Superintelligence: Paths, Dangers, Strategies
Nick Bostrom · first published 2014 · ISBN 9780199678112
Bostrom takes the scenario of artificial general intelligence seriously and explores what happens if machines reach and exceed human-level cognition across all domains. He examines how an AGI might self-improve, what could go wrong, and what safety measures might matter. The book is rigorous and doesn't shy from uncomfortable conclusions.
On our lists
The Best Books About Artificial Intelligence
This 2014 work anticipated concerns that have only grown more relevant. While some specifics have shifted, Bostrom's reasoning about intelligence, power, and alignment shaped how researchers now think about existential risk. It's the skeptic's classic, and worth reading if you're serious about understanding downside scenarios.