A conceptualization and general architecture of intelligent decision support systems

The paper presents an attempt to conceptualize decision support and various generic subtasks and to develop a general architecture of intelligent decision support systems. We decompose the task of decision support into subtasks whose input, output, and function are characterized. This is based on a small number of concepts: besides "decision", the essential ones are "observation", "situation", "goal", "action", and "process", which are in turn defined using elementary concepts for characterizing the system under consideration, or our model thereof. This is not an academic exercise aiming at providing definitions, but a prerequisite for a generic architecture of decision support systems with interfaces for certain generic functions, the comparison of basic modules implementing these functions, and the configuration of systems from a set of such modules. The primary subtasks whose (intelligent) solution is heavily dependent on domain knowledge are situation assessment, i.e. inferring what is happening in a system from a set of observations, and therapy proposal, i.e. developing plans for interventions to achieve certain goals starting from the current situation. Secondary tasks are situation and plan evaluation (checking whether and to what extent a situation or plan satisfies or violates goals), prediction (forecasting the future development starting from a situation with or without interventions), and observation/experiment proposal (designing activities to collect information, possibly after stimulating the system in a particular way, useful to disambiguate situation assessment and also situation evaluation).