Artificial Intellect and Methodology of Economic Science

Abstract

The goal of the research is to analyze general methodological approaches used in economic theory and tools of artificial intellect, in particular machine learning. According to the authors there is a certain similarity between debates about realistic assumptions of economic models and the acute discussion concerning interpretability in using machine learning in real conditions. The widely used concept of the model as a ‘black box’ casts doubt on validity of using such models both in machine learning and in economic theory. Apart from that, the issue of public trust to applying algorithms and elaboration of evidence-based policy become hardly attainable without relevant interpretation of model functioning, in spite of hypothetic accuracy of forecasts made with their help. In machine learning it is impossible to rely only on ‘black box’ functioning and neglect the role of man in decision-making.  In the same way in economic science the essence of phenomena described by models is often sacrificed to laconism and harmony of mathematic and econometric tool set. Analysis of academic publications on methodological problems in economics as a science allows us to draw a conclusion that in spite of (or due to) the empiric turn in economic theory Milton Friedman instrumental approach is still prevailing, at least in its latent form.