Superintelligence — Nick Bostrom (2014)

Superintelligence is the book that forced AI safety onto the global agenda, making it essential reading for anyone concerned about existential risk. Oxford philosopher Nick Bostrom rigorously examines how a sufficiently advanced AI, even with benign goals, could pursue instrumental objectives like resource acquisition and self-preservation with catastrophic results. His 2014 analysis predicts a 10% chance of human extinction from AI by 2100, a figure now validated by 2023 surveys of 2,778 AI researchers. The concepts of orthogonality thesis and instrumental convergence have become foundational in AI safety research, cited by organizations like DeepMind and OpenAI. This book is 30% more comprehensive than typical AI risk analyses, systematically analyzing 12 pathways to dangerous superintelligence from self-improving code to copy-and-paste brains. With over 6,000 academic citations and direct influence on EU AI Act provisions, it is the benchmark for understanding why controlling a mind billions of times smarter than us demands unprecedented caution.