Temporal-causal methods for constructing explanations in artificial intelligence systems

The subject of the research is the process of constructing explanations in artificial intelligence systems. The goal is to develop a temporal-causal approach to constructing explanations in artificial intelligence systems to present explanations both for the decision-making process and the obtained decision, and to make them transparent and understandable for solving practical user tasks. Tasks: structuring the levels of explanation representation considering temporal and causal aspects; developing a generalized method for constructing explanations using temporal and causal dependencies; developing a method for refining explanations using temporal dependencies. The structuring of the explanation representation in temporal and causal aspects at the local, intermediate, and global levels has been performed. The scientific novelty of the obtained results is as follows. A temporal-causal method for constructing explanations is proposed, which includes the stages of constructing temporal and causal dependencies at the local, intermediate, and global levels of explanation representation. The method makes it possible, based on temporal dependencies, to form an explanation in the form of causal dependencies that determine the actions of the process and the values of input variables as the causes of the obtained solution, which creates conditions for increasing the level of user trust. A method for constructing explanations at the global level of representation based on the ordering of input data has been developed. The method includes the stages of constructing temporal rules, determining the weights of rules based on the weights of the antecedent and consequent, constructing causal rules, constructing an explanation as a set of weighted causal rules, which makes it possible to consider the structure of input data when constructing an explanation.