Author(s) :

Rachel Rishita.

Volume/Issue :

Abstract :

As Artificial Intelligence continues to evolve at a rapid pace, the concerns regarding its capabilities circling around existential risk have increased at a similar rate. The concept of instrumental convergence primarily – the theory that intelligence AI agents may become indifferent to the programming and commands of humans and prioritize self-preservation – have alerted individuals, as this could endanger humanity. However, this paper critically evaluates the extent of the validity of instrumental convergence as a real-world AI safety concern by arguing that it overextends AI’s capabilities and is driven by anthropogenic assumptions. AI agents lack the fundamental characteristics such as intrinsic motivation, agency, and the capacity for self-generated terminal goals which would otherwise drive the theory of instrumental convergence. Moreover, other barriers to AI alignment are discussed in this paper, specifically the orthogonality thesis and the tricky notion of aligning complex human values with agents. Counterarguments to popular X-risk narratives are presented to support the conclusion that instrumental convergence is overstated. My findings suggest that a more accurate public and professional understanding of AI’s limitations and capabilities would bring us closer to achieving a safe and effective relationship between humanity and AI.