When AI is Wrong

And How We Must Put It Right

By: Richard Murch

Narrated by: Virtual Voice

This book is a reckoning.

It traces the history of artificial intelligence — from the optimistic experiments of the mid-twentieth century through the long winters of disappointment to the extraordinary and bewildering revolution of deep learning. It examines, in granular detail, the ways in which AI systems fail: not only how, but why, and with what consequence.

It examines evidence from criminal justice, healthcare, finance, employment, government, and the information ecosystem, and documents what happens to real people when the machine is wrong.

But documentation alone is not enough. The second purpose of this book is to argue — clearly and without equivocation — for what must change. Not what might change. Not what would be nice. What must.

Because the alternative to serious reform is a world in which the errors documented in these pages become not cautionary tales but permanent features: embedded in infrastructure, normalized in institutions, and insulated from challenge by the very complexity that makes these systems powerful.

That world is coming closer. Every month, new AI systems are deployed at scale. Every month, new populations of people are subjected to machine judgments they did not consent to and cannot contest.

The window in which we can insist on something better — in which we can demand accountability, transparency, redress, and genuine safety — is open, but it will not remain open indefinitely.