The article addresses the normative definition of artificial intelligence as provided in the Russian legislation by 2021 in the context of general approach to this term. Authors identify principal legally significant features of artificial intelligence that include autonomy of functioning (complete or relative), the ability to find solutions (including without predetermined algorithms) and “inorganic” nature that imply - strictly from formal legal, and not factual, perspective - potential risk and, subsequently, potential danger of artificial intelligence to be understood in formalized legal sense. The authors take into consideration examples of approaches to defining the term in the world, raise criticism of the existing definition and suggest more concise approach based on defining artificial intelligence as an information system (in formal legal sense) which provides the function of forming algorithms for solving problems different from those laid down by the system developers, including artificial neural network