Where is knowledge in computational intelligence?: On the reduction of the knowledge level to the level below

The fundamental conjecture in the present work is that it is actually possible to find effective procedures for reducing the knowledge level to the level below (from natural language to code) and vice versa, if the knowledge is structured. In order to support this conjecture, we prove that this procedure exists and is reversible at least for the case of hierarchical classification and its associated problem solving method (PSM) ("establish" and "refine"), according to B. Chandrasekaran's (1983; 1986) proposal and for neural networks. In conclusion, we emphasize two aspects: (1) the need for clarifying "where the knowledge is" with regard to any computation; and (2), our conviction that most of the work must be done by modelling knowledge on tasks and PSMs at the knowledge level and within the external observer's domain. This is where most of the knowledge remains, either explicitly or not.