The article describes the principles of application of self-learning artificial intelligence systems in the “Program package for modelling the city as a dynamic system”. According to the authors’ perceptions, the city’s functioning as a system is defined by the accessibility and availability of resources and energy flows. Accounting for the high complexity of the city functioning principles, traditional working methods with artificial intelligence designed for determined situations are practically not appropriate. The structural and parametric presupposition of traditional approaches will reflect the perceptions about the subject of operators training artificial intelligence and will not allow objectively representing the full complexity of the city functioning as an open dynamic system. With a large number of interrelated parameters concerning different types and hierarchical levels of information, it is desirable that artificial intelligence can be adaptive and self-learning. The approach developed by the authors to forming a program complex with the integration of artificial intelligence systems inside it will allow to connection diverse types of information in a single data field. In this case, this will make it possible to form the group cognitive maps about the principles of city functioning as a dynamic system. In addition, such an approach will allow to identify “agents of changes”, indicators of crises, sustainability and pre-catastrophe states, and optimise resource consumption required for positive transformations of the city system. Thanks to the extensive capabilities of artificial intelligence, the program complex potentially becomes a valuable tool for the formation of step-by-step strategies in the implementation of sustainable development goals.