OBS: Arkiv – underhålls ej mera

SVARNE - an expert system based on tacit knowledge

Charlotte Magnusson,	Center of Rehabilitation Engineering Research, Lund University, P.O Box 118, S-221 00, Sweden, phone: +46 46 222 4097, e-mail: charlotte.magnusson@certec.lth.se
Jan Eric Larsson,	Knowledge Systems Laboratory, Stanford University, USA Department of Information Technology, Lund University, P.O Box 118, S-221 00, Sweden, phone: +46 46 222 7523, e-mail: janeric@it.lth.se
Kirsten Rassmus Gröhn,	Center of Rehabilitation Engineering Research, Lund University, P.O Box 118, S-221 00, Sweden, phone: +46 46 133505, e-mail: kirre@certec.lth.se

Frontiers in Artificial Intelligence and Applications - Edited by G. Grahne
The proceedings of the Sixth Scandinavian Conference on Artificial Intelligence (SCAI'97),
pp. 279-280, Helsinki, Finland, August 1997

Abstract
1. The SVARNE program
2. Project and method
3. Results

Abstract

The SVARNE project investigates how an expert system may be used when dealing with occurrences of violence within the care of people with cognitive impairments. The method used is potentially applicable in many other areas where the expert knowledge is mainly based on practical experience and thus, at first sight, difficult to implement within an expert system. By attempting to use AI technology in such an environment the number of fields where AI technology may be applied is considerably increased.

The decision tree structure of the SVARNE system is one of the first attempts to give any structure to the knowledge in this area. The SVARNE project shows that it is possible to build an expert system in a field where much of the knowledge present is tacit / based on practical experience. In the project a method for knowledge acquisition which works in this type of environment has been obtained.

The SVARNE program

The SVARNE decision support program is written in C++, and runs under Windows 3.1 and later (also Windows NT). The expert system uses standard backward chaining, much in the style of MYCIN, and the rule base contains 110 rules at present. The SVARNE program is fast and easy to use.

The program can be described as an intelligent check list with practical advice included and is intended to serve as a tool for the transfer of knowledge. It gives advice applicable to different situations and provides a structured way of thinking about violence. Furthermore it is intended to counteract the well known cognitive tunnel vision that make people focus only on one or very few issues in a critical situation.

Picture 1. Example of an expert system question.

Running the expert system forces the user to consider several different hypotheses, presented in the shape of questions (picture 1). For each such question there is also practical advice available (picture 2). Due to the complex nature of the present field the expert system does not stop once it has arrived at a conclusion, but instead allows you to continue to get multiple advice.

Picture 2. Example of an expert system suggestion.

The program also has two facilities for additional documentation. The first of these is to simply add personal information to a small data base. The second one is to actually add rules and suggestions into the leafs of the expert system, which in this way becomes tailor made for each user group.

Project and method

The method used to extract the knowledge is to start off with an expert with deep knowledge of the field and who has worked for a long time in the field. The problem is then outlined together with this expert and a first prototype of the program is made. Gradually as the system grows the number of experts are increased. In this process the structure of the decision tree is determined. In order to fill the leaves of the expert system tree with practical information, a bigger number of experts are added at the end of the program development process. The large expert group is then divided into small groups who simultaneously run the program and discuss the rules and suggestions.

In order to evaluate the program, several group homes and day care centers have used the SVARNE program in their work.

Results

The SVARNE system shows that it is possible to use an expert system also in areas where most of the knowledge available is based on tacit knowledge. The project indicates that, when dealing with this kind of knowledge, knowledge acquisition may be performed by using the expert system in an iterative process. In this process each version of the expert system serves as a tool for the capture of further knowledge to be implemented in the next.

The SVARNE program has turned out to be particularly useful in the following two settings:

as a tool for direct human - human communication. One person may, by running the SVARNE program together with other people, more easily transfer his or her knowledge.
as a tool for indirect human - human communication. The program presents or activates knowledge and questions which may lead to a better understanding by a single user running the program.

The users finding the program most useful are:

temporary or new employees, for which the information in SVARNE is immediately useful.
people with a lot of experience in the field. These users enjoyed the structuring of their experience based knowledge.

The SVARNE system has turned out to be a powerful educational tool when demonstrated and discussed during a lecture or seminar. In this setting it is also an efficient way of eliciting new examples and knowledge from the auditorium, a fact which already has been used in the development process.