How can you trust Artificial Intelligence Systems? Objective, Subjective and Intersubjective parameters of Trust

Abstract

The article studies the most urgent problem of artificial intelligence (AI) - the development of a methodology for assessing trust in AI systems. A universal range of assessment is proposed: objective, subjective and intersubjective parameters. When studying the objective parameters of trust assessment, a formal definition of AI is introduced to increase the validity of trust judgments. When studying subjective parameters as the basis of expert evaluation indicators, their internal logical inconsistency is revealed along with demonstrating the significance of these parameters for analyzing the success and safety of methods for computer simulation of cognitive functions. When studying the intersubjective parameters of trust, a unified format of trust data is proposed, an interdisciplinary organization of monitoring trust in AI systems is considered, and the main blocks of an expert system project are formed.