In the last ten years, knowledge management (KM) has become a new fashioned managerial practice. Under this name, managers foresee the opportunity to control the processes of producing, distributing and using a ``new'' valuable resource: knowledge [6]. This complex matter cannot be reduced to the traditional concepts of land, labor and financial. As a matter of fact, we need a ``contamination'' process between managerial practices and theoretical disciplines that are historically devoted to the study of knowledge and cognition, such as philosophy, artificial intelligence (AI), social sciences, psychology.
From a managerial point of view, this process of contamination is not linear, and somehow contradictory. On the one hand, cognitive and social disciplines increasingly regard subjectivity, contextuality and distribution as intrinsic features of knowledge. On the other hand, following a traditional approach to organizational life, current KM systems seem to implicitly rely on an objective view of knowledge. In this situation, managers are facing the substantial failure of KM systems and have the intuition that this failure is somehow related to inadequate assumptions about the nature of knowledge (see e.g. [16]). However, though managers are attracted by the emerging approaches to knowledge, they are unable to accept the underlying paradigmatic shift. The result is a situation where KM systems are nominally consistent with emerging issues on the nature of knowledge, but are still substantially based on traditional assumptions.
Current KM systems are inspired to a traditional paradigm in organizational life based on the idea that the purpose of a KM system is to represent and organize knowledge into a single, shared, and coherent, structure (e.g. a taxonomy, an ontology), independently of when, how, where, and why it was originally produced. We call this paradigm, that descends from a traditional approach to cognition in social systems, the ``god's eye'' paradigm (GEP). The emergent paradigm in KM, which we call the ``distributed intelligence'' paradigm (DIP), is based on the idea that knowledge is always and irreducibly distributed into multiple contexts of knowledge production (individuals, groups, time periods and spatial locations, and so on), and therefore cannot in general be straightforwardly organized into a single, shared and coherent structure. This dichotomy between paradigms cannot be discharged as a purely philosophical issue. Adopting one view or the other inevitably leads to completely different conceptual and architectural choices in designing a technology enabled environment for KM, and these choices can be the reason for the success or the failure of a KM system. In this paper we discuss artificial intelligence (AI) theories and technologies that can support a shift from the GEP to the DIP in designing KM systems. We first discuss the basic assumptions of the GEP and the DIP using the evolution of KM systems within Arthur Andersen Consulting as our case study. Then we propose the framework of MultiContext Systems as a specification language for distributed intelligence KM systems, and sketch an agent-based architecture as an example of DIP based KM system for representing and integrating knowledge.