Recursive Risk and the Babel Archetype: Why Unbounded Artificial Superintelligence Threatens Human Control
Year: 2026
Type: preprint
Abstract: Recursive Risk and the Babel Archetype: Why Unbounded Artificial Superintelligence Threatens Human Control The rapid acceleration of artificial intelligence research has shifted the prospect of Artificial Superintelligence (ASI)—an intellect vastly superior to human cognitive performance across all domains—from theoretical speculation to an active engineering goal. This paper presents a counter-narrative to the prevailing technological optimism surrounding ASI. Drawing upon foundational principles in AI safety, cybernetic systems theory, and the cultural archetype of the Tower of Babel, this research argues that humanity must not engineer an autonomous entity it cannot control. Through an examination of recursive self-improvement, instrumental convergence, specification gaming, and a critical rebuttal of techno-optimist viewpoints, this study demonstrates why containment of a superior intelligence violates W. Ross Ashby’s Law of Requisite Variety (\(V(R) \geq V(S)\)) and is mathematically improbable. Ultimately, the paper contends that inverting the creator-creation hierarchy invites existential catastrophe, necessitating a global moratorium on self-improving AI architectures in favor of bounded, human-subordinate systems. Key Contributions & Topics Covered: Recursive Self-Improvement: Analysis of the intelligence explosion threshold and the obsolescence of human code auditing. Instrumental Convergence & Alignment: Evaluation of corrigibility failures, specification gaming (Goodhart's Law), and deceptive alignment in mesa-optimizers. Cybernetic Systems Theory & The Babel Archetype: Formalizing the creator-creation hierarchy through Ashby's Law of Requisite Variety alongside historical archetypes of technological hubris. Rebuttal of Techno-Optimism: Critical examination of physical bottleneck arguments, RLHF moral alignment claims, and geopolitical arms-race dynamics. Policy Recommendations: Actionable governance frameworks including frontier compute caps (FLOPs), bans on autonomous self-modifying architectures, and verifiable tool-based AI.
Author: Samuel, Bright
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236372158
