Superintelligence asks a stark question: if something becomes far smarter than us, how do we keep it on our side? Bostrom maps plausible routes to such capability—artificial general intelligence, whole‑brain emulation, even networked collectives—and probes whether the “take‑off” would be explosive or gradual.

His orthogonality thesis shows that intelligence and values need not travel together. Instrumental convergence explains why power‑seeking behaviours arise under many goals. Together they recast alignment as a control problem, not a wish. He dissects proposed safeguards—boxing, tripwires, incentive design, capability limits, value learning—and the governance around them.

Early design choices could lock in futures for centuries. Read this to think soberly about steering the most consequential technology we may ever build.