Causal Relationships in Explainable Artificial Intelligence

Abstract

The problem of explainability of artificial intelligence models has been solved for a long time by classical methods of explanation, generated by even more classical methods from the field of feature space analysis. This approach shows which of the parameters of the observed objects in the initial data set have the greatest influence on the decision being made (for example, in the problems of classifying brain MRI images by the presence of a disease). However, in the answer to the question about the parameters that have the greatest influence on the decision being made, there is no answer to the question about the reasons for the decision being made (it often takes a doctor a lot of time to explain to the patient the need for a particular action, for example, surgery. The problem of determining the significance of parameters is known due to the rich the history of decisions in the field of feature space analysis and is not essentially new.