Answerable

A woman in Michigan had just finished her training to become a police officer in Detroit. Then she failed the background check, and the job went away.

What the background check had flagged was a finding of fraud. Years earlier she had collected unemployment benefits, and a state computer system had decided, on its own, that she had lied to get them. She had never been told. The notice had gone to an online portal she had no reason to be checking, and when she did not respond within thirty days the decision became final and could not be appealed. By the time it surfaced on the background check it was simply true on the record: she was someone who had committed fraud. No person had ever looked at her case. No human had decided anything about her at all.

The system was called MiDAS. Michigan's Unemployment Insurance Agency installed it in 2013 to decide, automatically, whether a person had committed fraud in claiming benefits. That judgment could strip the benefits, quadruple the bill as a penalty, seize tax refunds, and garnish a quarter of a future paycheck, with no court and no hearing. In the same stretch, the agency laid off most of the workforce that had been reviewing applications and claims, nearly its entire fraud-detection unit. The humans who used to make the call were gone, and the call still had been made.

It was made wrong most of the time. Roughly forty thousand Michiganders were wrongfully accused, and later reporting and reviews placed the false-determination rate for the automated cases somewhere between eighty-five and ninety-three percent. At least eleven thousand families filed for bankruptcy. The accusation was not of a clerical error but of intent, of lying, and intent is the kind of thing we long believed only a person could weigh. Michigan had handed that judgment to a system and removed the people who might have caught it being wrong.

Undoing it took the better part of a decade, and the undoing was only partial. Litigation, federal scrutiny, and eventually new legislation forced the state to stop making fraud determinations by machine alone and to put a human back into the review. Holding anyone to account for what had already happened proved harder than the harm had been to inflict. One class action ended in a twenty million dollar settlement that reached only a few thousand of the people involved. Other claims were closed on immunity and procedural grounds, the kind of technicalities that eventually hold no one accountable, long after the damage is done. Tens of thousands had been wrongly accused. Far fewer were ever made whole. The one thing that actually fixed the machine was the simplest thing imaginable. Put a person back into the decision.

That is the whole of this book in a single case. A consequential judgment was delegated to a system with no accountable human owner, and the cost fell on people who were never in the room where it was decided, because by then there was no room and no one in it. This book is about that room, and about the person who is supposed to be standing in it. It is about the call.