Intelligent task solver self-learning approach using biological systems principles

The development of artificial intelligent systems usually requires a huge investment of time and effort from the developer to provide them with a small fraction of the capabilities of biological prototypes. Building knowledge models of intelligent systems is the most complex and time-consuming process. The application of biological systems principles is intended to accelerate progress in this direction. Biological systems have effective learning and, most importantly, self-learning abilities with the formation of skills and abilities, the rapid application of which helps them survive in an aggressive environment. Neural network mechanisms of memory and decision-making are capable of self-organization, rebuilding neural structures based on the accumulation of knowledge in changing environmental conditions. The organization of memory based on an adaptive neuron, capable of independently adjusting the ratio between various combinations of various signals coming to the inputs and signals of different shapes and levels at the output, allows even one element of the neural structure of the brain to store large amounts of relatively “simple” information. When interacting with other neurons, structural relationships are established between combinations of elementary information, forming experience, skills, and abilities, the generalization of which leads to the emergence of knowledge. Providing artificial intelligent systems with self-learning abilities should qualitatively change the process of their construction. An approach to improving self-organizing control systems based on the principles of biological systems, multilevel memory, mechanisms of structuring and generalization is proposed. The simplest example of the implementation of these principles is the use of plans for solving “simple” tasks as components of plans for solving more complex tasks. The use of structuring, generalization, and axiomatization of solutions to known problems allows you to create new knowledge that can then be used to solve new problems. An algorithm for solving problems based on the use of multilevel memory, structuring, generalization and axiomatization of procedural knowledge is proposed. An example of solving a declarative set task using self-learning is considered, demonstrating the effectiveness of the proposed approach.