Territorial AI Safety
Mapping and measuring the resilience of places to the erosion of human agency
Preface
This is the open, ongoing research record of territorial AI safety. It is published continuously: the version you are reading reflects the latest commit, and sections will be incomplete, revised, or wrong. That is by design, the work is meant to be built and argued with in the open.
Status: v0.x · See the roadmap in the Dialogoo white paper.
What this book is
A place is more or less resilient to the erosion of human agency depending on its structure: how distributed its decision-making is, how redundant its communication channels are, how dense its civic life is, how exposed it is to capture by a single mediating layer. This book defines that idea, turns it into a transparent and reproducible evaluation index, and maps the result.
It is one project under the Dialogoo umbrella, and the source of the materials produced by Dialogoo’s Communicate pillar.
How it is organized
- Foundations why agency erosion is the central risk, and why it is territorial.
- The framework what territorial AI safety is, and the pillars and indicators that operationalize it.
- Methodology how the index is built, tested for robustness, and matured from sketch to instrument.
- Application the pilot, the map, and the materials for community leaders.
Methodology chapters contain live code: the numbers in the book are produced by the same pipeline described in the text (see 9 Constructing the index).
How to read the living version
The HTML site rebuilds on every push. To run it yourself, see the README. quarto preview serves a live-reloading copy locally.
How to contribute
The most useful contribution is an attack on the framework: tell us where an indicator fails to measure what it claims to. Open an issue, suggest a source, or propose a territory for the pilot.