- Deep neural networks (primarily convolutional neural systems) are based on the principles of the human and animal visual system (LeCun et al., 2015),
- Reinforcement learning, a central approach in AI, is based on the theory of classical conditioned reflexes by I.P. Pavlova.
From the point of view of neuroscience, the structure of artificial neural networks (ANN) and the rules for their training are quite primitive.
This circumstance largely limits the functionality of ANNs in AI.
At the same time, most AI research is trying to build automatic decision systems that can be used in medical diagnostic systems, unmanned vehicles, pilot assistance systems (Davenport and Harris, 2005; Karanasiou and Pinotsis, 2017; Vatansever et al., 2017 ).
However, these automatic systems are based on well-known approaches that make it possible to approximate any continuous bounded function and do not use modern neuroscience data in any way.
In the laboratory of neurobiology of action programming, recent years have been focused on the study of functional neuromarkers of cognitive control in humans.
The second part of the report will deal with the experimental and theoretical work of the author in this area (see recent monographs in Academic Press, Elsevier, Kropotov, 2008, Kropotov, 2016).
In particular, the neural networks responsible for the formation of the current model of the world, maintenance of working memory, conflict detection, decision making, and suppression of irrelevant actions will be described.
Speaker: Kropotov Yury Dmitrievich, head lab. Institute of the Human Brain RAS, Professor, Laureate of the State Prize of the USSR, visiting professor at the universities of Norway, Switzerland and Poland. Author of more than 300 scientific publications, 13 monographs in Russian, English, German and Polish. Hirsch index - 29. In recent years, he has been studying the neural mechanisms of cognitive control.