Функциональная агентность искусственного интеллекта: типология и градуальный подход

Человек. 2026. Т. 37. № 5. С. 42–58.

Karpenko I.

This article examines the question of agency in artificial intelligence systems. A distinction is proposed between three types of agency: ontological, functional, and ascribed. The central thesis is that functional agency is a genuine rather than metaphorical type, which is grounded through the teleological theory of function and through a critique of the argument from derivative intentionality. A principle of gradualism is introduced: instead of posing a binary question about consciousness and understanding (yes or no), an empirically oriented question is proposed: “to what extent and in what sense.” The article accepts skeptical arguments that large language models lack phenomenal understanding, but shows that this does not affect the question of access understanding in the sense of N. Block. Drawing on the global workspace theory (with reservations about its applicability to feed-forward architectures) and acknowledging the seriousness of objections from integrated information theory, the article argues that the question of AI consciousness is empirically open. Based on the analysis of agent architectures (ReAct, Reflexion, generative agents), it is established that contemporary AI systems satisfy the key criteria of functional agency. The distinction between three types of agency combined with the principle of gradualism makes it possible to escape the impasse generated by the binary questions “Is AI an agent?” and “Does AI have consciousness?”

Keywords: искусственный интеллект, сознание, понимание, consciousness, understanding, agency, агентность, градуализм, Gradualism, большие языковые модели, large language models, artificial intelligence, functional agent, global workspace theory, функциональный агент, теория глобального рабочего пространства