BRAIN MODELING, NEURAL NETWORKS, ARTIFICIAL INTELLIGENCE

Abstract

This article is a generalised analysis of the knowledge of modern neuropsychology and neurophysiology. The article considers the possibilities of studying the work of the human brain using neuro-modelling on modern computers. The ways of further development of artificial intelligence research are shown. The theoretical approaches to creating models of different levels of detail, including biophysical ones, of individual neurons based on Hodgkin-Huxley equations, as well as simplified large-scale models of neuronal populations and brain regions are considered. Models of the latter type are used to reproduce the complex systemic effects of the physiological activity of a neural network, and they have a limited number of parameters and thus allow for their qualitative mathematical analysis. The simplicity of such models is achieved to the detriment of the rigour of the conclusion; they are derived either phenomenologically or strictly from the equations of single neurons, but under excessive simplifying assumptions. These limitations lead to the impossibility of quantitative reconciliation of models with experiments. Therefore, it is important to develop a theory of transition from biophysically detailed models of single neurons to models of cortical neuronal populations. At the same time, such detailed models require a consistent reduction to simple models suitable for analysing functional mechanisms. When applied to the analysis of the activity of any of the brain structures, the hierarchy of transient models, on the one hand, makes simplified models more reasonable and provides them with an interpretation in terms of intracellular characteristics, and, on the other hand, makes it possible to use the results of the analysis of simple models to study complex ones. The purpose of the article is to analyse topical issues related to attempts to model the human brain using neural networks.