A critique of the GEP

The GEP has some obvious advantages, as it promises of eliminating from the outset problems such as conceptual or semantic heterogeneity between different representations. However, our intuition is that the GEP simply begs the question of knowledge integration, without facing the real complexity of the problem. Indeed, it does not address a number of crucial issues, such as the following:

• first of all, it underestimates the complexity of knowledge within a big organization. Designing a single structure (e.g. a taxonomy, an ontology) where every piece of knowledge produced within an organization can be coherently placed is very hard work, and it is not obvious at all that it is feasible in practice, when the approach is scaled up to large and complex organizations (such as AA);
• second, the GEP is not elaboration tolerant, namely it is not adequate to cope with highly dynamic environments. Small changes in the organization may require a very complex revision or even a complete re-design of the entire knowledge structure, and this is not desirable;
• third, it overrides the categories, concepts, terms that a professional group uses in its everyday activity. This could be a reason for a sub-optimal usage of the system, or even a complete failure;
• finally, perhaps the most serious issue is the assumption that knowledge representation can be completely de-contextualized by mapping local (idiosyncratic) conceptual schemas into a general (shared) conceptual schema.

The fact that knowledge (and its representation) is irreducibly contextual has been advocated by several leading researchers both in AI [15,14,11,9] and in other knowledge related disciplines [7,5,18] (see [3] for an interdisciplinary collection of papers on this topic). The common intuition is that the content of any linguistic representation depends on a collection of assumptions, on the purpose for which it was produced, on the intentions of an agent, and so on (a similar argument can be made for non linguistic representations, but here we are not concerned with this aspect of the problem). Whatever the right explanation for this irredicibility (the non existence of a so called reality, its inaccessibility through our perception, of its complexity as a constraint to understanding), the immediate consequence is the impossibility to establish a non ambiguous relation between words, meanings and objects. The meaning of a word is not always self evident. Thus it is perfectly conceivable (and we see it happen in everyday life) that speakers who share the ``same'' language do not necessarily share its semantics, and so associate different content to the same word.

This is what happened within the AAOnLine approach. Different professional communities produce (and represent) knowledge autonomously, use idiosyncratic lexicons, different concepts and implicit taxonomies. Managing knowledge is not just a matter of creating a shared repository, or having a team of Knowledge Managers that impose on it a general structure. We need new concepts that allow us to deal with conceptual and semantic heterogeneity. In short, we need a different paradigm.