Causal Relationships as a Basis for Diagnosis and Decision Making

The effectiveness of modern AI applications, despite their tremendous capabilities, is limited in a number of crucial areas, such as security, transport, and medicine, etc., due to the inability to explain their decisions and actions. The Explainable AI (XAI) project was developed to create new or modified machine learning methods that combine high performance with decision transparency. One of XAI’s focuses is the development of data analysis techniques that study interpretable and causal patterns. These models, however, often inherit some of the characteristic problems of machine learning, in particular the need for a large and representative dataset. The paper proposes an approach that implements plausible reasoning by logical means to inductively generate causal relations from limited (but potentially updated with new data) datasets. Empirical induction is represented by formal specifications and extensions of J.S. Mill's inductive methods, based on the analysis of the algebraic similarity of objects with common properties. The revealed causal relations can provide a basis for explaining the mechanisms behind observed phenomena, and thus become a tool for diagnosing (or meaningfully classifying) and developing pragmatic solutions. These opportunities are particularly relevant for poorly formalized areas. Some results of generating cause-and-effect dependencies of different types for problems of medical diagnostics and social behavior analysis are presented. In medicine, these dependencies are used to predict the remission or recurrence of a disease and the formation of screening groups. In sociology, they allow us to make applied conclusions for social policies and managerial decisions.