Chapter 10 Final Words

A reader who has followed the chapters in order has travelled a long way, and probably without noticing how long. We began with two parameters estimated on a straight line, and we ended with networks of several million weights that generate images. Between the two, the vocabulary changed, the tools changed, and above all the question changed.

This closing chapter does three things. It puts the methods of the book back into the chronology that produced them, because a technique is easier to judge when one knows what problem it was invented against. It isolates the thread that runs through all of them. And it ends with what seems likely to happen next, stated as an opinion rather than as a fact, since the history that follows shows how badly this field has been predicted before.