In a continual quest for excellence and application of leading edge technologies, The Institute of Technical and Adult Teacher Education (ITATE) introduced Expert Systems and Education, as a final semester option, into their Graduate Diploma in Educational Studies (Computer Education) during Semester 2 of 1989.
This paper presents the lecturer's perspective of the inaugural launch, method of presentation, evolution of format, conceptual pitfalls and teaching challenges together with consensus strategies for hurdling foreign jargon, alien methodologies and preconceived mysteries.
Despite unexpected manifestations of seemingly myriad problems, some bizarre computer feedback and tremendous emotional drain in continual attempts at demystification of strange concepts, the author found Expert Systems and Education the most difficult, unconventional yet fascinating and rewarding experience in a twenty-five year tertiary teaching career.
The lecturer's qualifications range across several disciplines [Electrical Engineering \ Applied Science \ Operations Research \ Computing \ Education], 14 years in industry, 21 years full time tertiary teaching. He qualified for the Grad Dip Ed (Computer Education) in 1988 with the project Artificial Intelligence - Expert Systems - Impshell and has studied educational techniques during the past 15 years.
Available software consisted of ESIE and IMPSHELL-87 Shareware Expert Systems Environments, evaluated by two ITATE students in partial fulfilment of requirements for the award of Grad Dip Ed (Computer Education). Every other resource had to be begged or borrowed. For example, IMPSHELL-89 was begged, while EXPERTEACH III was borrowed.
Documentation for IMPSHELL was finally published in text book form and became available in Australia during April, 1989. The book Expert Systems Programming in TURBO PROLOG, Daniel H Marcellus, Prentice- Hall Inc 1989, which included IMPSHELL-& diskette, was used as a primary reference for Expert Systems and Education. In addition, the impressive Online Study Course, EXPERTEACH III, IntelligenceWare Inc., 1988, was introduced, for consolidation purposes, as a self paced tutorial.
Under the prevailing conditions, one's teaching ability and use of inner resources, guided the direction of the course and proved the main contributing factor to the degree of success of Expert Systems and Education. The author was prepared to face peer and student judgment at the end of the inaugural course during December, 1989.
Essentially, this course was designed for maximum effect. It would demystify Expert Systems and explore the depth as well as breadth of the topic in a graphic, diagrammatic, jargon free mode. The group would focus on understanding, creating, achieving and evaluating, rather than arguing about shades of meaning in a large vocabulary of strange nomenclature. In the words of Anna Hart, Knowledge Acquisition for Expert Systems, Kogan Page, 1989:
a no nonsense approach to a subject which is too
often obfuscated by jargon and technicalities.
Assessment tasks were presented together with the notion that the group would be studying an emerging body of knowledge. The introductory phase concluded with some suggested areas of research and the presentation of a substantial list of reference books.
Expert systems demystification commenced with individual student participation in hands on attempts to solve the classic, four coin, nickel and dime problem. After three to four minutes a visual representation of the complete problem space, in traditional state space decision tree format, was displayed using the overhead projector and a copy was distributed to each student.
The historical perspective took the form of a chronological scan through fifteen important milestones of AI research between 1947 and 1990. Artificial intelligence in education was illustrated by a transparency featuring intelligent computer assisted instruction (ICAI) placed in context of AI activities and lessons learned during the two decades, 1960-1980.
The basic elements of artificial intelligence were conceptualised in the form of a segmented wheel, each segment representing a unique subset of AI. Expert Systems was highlighted in the transparency. A later transparency illustrated the distinctive similarity between the User - Inference Engine - Working Memory - Knowledge Base Expert System Architecture and Student - Tutor - Prior Knowledge - Knowledge Base, ICAI Educational Model.
Expert Systems in Management were presented from the point of view of the "Intelligent Assistant" as a facilitator in the decision process. The module concluded with several applications of Expert Systems in Crisis Management.
After approximately 30 minutes, it was time to introduce a change of pace in the form of an exploratory walk through the various menus plus all frames comprising the introductory Item 0: Background on Expert Systems within the EXPERTEACH III tutorial.
The EXPERTEACH III learning environment posed a series of pertinent questions and provided simple, jargon free answers, in an impressive colour coordinated graphical format. For example:
EXPERTEACH III thus led the group through Methodologies for Dealing with Knowledge, Knowledge Oriented Applications, Differences Between Traditional Programming Languages and Logical, Object Oriented Programming. The penultimate frame displayed a chronological development of programming languages in block graphical format. The initial contact with PROLOG and LISP took place in the penultimate frame of Item 0.
The next change of pace was designed to slot the presented terms and concepts into context by conceiving a possible application. A classic mystery "Electronic Expert Solves Murder at Mumfrey Manor" was thus specifically chosen to reinforce the group's interest and to blend the new concepts with previous knowledge.
SHERLOCK initiates this dialogue, one question per frame:
"Murder at Mumfrey Manor" clearly illustrates that SHERLOCK, the Electronic Expert, finally unmasked the killer by following a simple chain of logic, not unlike a trail of footprints. Each new step followed on the heels of the last until the destination was reached.
Students were familiarised with formal problem representation, consultation paradigms and five strategies for representing knowledge. "Semantic Networks" were thus related to "Object Attribute Value Triplets", "Rules", "Frames Inheritance", "Logical Expressions".
Forward and Backward Chaining was introduced as techniques of "groping for a solution" or "confirmation of hypotheses" respectively. Various search patterns, Exhaustive, Random, Heuristic were illustrated and evaluated. The Animal Classification system was explored within ESIE and "Investment Adviser" within IMPSHELL as hands on illustrations of the differences between forward and backward chaining respectively.
The perceived future of expert systems was illustrated in the form of two waves. The first, smaller wave, is generally predicted to crest at the present time, while the second, much larger wave, is just beginning to swell and is not expected to crest until 1992 to 1995. The second wave is expected to have a significant impact on business and industry.
Because IMPSHELL is coded in PROLOG, all students were first familiarised with basic differences between Procedural and Declarative programming styles as two perspectives of the same knowledge. Tracing, checking, editing rules required an exhaustive analysis of the internal, IMPSHELL eight segment, rule storage format. The rule structure was visually related to simplified real life situations such as: Transportation Planning, Housing Loan Evaluations, Aircraft Recognition, Management of Minor Ailments.
IMPSHELL Rule Types, Reversibility, Positive, Negative applications, Premises, Conclusion were illustrated in "field-record" file processing format and related to simple examples. AND, OR, and NOT operators were defined within "Crisp Logic" and extended to "Fuzzy Logic".
The MYCIN style confidence (certainty (ct)) representations of "definite", "almost certain", "probably", "slight evidence", "don't know", "practically not", "almost certainly not", "definitely not", were marked on a number line between limits of -1.0 and +1.0. The visual impact of this representation proved to be immediate and total.
[ct(conclusion) = ct(evidence)*ct(implication)]
[ct(evid1 AND evid2) = min(ct(evid1),ct(evid2))]
[ct(evid1 OR evid2) = max(ct(evid1),ct(evid2))]
0 +0.76 +1.0 [-----------------------------------] (First conclusion .6) ^ ^ 0 ^ +1.0 [---------------] (Second conclusion .4) ^
[ctotal = (ct1 + ct2)/(1 - min(abs(ct1),abs(ct2)))]
Equivalence between "Inference Net Notation" and "Rule Notation" was highlighted on a transparency and immediately interconnected, coded and saved within IMPSHELL as a "Medical Rule" (MEDICAL.RUL) inference net. Hand calculations were performed on MEDICAL.RUL inference net for several sets of client responses and compared with results returned by IMPSHELL. Similar tests were carried out for a Mortgage Loan example and on FUZZYNET, an inference network especially constructed to reason about competing hypotheses.
Hand calculations were applied to the substantial, 25 question - 6 Hypothesis, "Investment Adviser" supplied on the IMPSHELL-9 disk. A comparison with results returned by the Expert System disclosed a discrepancy in the degree of certainty in one of the six Investment Hypotheses. A careful examination revealed a missing full stop (.) in a single statement within the IMPSHELL knowledge base. The IMPSHELL environment was particularly unforgiving but our methodology withstood the acid test.
The simple inference net discloses that numerical and logical inferences converge at the two concluding nodes and have to be reconciled at both locations.
IMPSHELL is designed to manipulate algebraic expressions by conversion to Reverse Polish Notation (RPN) and returning a TRUE(+1) or FALSE(-1) result by an embedded RPN interpreter. This aspect was thoroughly tested and extended by introducing the group to the PROLOG environment, and in particular, PROLOG-C, PROLOG-PASCAL project modules. System execution traces were often saved and evaluated as aids for error diagnosis.
The main segments of IMPSHELL were interpreted in PROLOG source code and a small test program FUZZYNET was executed using standard PROLOG and within the IMPSHELL environment. Both results were verified by hand calculation. The group was introduced to the inbuilt WHY and HOW capabilities and a detailed exploration of the WHY stack was undertaken. The HOW function did not appear to operate within IMPSHELL-89.
The prelude for individual student project work consisted of a detailed tour through the Expert System Creation Sequence. The mode of presentation consisted of successive computer screen dumps transposed on transparencies. The creation dialogue was emphasised and applied to a small example.
Students were acquainted with current activity in Expert Systems applications. These included PICON, PREDICTE, WES, APES, CLASS, JET-X, CARE, ROSES, ECAS together with a brief description of the commercial shell NEXPERT. Current openings for Artificial Intelligence and Expert Systems professionals were mentioned.
After establishing interconnecting nodes and defining the backward chaining process with a flow chart it seemed like an easy task.
Could anything be more straight forward than this backward thinking?.
I found the development of my own non expert 'expert system' a time consuming but very rewarding exercise.
I entered 'you have tender abdomen' in my implication statement instead of 'you have a tender abdomen'. The result of this was that in the printout of the 'inference summary' statements beginning with db occurred which I could not understand.
The "post-run" rule set, however, is typically much longer. Additional information, provided by the system, consists of evidence, "infer_summary", "danswer" together with equivalent terms for mathematical implications.
These extra terms establish a trace and display IMPSHELL certainty calculations at all nodes.
There are no traditional diagnostic messages. Character mismatches, however, appear to be indicated by "dbimp" from the WHY stack and the undefined "tdbimp" statements in the "post-run" rule set. Hypotheses, which exclude contributions from mismatched paths, are evaluated by IMPSHELL.
Student essays, designed to broaden their knowledge of applications, featured the following titles:
The author was amazed by the rate of progress during the 15 week course. Course content was exceeded while most assessment tasks were original and of exceptional quality. It appears that the extensive coverage was possible by continual application of the Confucian Dictum - "one picture is worth a thousand words". Top priority was given to "what it is" rather than "how you say it".
The style of presentation was adaptive to student needs and aspirations. Group dynamics influenced the agenda for each session. The numerous books, journals, articles, consulted by the group makes a bibliography impractical. However, books by Anna Hart, Chris Naylor, C. F. Chabris, Harmon and King, P. S. Snell, A. C. Staugaard, W. J. Black, B. G. Silverman, Clocksin and Mellish, V. D. Hunt, D. N. Chorafas, together with ITATE Theses by B. J. Smith and G. I. Gedgovd, were frequently examined by the participants.
In many ways Anna Hart, Knowledge Acquisition for Expert Systems, 1989 summarises the cooperative effort of the ITATE class of 1989 by the following words:
I have seen many good projects. They all had somebody in the development team who had a questioning spirit, who was able to take on ideas from different disciplines and who was not afraid to change a decision.Despite unconventiality, selective usage of emotions added "enthusiasm", "flair", "a sense of adventure" to a subject "often obfuscated by jargon and technicalities". Perhaps our collective appreciation of the intuitive elegance of AI played a major part in the success of Expert Systems and Education.
Recognition of achievements have been swift and positive. One participant has been invited to teach Expert Systems while another intends to pursue the Expert Systems\ Computer Aided Learning (CAL) interface to master's level. With gradual resolution of teething problems, acquisition of worked examples plus additional Expert Systems Shells, it is anticipated that the 1990 session will offer a substantial advance on the described course.
|Please cite as: Gedgovd, G. (1990). Expert systems and education: An Innovation in Grad Dip Ed (Computer Education). In J. G. Hedberg, J. Steele and M. Mooney (Eds), Converging Technologies: Selected papers from EdTech'90, 194-204. Canberra: AJET Publications. http://www.aset.org.au/confs/edtech90/gedgovd.html|