Toward the problem of unexplainability of artificial intelligence decisions

The article investigates the fundamental principles of modern artificial intelligence systems in the context of their cognitive abilities. The main attention is paid to the problem of correlation between efficiency and transparency of intellectual systems. The study considers the phenomenon of decreasing explainability of solutions when the complexity of artificial intelligence systems increases. The author analyzes this phenomenon not as a technological limitation, but as a fundamental property of complex cognitive systems. The study is based on a comparative analysis of decision-making processes in natural and artificial intelligence, which makes it possible to identify common regularities of their functioning. Special attention is paid to the study of parallels between natural cognitive processes and the work of modern neural networks. Current problems of verification and validation of intelligent systems under conditions of their increasing complexity are considered. The influence of digital transformation on the development of new approaches to assessing the reliability of artificial intelligence systems is studied. A verification system based on four key criteria has been developed: stability of results, correctness of behavior in boundary conditions, robustness of solutions and the possibility of external validation. The existence of a fundamental relationship between the efficiency of intellectual systems and the inevitable decrease in their explainability is proved. The necessity of transition from traditional verification methods to more flexible approaches that take into account the specificity of complex cognitive systems is substantiated. The proposed methodology opens new perspectives for development and implementation of artificial intelligence systems in the conditions of digital transformation of society. The study makes a significant contribution to the development of the philosophy of artificial intelligence and offers practical solutions for assessing the reliability of intelligent systems. The results of the work create a theoretical basis for further development of verification methods for complex cognitive systems.