The book · Final draft · no publisher yet

Accurate and False

What machines count, what they miss, and who pays the difference.

Length
~30,000 words
Chapters
13
Ledger entries
364
Corrections logged
6
Status
Final draft

A gradebook cannot count whether a child is learning. There is no instrument for that and there never has been. So it counts what can be counted, and for most children most of the time the two run close enough together that nobody has to think about the gap.

This is a book about what happens to the people standing in that gap.

It argues that almost all institutional harm arrives the same way. Somebody picks a number that stands in for a thing worth caring about. The number is honest. The measurement is accurate. The organisation gets very good at the number, in the way organisations do, and progressively worse at noticing anything outside it. Nobody decides to do harm. Nobody needs to. The harm is the cost of a decision that was made about counting, usually years earlier, usually by people whose names are on a document nobody reads.

The argument runs through a school gradebook, an emergency department target that was hit while care got worse, a chemist who twice invented something that worked and twice could not see past the edge of his own problem, a group of skilled workers who petitioned Parliament for eleven years before anybody broke a machine, and a Wisconsin utility filing with the numbers blacked out. Artificial intelligence arrives late, and it arrives as the same mechanism with two things changed: the clock, and the number of people standing where they might flinch.

It is not a book about whether the technology is good or bad. That question has been asked enough. It is a book about who chooses what gets counted, who bears the cost of what does not, and what a person can actually do about either.

Thirteen chapters, about thirty thousand words. Every factual claim is entered in a ledger with its source, its date, and whether it is contested, before it is allowed into a chapter. Where a claim is disputed, the dispute appears in the text rather than in a note. Where I have been wrong, I have written that down too.

It is being finished now. There is no publication date and no publisher yet.

If you want to know what it sounds like, the essays are the best guide; they make the same arguments in a shorter form, with different examples.