agency-problems
Inefficiencies in human-run organizations arising from divergent individual preferences, which AI systems might avoid due to perfectly loyal components.
1 chapter across 1 book
Superintelligence: Paths, Dangers, Strategies (2014)Nick Bostrom
Chapter 5 explores the possibility of a single superintelligent AI project gaining a decisive strategic advantage over competitors, potentially enabling it to dominate the future and establish a singleton global order. It analyzes how the speed of AI takeoff—fast, medium, or slow—affects whether multiple projects can advance concurrently and how gaps between frontrunners and followers might evolve. Historical technology races and factors such as diffusion rates, imitation, and organizational efficiencies are used to contextualize plausible time lags and strategic advantages in AI development.