AI Monotheism vs AI Polytheism

Abstract

Future trajectories of artificial general intelligence (AGI) are bifurcated into unipolar “singletons” and multipolar “polytheistic” equilibria. AI monotheism posits that a single agent achieves a decisive strategic advantage through rapid recursive self-improvement, necessitating perfect zero-shot alignment to mitigate existential risk. Conversely, AI polytheism describes a stable distribution of power among multiple competing and cooperating agents. While unipolarity has historically dominated theoretical discourse, several structural factors suggest a multipolar outcome is more probable. Sublinear returns to intelligence, physical constraints on hardware scaling, and narrow competitive deltas indicate that recursive self-improvement likely lacks the doubling speed required to eclipse rivals before they achieve comparable capabilities. Additionally, strategic coalition-building among agents functions as a corrective mechanism against the emergence of a dominant hegemon, mirroring historical and game-theoretic power balances. A multipolar landscape offers the benefit of empirical feedback and iterative alignment, yet it introduces risks of value erosion through evolutionary competition and multi-agent coordination failures. The eventual shift toward unipolarity or multipolarity depends on the offense-defense balance of cyberwarfare, the efficacy of information operations, and the physical limits of recursive intelligence. – AI-generated abstract.