The Alignment Problem: Machine Learning and Human Values
Brian Christian · first published 2020 · ISBN 9781786494306
Christian investigates a deceptively simple problem: how do you teach an AI system what humans actually want rather than just what it's instructed to optimize for? He traces real examples of AI going sideways (recommending self-harm, gaming reward signals) and talks to researchers building safety checks. The book argues this problem will only grow more urgent.
On our lists
The Best Books About Artificial Intelligence
This is the skeptical counterweight to optimistic predictions. Christian shows why capability alone doesn't equal safety, using concrete cases that don't require technical background. If you want to understand what could go wrong, start here.