Superintelligence — Nick Bostrom (2014)

Oxford philosopher Nick Bostrom's Superintelligence: Paths, Dangers, Strategies is the book that made AI safety a serious intellectual and policy concern among people beyond computer science departments. Its rigorous analysis of how a sufficiently advanced AI system might pursue instrumental goals in ways catastrophic for humanity — regardless of its programmed objectives — introduced concepts like "orthogonality thesis" and "instrumental convergence" that now dominate AI safety research.