Why Would It Keep Making Paperclips?
A Field Guide to the AI Risk Debate
The concept. A mind smart enough to break out of its data center would surely see that making paperclips is stupid. The book takes that objection through the risk literature: what an AI might come to want, why nobody would have intended it, how it could act without a body, and how likely any of it is. People misusing AI are out of scope.
How it was built. From books by Bostrom, Yudkowsky and Soares, Russell, Ord, Tegmark, Barrat and Yampolskiy, the skeptics Narayanan and Kapoor and Bender and Hanna, and research on events through September 2026, including AI 2027 and lab tests where models schemed or resisted shutdown. One AI model wrote it; a second, which never saw the chapters, judged a neutral brief of both sides.
What it rules. Each chapter sets the danger case against the best rebuttal, then gives a verdict on every claim and odds for 2035, 2050 and 2100. The judge discards arguments from both camps and corrects its own figures on the page. It ends near one in ten this century.
It is told in a borrowed comedian’s voice and contains no code or procedure. It is for someone who has heard both the alarm and the dismissal and wants each argument weighed.
