EMOTIONAL SIGNIFICANCE IN MACHINE PERCEPTION AND COGNITION

摘要

Artificial cognitive systems utilizing associative distributed signal processing face the problem of combinatorial explosion. Attention process can be used to select the pertinent signal representations and guide the line of computation, but how is attention controlled then? It is proposed here that attention could be controlled by "emotional significance" generated by value-giving detections functionally similar to the sensations of pain, pleasure, goodness and badness. It is further proposed that generalized emotional values could also be associated to representations that do not initially involve any emotional value at all. Emotional significance would then constitute a kind of "emotional soundtrack" that would accompany memories and allow quick evaluation of the significance of new percepts, situations as well as imaginations.