Neural and behavioral substrates of emotions as actions to reduce embodied dissonance

Toward better understandings of emotions, we argue how emotions should be synthesized, from phenomenological, evolutionary-psychological, and embodied perspectives. In particular, based on the assumption that an emotion consist of appraisal process and reaction/expression process, we propose a novel model of primitive emotions, in which an appraisal mechanism exploits informations of "embodied dissonance" existing as conflicts among desired and actual states. The model consists of action modules representing motivation, and evaluation modules representing actual states, both of the modules are coordinated by a constraint satisfaction neural network which behaves by a minimization principle of its neural energy. By using a physically simulated bipedal creature, we show that the interplay of neural and body-environment dynamics leads to syncope/fainting-similar states which emerge as adaptive responses to embodied dissonance, that "drastic and random movements" coupled with dissonance leads to the transitions of stimulus-response patterns from approach to avoidance, and that the transitions let the creature to cope with high-order problems, the causal structure of which is not perceived nor recognized by the creature. Taking these results into account, we discuss three things; (1) a principle to discriminate emotional and non-emotional cold behaviors of artificial agents; (2) a prospective architecture of appraisal processes of animal\´s emotions; and (3) a prospective account for the difference among fear, anxiety, hysteria, anger, and sadness. This study may shed light on a link of emotions and intelligence.