Emotion Eliciting and Decision Making by Psychodynamic Appraisal Mechanism

摘要

We describe an architecture named Psychodynamic Cognitive Construction (PCC) for artificial systems to develop affective attitudes from the history of human-machine interaction. The evolving emotion states are used as an appraisal for decision making. Psychodynamic appraisal mechanism is proposed based on our understanding that emotion is a physically grounded, dynamically constructed, and personally experienced process. Different from other mechanisms, the mapping between a stimulus and affective attitudes, as well as behaviors, is not predefined, but learned. A semi-embodied system, named qViki, is built as an implementation of PCC architecture. The emotional incentive makes the system trying to live for its own well-being and improve its constructs of the external world according to its experience. The fluctuation of tensions, which is induced by innate needs or acquired anticipations, drives associative learning and expressive behaviors. A growing network is devised to memorize and retrieve its past experiences. The preliminary experiment shows how the system's attitude towards certain stimulus is elicited and regulated by interaction. We discuss the role that emotion may play in affective agents as a motivation system to facilitate cognitive development through social interaction.