Decision AI
Start here
What the book is, the ways in and the volumes.
Decision AI is a set of case studies on decision models: systems that take evidence and typed questions and return a choice, a rubric score or a probability that application code acts on. Each chapter takes one business or workflow decision and poses it as questions. It compares the policies built on the answers with rules, trained classifiers and language models, and prices the result with costs that the reader can change. Each chapter ends with what its data cannot show.
Every statistic in the text is printed from the chapter's data file, not typed by hand. Hover over a number or focus it to see its unit, population, period and source, and its interval where it has one. Costs that a reader sets in a scenario are marked as assumed, and quantities measured from the data are marked as measured.
All the chapters are drafts, and their figures may change as the analysis is checked.
7 of 7 chapters are written, all of them drafts that are still being checked. The other 0 are planned: each shows the question it will ask and the data it will use, and states no result.
Ways in
All chapters lists every chapter with its finding and status. Explorers are interactive tools that sit beside the parts. Reference holds the data sources, glossary, figure gallery and methods.
The reading guide explains the kinds of claims the chapters make, how to read a figure and an interactive scenario, and what the grain of a chapter's data means.
The parts
Volume 1
Case studies
Case studies that follow a model's judgment into an action, from reading a review to deciding when an agent's work is done, each compared with a rule or a trained classifier.
Part I
From judgment to action
What a decision model returns, what labels buy, and why a good ranking pays only when its probabilities can be trusted near the cutoff.
Part II
Scarce attention
Who should get a person's limited time, on a complaint desk and in a queue of code changes, and how far a tidy test set carries to real work.
Part III
Acting and checking
A case where each stopping policy was run to the end of the task, so a decision changes the evidence that follows it.
Part IV
Deciding whether to adopt
What the cases support across them, and how to weigh a system against the simple alternative.