New methodology for measuring information, knowledge, and understanding versus complexity in hierarchical decision support models

There is much on-going effort to develop new methods for paring down complexity in decision support models (DSM). Many of these methods are so intricate and prone to bias introduction that they are rarely used. The first part of this work developed, tested, and evaluated a new methodology called Knockout (KO) for pruning unnecessary complexity from a Testbed DSM. Complexity is defined as the number of semantic nodes in the DSM. Unnecessary complexity is the maximum quantity of complexity that can be pruned without violating the requisite DSM fidelity. KO identifies all of the semantic nodes that make up the DSM, and determines their individual semantic contribution to DSM fidelity in a manner that avoids bias introduction. The node of least semantic significance to DSM fidelity is always pruned first. KO is shown to efficiently prune complexity from a Testbed DSM, pruning complexity by 36% while reducing fidelity by only 1%. Thus, the first result of this work is a new methodology to enable organizations to trade DSM fidelity for a reduction in DSM complexity.;The second part of this work used KO to investigate the ratio of information nodes (parameters) to knowledge nodes (functions) as the complexity of a Testbed DSM was pruned. The a priori expectation was that this work would support one of two learning models in the literature: (1) the bottom-up model known as the Wisdom Hierarchy in which information is accumulated prior to the mental construction of knowledge, or (2) the top-down model known as the Reverse Knowledge Hierarchy in which knowledge is accumulated prior to the mental construction of information. But this work found that the baseline Testbed DSM (the full DSM prior to pruning) has nearly an equal number of information and knowledge nodes (188 to 191), and the ratio of information-to-knowledge remained within a few percent of unity as the DSM's complexity was decreased by successive pruning of the least-semantically-significant node. Thus, the second result of this work is a new model of human goal-driven learning in which information and knowledge accumulate simultaneously and contribute equally to model fidelity and complexity.