Categories of subjectivity in artificial intelligence research: systematization of approaches and criteria of distinction

The light and shadows of digital reality. Living intelligence and its avatars

Abstract

(Novosibirsk)

Contemporary interdisciplinary AI research employs various concepts to describe the agential status of AI algorithms, often without clarifying the precise meaning of the terminology used. This leads to methodological uncertainty and disputes over the legitimacy of attributing agential characteristics to algorithmic systems. By systemizing the philosophical and sociological approaches, this paper develops a differentiated conceptual apparatus that specifies the conditions for applying each term, avoiding both excessive anthropomorphization of AI algorithms and their reduction to a purely instrumental status. The paper formulates four analytical criteria for systemizing the conceptual apparatus: ontological, epistemological, causal-agential, and contextual-relational. On this basis, the study demonstrates that the categories of actor and agent are primarily associated with the causal-agential criterion and characterize the functional participation of an AI agent in a specific interaction situation. The category of subject in the strict sense relates to all four criteria and presupposes phenomenal consciousness, reflexivity, and ethical agency, making it relevant primarily for describing hypothetical strong artificial intelligence. The concept of technosubject describes decentered, self-optimizing systems oriented toward operational efficiency under minimized human involvement; the concept of quasi-subject characterizes situations where an AI algorithm is attributed with individual features of a subject, with the specific set varying depending on the research context; finally, the term “situational subject” treats agential status as a temporary predicate, enabling analysis of interaction situations with dynamic role changes, where the same entity can alternately act as subject and object. The choice of a specific concept to describe an AI agent should be accompanied by the indication of its determining criterion, this preserving methodological and terminological precision. Such differentiation prevents conceptual substitution when analyzing human-machine interaction. The results demonstrate that abandoning a universal terminological template in favor of context-dependent choice of concepts significantly expands analytical capabilities in studying various types of AI systems.