A model ranks the job applicants before a human reads a single resume. A feed chooses what you will believe is happening in the world. A score sets your interest rate. A dashboard tells your team what is working, and the number gets trusted more than the people who watched the work happen. Each one arrives as a default, a suggestion, a score. And we wave it through without ever feeling that a decision was made.
That is the quiet handoff Who Decides? is about. Not a robot uprising. A slow, almost invisible transfer of judgment from people to systems, one reasonable convenience at a time.
The conversation about AI tends to arrive in two unhelpful flavors. One is the sales pitch, where everything is about to be transformed and any hesitation is nostalgia. The other is the alarm, where the machines are coming for our jobs, our truth, and our children. Both make for good headlines. Neither helps the person who has to decide whether to let the new system screen the applicants, set the prices, or answer the customers.
This book is for that person. If you build, buy, deploy, or sign off on automated systems, the chapters ahead are a field guide to one question that hides inside all of them. Not whether AI is good or bad, but where human judgment still belongs, and how to tell.
The systems that make decisions about people, most of the time without the people ever seeing it happen. Facial recognition and the wrongful arrest of Robert Williams. Recommender systems and what they optimize for when attention is the proxy. Tenant-screening algorithms and the story of Mary Louis, denied an apartment by a score she could not appeal.
The numbers underneath the decisions. The Cambridge Analytica scandal and what your own customer data is actually worth. Wells Fargo's "eight is great" and Goodhart's Law: what happens when the measure becomes the target, and why it always does.
The machine that writes. A lawyer who filed a brief citing six cases that did not exist. Air Canada's chatbot promising a policy the airline did not have. The Arup finance worker who transferred $25 million after a deepfake video call with people he thought he knew. And what happens to skill on the day the machine does the work for you.
Five questions to ask before you let a system make or shape a decision. Proxy: what is the number really standing in for? Incentive: who benefits if you trust it? Recourse: can the decision be questioned? Reliability: is the machine or the human the better judge here? Stakes: does the speed match how reversible the call is? A clean pass on all five means it is probably safe to lean on. Any clear fail is a decision to keep your hands on.
Almost every failure in the book begins with the same move. You cannot measure the thing you actually care about, so you pick something you can count and let it stand in. Cost stands in for need. Engagement stands in for value. Test scores stand in for learning. The system optimizes for the proxy, and the gap between it and the real thing becomes invisible. Learn to see the proxy, and you can see the failure coming.
A short toolkit at the back of the book. The Automation Trust Check as a working checklist. A vendor questionnaire for anyone about to buy a model someone else built. A verification protocol for confirming what is real when anything can be faked. Plus a closing note on the law and where it is going. Practical enough to take into a meeting tomorrow.
The machines will keep getting better at producing answers. They are already good, and they improve by the month. What they cannot do is tell you what an answer means, what it is worth, or whether to trust it. That judgment was never going to be automated. It is the part that turns data into wisdom.
So, who decides? Every system in this book offered to take that off your hands, and the offer always looked like help. Sometimes it generally was. But the task is knowing the difference, and keeping a hand on the wheel even when the car is steering well.
There is one decision left that no machine will ever make for you. Not whether to trust the model in front of you, but which decisions you keep, and which you give away to the machine. You can automate almost anything now, but you should not automate the choice of what to automate. When the system offers to decide, the real question was never about the system. It was always about you.