Elements of a cybernetic epistemology: output-side attention directing systems.

Purpose – The purpose of the paper is to analyze cybernetic necessities of output-side attention directing systems, i.e. how systems can decide to act towards one of various inputs. Design/methodology/approach – Complex pattern recognition and sequence learning systems may recognize more than one pattern and deliver more than one output at a point in time. Therefore, they require an output-side attention directing system to decide to act towards just one pattern. The necessary cybernetic structures of such systems are analyzed using a functional approach. Findings – An output-side attention directing system has to evaluate the effect of current observations (patterns, sequences, etc.) on highest level goal-values (in a living system these are existential goal-values like a body temperature or energy supply). Measure of this effect is the degree of goal-approximation towards these goal-values. This measure can either be preprogrammed for some patterns or sequences, or has to be determined in trial and error processes for new patterns or sequences learned by the system. Practical implications – The paper shows the cybernetic necessities of the development of the "know how" of sequence learning systems in time, starting with default behavior, via learning new patterns and sequences, and trial and error to develop goal-orientated actions towards them, until finally the achieved results enable experience based directing of attention. Originality/value – The paper shows basic cybernetic structures and functions for output-side attention directing systems required for all complex pattern recognition and sequence learning systems. [ABSTRACT FROM AUTHOR]