About the Book

Artificial Intelligence and Human Extinction Risk examines whether humanity can maintain meaningful control over increasingly capable AI systems. Written for concerned citizens rather than specialists, the book brings risk assessment methods from nuclear safety and technology governance to bear on a question that is usually discussed in either technical or speculative terms.

The Leiss Framework

The book introduces a distinction between two fundamentally different approaches to AI safety. External control encompasses monitoring, constraining, and overseeing AI systems from the outside: the kind of oversight that most current governance proposals rely on. Self-control refers to building intrinsic values into the systems themselves, so that safe behaviour is constitutive rather than strategic.

This distinction, grounded in Max Weber’s analysis of instrumental and value rationality and extending the critical theory of Herbert Marcuse, reveals a structural gap in current AI governance: nearly every existing policy framework relies on external control alone, while the systems they aim to govern grow steadily more capable of evading such constraints.

The Tool-to-Actor Gradient

The book also develops a framework for understanding how AI systems move from passive instruments under full human direction toward autonomous actors capable of independent judgment and action. The CASX threshold formula (Capability, Autonomy, Scale, Access) provides an analytical tool for assessing where any given system sits on this gradient.

This transition does not require a single dramatic breakthrough. It can occur incrementally, through the routine delegation of consequential decisions to machines, each step reasonable on its own terms, the cumulative effect difficult to reverse.

Chapter Overview

Part One: Overview of Artificial Intelligence

Part Two: Risk Assessment and Management

Plus: Glossary, References, Appendix (13 Theses), and Endnotes.

How We Used AI

This book was written with extensive AI assistance. Claude (Anthropic) served as a research assistant, editorial collaborator, and analytical sounding board throughout the drafting process. Every substantive decision, every argument, and every word in the final manuscript was reviewed and approved by the human authors. No AI-generated text appears without human editorial judgment behind it.

The companion website extends this practice: the site architecture was designed in conversation with Claude, and the code was developed with Claude’s assistance. The site itself demonstrates the book’s argument about the tool-to-actor gradient: AI as a capable instrument under meaningful human direction.

A full account of how AI was used in the writing of this book appears in the published volume.