The ``distributed intelligence'' paradigm

The paradigm we sketch in this section aims at overcoming some of the drawbacks of the GEP. It is inspired to recent work in distributed AI, and its basic assumptions can be summarized as follows:

• knowledge is context dependent, and its representation depends on contextual factors such as purpose, background assumptions, available resources, and so on;
• linguistic heterogeneity and semantic differentiation are not accidents that must be eliminated, but essential properties of knowledge that must be dealt with in the process of knowledge integration.

The basic intuition is that locally produced knowledge cannot be represented by mapping it onto a universal structure, because we cannot assume that this structure is shared and understood in the same way by different knowers (or groups of knowers). This does not mean, however, that knowledge cannot be integrated. Only, a different mechanism should be used. If there is no single, objective perspective from which knowledge can be represented, knowledge integration can only be the result of a process of meaning negotiation between autonomous entities. Integrating knowledge is therefore a mechanism of social agreement. As a consequence, knowing appears as the intersubjective process of negotiating relations between symbols, and knowledge as the system of negotiated symbols and meanings at a given time and place. Given this general assumptions, knowledge becomes intrinsically social, as it cannot be separated from the social agreement that sustains its constitutive elements.

Many are the consequences of these assumptions. We will emphasize only those that are relevant for our purposes.

• First, since knowledge ``exists'' in the context of a negotiation process, knowledge has no existence when considered apart from its context. That is to say, contextual elements become constitutive of knowledge itself. As a consequence, any knowledge, when intended as a system of negotiated meanings and symbols, is to be considered as a two face entity of symbolic and mental representation within an interpretive context.
• Second, it is possible to imagine as many ``knowledge sets'' about the same domain as the possible interpretive contexts through which this domain is readable. As a consequence, the abstract concept of knowledge can concretely be viewed as a system of possible local knowledge sets.
• Third, learning appears, on the one hand, as the social process of generating, criticizing and reviewing interpretive contexts and, on the other hand, as the process of generating propositions within interpretive contexts. The problem of knowledge transfer and exchange involves the transfer and the accessibility of both symbolic propositions, and interpretive contexts through which symbols are able to gain a particular meaning.
• Fourth, diversity, redundancy and heterogeneity are seen as intrinsic qualities of a knowledge system. Basically, the many to many relationship that links a symbolic to a meaning system is the direct outcome of the many possible ways to read and represent the same domain. That is to say, a given symbolic system is able to represent as many thought worlds as the possible social perspectives through which a world can be red.
• Fifth, for our purpose, a context is an unspecified expression of social subjectivity. In other words, whether we assume a context to be a collective identity, a social paradigm, a task oriented use of a world or a mental model, the fundamental consequences of our description remain unchanged.

Within this framework, a KM system, viewed in its cognitive aspects, appears as a "distributed intelligence system", and the emerging paradigm becomes a distributed intelligence paradigm (DIP). With this term we refer to a system where knowledge is distributed between social actors, each social actor being involved in the process of constituting knowledge by considering knowledge domains through negotiated and subjective interpretive perspectives.

The development of KM systems based on the DIP poses a great variety of challenges for researchers and practitioners. First of all, we need to take into account the social aspects of cognition and investigate how knowledge is generated and represented through social structures, processes and roles. In particular, assuming the community as a lens to read organizations as social learning systems, a distributed intelligence system appears as a constellation of communities, each community being a social context and an interpretive perspective [17,4,23].

From a cognitive point of view, a second challenge is explaining how groups of individuals can communicate without sharing the interpretive contexts. In particular, assuming mapping processes as a mean to meaning making, a distributed intelligence system appears as a constellation of interpretive maps and of mapping processes between different individuals and communities.

Third, from a technological point of view, the representation of interpretive context and the analysis of meaning negotiation processes can be a valid basis for research in the intelligent agents field. In particular, a distributed intelligence system within a technology enabled communication environment can benefit of technological agents able to ``socially negotiate'' information under the light of locally and contextually represented ``knowledge''.

In this paper we do not address the first issue, as we are not personally involved in that research area (though we should mention the fact that the study of social aspects of cognition is part of the project within which the research presented in this paper was developed). Instead, we concentrate on the description of a formal framework for context-based knowledge representation, and sketch a high-level architecture of a multi-agent system for implementing a distributed intelligence KM system.