Problems of Development of Artificial Intelligence, its Errors and Hallucinations. Psychoinformational (Socionical) and Quantum Methods of their Elimination.

The causes of errors and hallucinations of artificial intelligence models are analyzed. The work of artificial intelligence is compared with the structure of the psyche, which is well described by the socionic model of information metabolism by Augustinavichyute–Bukalov. This model has been successfully used in the analysis of the work of the psyche for over 40 years. It is shown that modern AI models lack an independent block for monitoring the implementation of established rules and assessing the importance of the information being processed. It is also necessary to rank the processed data according to specified criteria. It is shown that the most pressing problem of AI - the emergence of hallucinations - is due to an increase in semantic entropy in closed probabilistic models. This problem is fundamentally unsolvable without changing the principles of construction and operation of AI. In addition, AI models lack a number of structures, including aspects of true creativity, which are an integral part of the real psyche and socionic models. In the context of the creation of quantum computers, the quantum aspects of creating a new type of AI are also considered, by analogy with the work of the psyche. Such computers must have a fundamentally different architecture, compared to existing ones, and consist of a whole system of specialized quantum processors.