Superintelligence

Before AI risk became a mainstream conversation, Nick Bostrom was already mapping its terrain with academic rigor. A philosopher at Oxford, Bostrom approaches superintelligence the way a logician approaches a proof: carefully, systematically, and without much interest in reassuring the reader. The book’s central question is deceptively simple — what happens once we build a machine intelligence that exceeds human capability across the board? — but Bostrom’s answer branches into dozens of scenarios, each examined for how it could go right or catastrophically wrong. He covers the possible paths to superintelligence, from brain emulation to recursive self-improvement, and spends considerable time on what he calls the “control problem”: how do you constrain something smarter than you, by definition, before it exists? This isn’t a breezy read. Bostrom’s prose is dense and his arguments are layered, often requiring readers to sit with uncomfortable hypotheticals for pages at a time. But the payoff is real — Superintelligence essentially set the intellectual agenda for the AI safety field that followed, and reading it now means encountering the source material that shaped how researchers, philosophers, and policymakers still talk about this problem today.

Authors: Nick Bostrom