Ref : Colloc J. (2022) « A Multi-Agent Model Approach to Fill up the Gap Between Emotion, Psychologic and Brain Computing », in proceedings of SCIFI-IT’2022, ed by P. Geril and  M. Polanska, EUROSIS, University of Leuven, campus Ghent,  Belgium, 3-6, pp 8-15

Auteur :

Joël COLLOC est professeur émérite des universités en Informatique, membre du Laboratoire UMR IDEES Le Havre. Docteur en médecine, ancien médecin légiste des hôpitaux de Lyon, il est docteur en informatique de l’INSA de Lyon. Ses recherches concernent l’intelligence artificielle (IA) en e-Santé, la sémiologie, le raisonnement clinique. Il préconise une éthique de l’IA respectueuse de  la relation patient-soignant, la vie privée et la liberté de choix des patients. Il s’intéresse aux systèmes complexes, la nature de la pensée, aux évolutions de l’IA générative autonome (Chatbots) vers la conscience artificielle et aux conséquences sociologiques pour  l’avenir de l’humanité.

Résumé :

Deep learning and artificial neural networks are used to simulate synapses and recognition features of brain perception. However, they do not take into account, the dynamic, the plasticity and the diversity of the inter-neuron local communication: electrical, chemical and paracrine synapses. Moreover, these models do not implement the glia that conveys long distance hormonal messages that interact with neurons and make them available everywhere in the whole body at the same time. These internal hormonal messages strongly influence the state and the behavior of all the brain layers and nuclei (fear, emotions, sleeping and awakening). We have shown in our previous article that these hormonal messages are essential to consciousness and thinking. We know that the nervous system is always evolving and that this evolution is necessary to achieve long- term memory, cognitive, sensitive, motor and language tasks. We have also described the features of the object oriented subsymbolic perception pyramid and those of the object oriented linguistic pyramid. In this work we provide a junctions object ontology to implement the different type of synapses and hormonal messages and a multi-agent system that integrates and control them. The supervisor agent represents anatomical and connection constraints of the brain layer and nuclei and control the diffusion of hormones and neuromediators. The temporal fuzzy vector space (TFVS) is used to tune the composition of neuron objects in layers and nuclei objects and to implement the emergence of their states and features according to a holistic systemic approach. We present the necessary tools and TFVS object classes to implement the MAS subsymbolic perceptive and psychological layers. To illustrate the approach we propose a simulation of object perception and recognition, emotion tagging, indexation to store the cognitive experience corresponding to a set of threatening objects in the environment

Source : https://hal.science/hal-03661339/document