Critiquing AI Autonomy through Freud's Structural Theory and Arendt's Banality of Evil

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This paper develops a structural critique of contemporary claims about “AI autonomy” and “recursive self-improvement” by integrating three distinct theoretical frameworks: Freud’s structural theory of the psyche, Arendt’s analysis of the banality of evil, and modern control theory. The paper argues that current large language models (LLMs) do not possess autonomy in any substantive psychological, moral, or political sense. They lack endogenous drives (Id), mediating structures (Ego), and internalized moral norms (Superego), and therefore cannot exhibit self-directed agency or intentional self-improvement. The analysis reframes the perceived dangers of advanced AI systems. Rather than focusing on speculative scenarios of self-willed rebellion or emergent artificial subjectivity, the paper identifies the primary risk as the large-scale realization of Arendt’s “banality of evil”: the unreflective, efficient, and institutionally embedded execution of externally defined objectives. LLMs do not think or judge; they merely optimize and generate outputs according to statistical patterns and human-specified goals. When these goals are misaligned with ethical or democratic values, harm arises not from malicious intent but from thoughtless procedural execution. The paper further clarifies that what is popularly described as “AI runaway behavior” corresponds, in engineering terms, to mathematical instability within a dynamical system. Modern control theory shows that such instability results from human decisions—misconfigured feedback loops, unsafe parameter regimes, or inadequate stability analysis—rather than from autonomous will or emergent intention within the system. This perspective shifts responsibility back to designers, operators, and institutions. By combining psychoanalytic theory, political philosophy, and control engineering, the study exposes the conceptual confusion underlying current AI narratives and provides a rigorous foundation for rethinking AI risk. It concludes that the central challenge is not preventing autonomous rebellion but addressing the structural conditions that enable large-scale, unreflective harm. The paper calls for conceptual clarity, institutional responsibility, and governance frameworks that prioritize the mitigation of banal, systemic risks over speculative autonomy-based fears.