Abstract

Artificial Intelligence (AI) is hard to sell as a matter of fact solution. Even knowledge systems, AI's economically most successful prodigy, have a bad reputation concerning their bottom-line. The main problems in the development of knowledge systems are the difficulty of knowledge acquisition and maintenance, and the gap that exists between prototype and industrial strength knowledge systems. These problems make their development a costly and risky enterprise, and create uncertainty as to the benefits of knowledge systems. This leads them to be perceived as experimental technology.

The wish for economically sound, principled development of knowledge systems coupled with a symbolic view of cognition has led to an ingrained engineering metaphor. This metaphor inspires the current approaches to these problems to employ structured methodology to design and realise them according to specification. Counter to this a science metaphor can be placed, motivated in part by situated theories of cognition. This metaphor sees knowledge system development as akin to working on a scientific theory, successively changed and subjected to critical review. This intends to indulge the learning character of knowledge rather than attempt to curb its fluid nature.

To investigate the differences between these two metaphors and their influence on the development of knowledge systems this research formulated a synthesis approach called 'continuous knowledge engineering'. This approach is oriented to support an ongoing learning approach through its guiding principles: participation of experts in knowledge modelling, project management as stewardship over the complete lifecycle, use of knowledge systems as a medium rather than a product, and a cyclic development process. The approach does not aim to replace current practices, but reorients them, augmented with additional facilities.
To make the approach practical, it requires support from development tools. Two tools were developed with this in mind. The first is a simple tool based on a visual knowledge representation, and is limited in its scalability. The second is a more advanced system, building on the accomplishments of the first tool. It extends the number of knowledge representations and introduces object-oriented domain models to enable vivid models. Both tools were put into daily practice at the Knowledge Based Systems Group at TNO in a number of case-studies.

These case-studies allowed analysis of these tools and their support for continuous knowledge engineering. The results show that experts are indeed able to participate in the modelling of knowledge. The case studies further demonstrate the viability of stewardship through a dedicated organisation. In addition, the support to develop a professional system as a medium from the first knowledge model onwards is shown, with the knowledge system growing in specificity and completeness as the knowledge itself develops. Finally, the case studies show a highly cyclic process of development. The benefits of the approach are a gradual knowledge acquisition process, and support for creation and evolution of an advanced knowledge system. This lessens the cost and risk of knowledge system development, while it increases the clarity on the realised benefits. This significantly improves the bottom-line of knowledge systems.