Kinetic-Statistical Neuromodeling and Problems of Trust in Artificial Intelligence Systems

The focus of this article is related to cognitive sciences: the problems of modeling neural networks and consciousness in general are discussed. Examples of the inadequacy of artificial intelligence systems based on neural network modeling are given, and problems of the modern numerical model of neural networks are considered. In order to solve these problems, the methodology of consciousness neuromodeling based on kinetic-statistical methods is proposed. It is shown that, in the process of quasi-chaotic complication of the neuro-like graph during the percolation transition, large structures (clusters) are formed, which can be interpreted as a manifestation of the primary elements of consciousness. The manifestation of the phenomenon of self-consciousness in the process of complication of the system of neural clusters is discussed. It is possible to use the results of this research for generation of complex neural networks with criteria of finite structure evaluation (number of cycles, simple paths, and Euler's number).