PrimEmo : a neural implementation of survival circuits supporting primitive emotions

Affective and cognitive sciences are both fields interested in the inner workings of the brain. While affective science focuses on the concept of emotions and how they are produced, cognitive science considers the brain as a whole, treating it as a system made of a multitude of independent subsystems. Since its inception, affective science has produced a plethora of theories and models each trying to solve the mysteries surrounding the definition and origin of emotions in the brain. Cognitive science on the other hand has had a rather chaotic history shifting its focus every decade, before finally being influenced by computer science and adopting the point of view of the brain as the most elaborate computational device. In this point of view, the gray matter residing within our skull is reduced to a system that takes sensory information on its inputs, processes it and outputs more data or actions. Each field has, more or less, evolved independently from one another so far, but both are now facing fundamental problems. On the one hand, the concept of emotions has yet to be completely defined and modeled. On the other, cognitive science is still trying to produce architectures imbuing artificial agents with human-level intelligence. This paper introduces a neural architecture based on the "survival circuits" framework (LeDoux, 2012) supporting primitive emotions and providing survival skills to artificial agents. Side-stepping the problem of defining the concept of emotions, the suggested neural structure focuses on identifying and modeling parts of the brain involved in survival functions (defense, thermo-regulation, maintenance of energy, nutritional supplies, reproduction and fluid balance). The neural implementation of this system provides a proto-brain for groups of artificial agents trying to survive in a dynamic virtual environment. By comparison with a hard-coded control logic (Scheutz, 2004), our architecture allows for a more equitable sharing of the resources and a longer life expectancy. It is our belief that, in the long term, the system suggested in this paper could become a robust basis upon which more elaborate cognitive architectures, such as ACT-R or SOAR, could be built, hence moving one step closer to endowing artificial agents with human-level intelligence.