Brézillon P (1997)

Preface of the Special Issue on Successes and Pitfalls of Knowledge-Based Systems in Real-World Applications.

Failures & Lessons Learned in Information Technology Management, 1(2).


There is a number of industrial Knowledge-Based Systems (KBSs) since the implementation of the first expert systems. However, we do not know how many KBSs are really operational within companies or administrations. Durkin (1993) lists about 2500 systems, but their exact operational status is often not precisely stated. Majchrzak and Gasser (1991) point out that more than 50% of the systems, which are installed in companies, are not used. A reason is that KBSs solve 80% of problems when users mainly need a support for the other 20%. Another reason is that integration of KBSs in an organization implies interaction of KBSs with classical softwares and changes in the organization itself. A special issue of the AI Journal (Volume 59, 1993) presents the most successful systems. Mizoguchi and Motoda (1995) say that, in the top five Japanese companies, there are from 20 to 30 operating KBSs. Thus, some clues about system use exist. However, other systems--successful and unsuccessful ones--represent considerable experience that may provide many useful lessons for the community. Successes and failures occur for a variety of reasons that, once shared, may permit others to save time, work and money. Papers in this Special Issue deal with such problems.

At the International Conference on Successes and Failures of Knowledge-Based Systems in Real-World Applications (Bangkok, Thailand, October 1996), one finds some interesting conclusions. For example, the knowledge and reasoning of a KBS is not a simple translation of those of the human experts. Beyond the problem of the knowledge-engineer's translation of the expert's knowledge, there are deep problems. A problem is that structures generally are imposed on knowledge bases and new concepts are introduced by the knowledge engineer. Such a macro-reasoning is useful for an incremental development of the knowledge bases but does not correspond to the human expertise. Another problem is the alive nature of knowledge that implies that a KBS must be able to incrementally acquire knowledge during the problem solving. The incremental knowledge acquisition permits the KBS to have the knowledge and its context of use. Moreover, KBSs do not exist in a technological context alone but are embedded within dynamic human organizations. Thus, they cannot be completely designed prior to use and must evolve in the hand of the users.

For Ware, KBSs solutions are perceived as the ultimate solution to management problems within the business, some of which were mission--critical under organization--wide restructuring. Most of these projects related either to management issues precipitated or exacerbated by the restructuring, or to the impending loss of irreplaceable expertise from the reorganization. The author thus presents a number of tentative conclusions concerning KBSs development teams, relating to issues of politics, leadership, KBSs personnel, client perceptions, project selection and management, development procedures and platforms.

Pomerol and Brezillon think that most of the references in the literature do not address some important dimensions as the differences between stand-alone KBSs and interactive KBSs, the types of data acquisition that is required and the nature and the role of future and preferences in the involved decisions. They note also that most of the previous failures and difficulties have already been encountered in several domains, as decision support systems, in which interactivity plays a crucial role. An interactive system must be able to cooperate in a way that does not constrain the user.

Migliarese proposes an information-system design method based on the analysis of the system's organizational impact. The author considers three organizational aspects: (i) the analysis of decisional requirements of managers, (ii) the changes the DSS causes in users work habits and (iii) the changes the DSS causes in organizational power equilibrium. The main concern is the power equilibrium changes: the analysis of organizational relations overcomes some of the limits of the classical organizational perspectives to political analysis of information-system introduction.

Burrell and Duan encountered limitations of KBSs through their involvement in the development of a system for strategic marketing planning. The original purpose of the research was to develop a knowledge based system to support the strategic marketing planning process. They found that a knowledge based system alone was not suitable for some types of tasks. For the authors, clearly one needs hybrid systems in which a KBS works jointly with other types of system as a decision support system.

Fischer's claim concerns the need to emphasize a human-centered and domain-oriented approach in developing KBS and their use as intelligent support systems. A keystone for the KBSs is the necessity to be able to evolve when the knowledge change, the goal of the task changes, etc. The author discusses in this paper a component architecture (the multifaceted architecture) and a process model (the seeding, evolutionary growth, reseeding model) underlying DODEs by focusing specifically on their support for evolution. Two applications developped for voice dialog and computer network design are described. Stathis and Sergot present a framework for building complex interactive systems from simple components, with particular attention to the development of Knowledge-Based Front-Ends to software packages. The work was motivated from problems with the interactions of GLIMPSE and FAST, two Knowledge-Based Front-Ends developed using logic programming tools and techniques. They propose to view interaction as a rule governed activity that may be usefully regarded as a game.

Frasson and Aïmeur present the lessons learned from a project concerning Intelligent Assistant Systems, project that is developed in cooperation between universities and industries. They show which problems have to be avoided to maintain a coherent and progressive development approach. The main point of the authors is the need to account for the industrial dimension of any KBS as soon as possible in the KBS development. This implies a twofold effort: a long term effort (e.g., for research) and a short term effort with immediate goals and results.

Killin and Curet analyze four KBSs that have been developed in the finance domain by their company since 1988. Three of the systems are intended to highlight possible fraud risks, the other makes decisions in the domain of new business underwriting (which includes an element of fraud risk identification). They present an evaluation of the positive and negative points of these systems.

The relationship between user and interactive system would be one of controller and assistant and would not threaten the autonomy of the user. This implies that a KBS must: - Rely on some concepts of biological systems as self-organization and self-adaptation, without trying to mimic biological systems but transforming biological concepts into computer concepts; - Be able to learn (e.g., learning-by-doing) and incrementally acquire fresh knowledge in order to improve its support to users; - Be equipped with a number of functions as symbolic learning and incremental knowledge acquisition. For example, a KBS must be able to store its own experiences with users to be able to improve its support; - Be comprehensible to users as any computer system. This implies in turn that the KBS must a transparent box, not a black box. Several paths are now explored as intelligent interfaces, intelligent assistant systems, active environment, etc. This is the new challenge for the future.

REFERENCES

Durkin, J. (1993) Expert Systems. Catalog of Applications, Intelligent Computer Systems Inc., PO Box 4117, Akron, Ohio 44321-117, USA.

Majchrzak, and Gasser, L. (1991) On using Artificial Intelligence to integrate the design of organizational and process change in US manufacturing, AI and Society Journal, 5, 321-338.

Mizoguchi, R. and Motoda, H. (1995) Expert systems research in Japan, IEEE Expert Magazine, 14-23.