Rl: a rule-based configurer of computer systems (in Artif. Intel. 19, 1 (Sept. 1982), 39-88
|
 |
|
Post a Comment
|
|
|
|
|
CONTRIBUTORS:
|
|
|
JOURNAL:
|
Computing Reviews,
24(6),
263 -
264.
|
|
|
|
YEAR:
|
1983
|
|
PUB TYPE:
|
Book Review
|
|
SUBJECT(S):
|
DESIGN, HUMAN FACTORS. MANAGEMENT
|
|
DISCIPLINE:
|
Computer Science
|
|
HTTP:
|
|
|
LANGUAGE:
|
English
|
|
PUB ID:
|
103-426-837
(Last edited on
2006/05/23 12:57:14 GMT-6)
|
|
ABSTRACT:
A prerequisite for reading this paper is the examination of "XSEL: A Computer Sales Person's Assistant" by the same author [I]. XSEL is a computer program designed to be inserted into the customer-salesperson relationship when the customer is buying a particular computer configuration (or system) and the salesperson is selling products of the Digital Equipment Corporation (DEC). It makes technical decisions for the salesperson so a system design is produced which fits customer needs and will function as intended. Used in conjunction with XSEL is RI, a program which actually does the configuring and produces the end
product which the salesperson uses to price and deliver. RI is written in various versions of OPS (OPS4 in this paper), a language sponsored by DARPA and monitored by the Air Force Avionics Laboratory. While RI is briefly mentioned in the earlier paper, it is center of attention in the paper under review.
One of the joys of attempting to comprehend what artificial intelligencers are talking about is the effort to learn their lingo. AIers, like most computer scientists, continually generate new terms which are exciting and fun but make the reading of a technical paper hazardous to one's mental health. McDermott is no exception to this behavioral rule. A good method to offset this is to jot down such terms as instantiation, component-token, gettemplate, fan-in, fire. and context spawning, along with their definitions, while following along in the text. Once the reader achieves an understanding of the jargon, the concepts expressed by the author are crystal clear.
Generally speaking, McDermott extracted knowledge by various kinds of interviews, known to a vocational educator as job analysis and to a military or engineering psychologist as task analysis. He simply asked the experts what they did in the process of system configuration. He probably observed them during the process as well. One of the problems which emerged was that exceptional configurations could not be generated on demand. This is probably because experts store their knowledge in such ways that it is extremely difficult for them to access constraints given only actions. Of course, the basic goal was to create a rule structure which would produce a suitable configuration without human intervention. Once the human thought process had been thoroughly delineated, it was relatively easy to produce this structure. It consisted of 772 rules in two major areas: domain-specific (480) and general (292). The domain-specific consisted of context generation (96), prerequisites (127), component association (156), retrieval (54), and computation (47). The general included: output generation (106), context deactivation (84), counting (54), and set-up (48). Thus there were really nine different kinds of rules which RI had to incorpora te in the logical structure of the program. Examples of these rules and the statistics of test runs are interesting reading. (Incidentally, a newer version of R1 using OPS5, reported in the earlier paper, has about 850 rules, testifying to the continued interest for improving and expanding the programs.)
As computer-based automation, including robotics, evolves, the need to examine human decision-making and to record, test, and verify the process will grow. This method of analysis has been used since 1920 by both industrial engineers and vocational educators to elicit job and task knowledge from trade experts and to convert this into job standards and job training where humans applied the standards and humans performed the training. Today an identical process using new terminology and much higher resolution is under way to elicit job and task knowledge from technical experts and to convert this into computer programs where machines perform the process. Something which began with Taylor in engineering and Allen in education continues with McDermott and others in artificial intelligence.
|
|
|
|
STATISTICS
|
|
Click on # to view
|
|
Citations
|
|
1
|
|
References
|
|
0
|
|
Comments
|
|
0
|
|
Quality
|
|
0/0.00
|
|
Interest
|
|
0/0.00
|
|
View(er)s
|
|
2/80
|
|
|
|
|
|
|
| Prev |
Next |
|