In this position paper we present our view on how AI techniques could contribute to the learning capabilities of an organization. The analysis is based on a particular model of organizational learning which we call the knowledge pump. For each of the components of the knowledge pump it is discussed which Al techniques could potentially support it.
Kenniscentrum CIBIT is a Dutch non-profit organization which, amongst other activities, provides consulting services in the areas of knowledge management and knowledge technology. In our work, we make a distinction between three viewpoints on knowledge management: the strategic, the logistics and the learning viewpoint (Van der Spek and Spijkervet, 1996).
From this viewpoint, knowledge management addresses issues as, "what knowledge does our organization have?", "which knowledge do we need in the future to achieve our strategic goals?", "how can we acquire this knowledge?" etc.
From this viewpoint, the main target knowledge management is efficiency. Typical questions addressed from this viewpoint include "how efficiently does our organization apply its knowledge?", "is knowledge available at the right time and the right place?", what are the knowledge bottlenecks and how can they be solved?" etc. From this viewpoint, the relation between knowledge and business processes is the main object of study.
Here, the emphasis is on the ability of an organization to learn from experience, including the experiences of others, and to distribute these lessons learned throughout the whole organization.
Each viewpoint has its own set of typical problems and its own set of tools and techniques to solve these problems. In this position paper, we will articulate our approach with respect to problems associated with the learning viewpoint.
When looking at an organization from the learning viewpoint, we often make use of a thinking model for knowledge infrastructures which we call the "knowledge pump" (Van Heijst et al, 1996; Kruizinga et al, 1996). In this model, which is depicted in Figure 1, learning is viewed as a cyclic process of collecting experiences, analyzing and organizing these experiences, using the new knowledge to update the organizational memory, and then distributing the new insights to relevant parts of the organization. For each of the activities in the knowledge pump various techniques are available, some of which have their origins in artificial intelligence and knowledge technology. In the sequel of this position paper we will discuss these.
The term work experience covers such a wide area of activities that it is not possible to associate this part of the knowledge pump with particular Al tools or techniques. Any tool or technique&emdash;based on artificial intelligence or not&emdash;that facilitates experimentation on the work floor could be useful. In some situations, one could consider simulation techniques of the kind developed in qualitative reasoning as a way of facilitating experimentation.
In most of the cases, the most difficult part cf collecting lessons learned is to motivate coworkers to submit their experiences in the first place, or to extract them in some other way. Al is not of much use here. However, in situations where lessons learned are collected, and when they are available in an electronic form, techniques developed in natural language processing can be used to recognize lessons learned about particular subjects. In our own practice, we have used such techniques, for example, to recognize interesting case descriptions in a psychotherapeutic setting, by making use of an ontology of symptoms and diagnoses in the area of geriatric psychiatry.
The main goal of this activity is to generalize individual experiences to make them applicable in a broader context. In cases where large numbers of lessons learned documents are generated, and where they have a standardized format, machine learning techniques can be used. ln other cases, manual knowledge acquisition tools are more applicable. Both machine learning and manual knowledge acquisition can be facilitated by the availability of a body cf background knowledge (usually called bias in ML and skeletal models in KA). Therefore, also knowledge modeling techniques as developed in knowledge technology can be of help here.
Intuitively, the issue of storing lessons learned is related to the issue of knowledge representation in art)ficial intelligence. However, perhaps surprisingly, we have so far not seen a situation where it was worthwhile to formalize the lessons learned to the extent which is typically required by knowledge representation formalisms as developed in art)ficial intelligence. Typically, lessons learned databases, or more generally, corporate memories, consist of structured documents with associated attributes. In some cases, the documents are related through hyperlinks.
The low utility of knowledge representation theories for corporate memories is in our view due to the fact that companies that develop a corporate memory tend to concentrate on the knowledge that will give them a competitive advantage. Typically, this is knowledge which is not yet fully christalized, and which is likely to change rapidly. The first factor makes it difficult to formalize the knowledge, and the second factor makes it uneconomic.
Which Al-techniques are useful for retrieving knowledge from the corporate memory depends on how the corporate memory is structured. In our own practice, we have mostly worked with companies where the corporate memories were organized as flat case bases. In such situations, the obvious AI-technique to use is case based reasoning. We have applied this technique succesfully in a number of situations (e.g. in the psychiatric setting mentioned above). The advantage of using case based reasoning as the main mechanism in a corporate memory is that it requires only a limited effort of the employees. A disadvantage is that the newly acquired knowledge is not explicitly articulated in the form of rules and guidelines. This may decrease the learning speed of the organization.
As was the case with the first step in the cycle, not much can be said about the utility of AI techniques for this activity.
In the above, we have described how AI techniques can be used to enhance the learning capacity of organizations. By no means we want to imply though, that these are the only AI techniques that can support knowledge management. When taking the knowledge logistics perspective, expert systems technology is often a useful tool. Furthermore, in the context of strategic knowledge management --- where scenario planning is an often used technique&emdash; qualitative simulation tools could be used to compute the possible implications of future developments. With the latter, however, we have not yet experimented.
van der Spek, R., and Spijkervet, A. 1996, Knowledge Management, Dealing Intelligently with Knowledge. Kenniscentrum CIBIT, Utrecht, The Netherlands.
van Heijst G., van der spek, R. and Kruizinga, E.1996, Organizing Corporate Memories, paper presented at the 10th KAW, workshop for knowledge acquisition, knowledge modeling and knowledge management, Banff, Canada
Kruizinga, E., van Heijst G. and van der Spek, R. 1996, Knowledge Management and Knowledge Infrastructures, Sigois Bulletin, Vol 17, No 3 (December 1996).