The Epistemic Dependence of Artificial General Intelligence and the Limits of Its Existential Risk

Abstract Artificial General Intelligence (AGI) is often portrayed as a potential existential threat to humanity. However, this paper argues that such fears are largely speculative and misrepresent the fundamental limitations of AGI systems. We examine how AGI remains epistemically dependent on existing human-generated data, lacks autonomous goals, and cannot operate beyond the physical and logical constraints imposed by its human creators. The core risks associated with AGI are sociotechnical and arise from human misuse or design flaws, not from AGI independently seeking power or dominance. This paper concludes that, while AGI poses governance and safety challenges, it does not currently present a true existential threat in the absence of deliberate human error or malice.