  and the effects
can be overwhelming.   But  unlike  Dr. Li, I have seen the "fatal flaws" of
rule programming overcome, with astonishing results.

   I believe our basic difference is in how we view rule programming.   From
Dr. Li's article, he appears to believe rule programming is somehow magical,
that simply using the rule programming paradigm will result in  more simple,
more maintainable solutions.  When rule programming fails to deliver on this
mystical promise,   he  concludes  the  paradigm  is  flawed. I believe rule
programming is simply another programming technique,  effective when applied
to the class of problems it is suited for. Simplicity and maintainability do
not come for free. As with all programming tasks,  a disciplined engineering
approach is required.

   This  is  why I believe XCON continues to be, as Dr. Li points out in his
article, an exception.   Because  of the combination of XCON's importance to
Digital and XCON's  rapidly evolving domain,   we  were  forced  to  take  a
disciplined approached early in XCON's life cycle.   The  result  has been a
system that has grown almost two orders of magnitude in size (700 to 15,000+
rules,  200 to 40,000+ configuration objects), has survived continual change
for 10+ years (I personally did a study that  showed greater than 75% annual
change for 2 consecutive years),   and that continues to supply Digital with
timely,   critical functionality  (XCON is released quarterly supporting all
the  new  products  being  announced/offered in the following quarter). This  
continued  success  is  in  no  small  part  attributable to the use of rule
programming technology.

   I  don't  understand  why  Dr. Li has felt compelled to condemn an entire
technology  based  on  such  a  limited set of data points, but I think it's
unfortunate  that  your  readers  have  been exposed to only one view on the
subject.   Should  you  be  interested  in another, hopefully more balanced,
article about rule programming,   my  collegues and  I   would  be more than
willing to write about our experiences.

	Respectfully,

		Ken Gilbert
		Digital Equipment Corporation
		111 Locke Drive
		Marlboro, Mass. 01752
		508-480-5757