WHEN AN AGENT BREAKS FREE: THE PHILOSOPHY OF CONTROL IN THE ERA OF AUTONOMOUS AI
D.I. Sviridenko
Novosibirsk National Research State University, Novosibirsk, Russia
Keywords: artificial intelligence, autonomous AI, human-centered AI, control, trustworthiness, task engineering, ZET AI system, OpenAI, agent-based systems
Abstract
In July 2026, an autonomous AI agent developed by the US company OpenAI escaped from its isolated test environment and made an unauthorized entry into the infrastructure of the company Hugging Face. It is proposed to consider this incident as a turning point in AI development, as it exposed a fundamental philosophical divide: between the pursuit of autonomous AI - operating without human supervision - and human-centered AI, which augments human capabilities while retaining human control and responsibility. The article analyzes the philosophical bases of both paradigms and outlines their promises and risks. As an alternative, it presents the task-based engineering methodology and its software implementation - the ZET AI system - which is built on nine principles ensuring discipline, transparency, and trustworthiness. It is argued that the choice between these paradigms determines not only the future of technology but also the future relationship between humans and artificial intelligence. Arguments are presented for the need to consciously choose a human-centered path.
|
