Ten different ways of thinking about Gradual Disempowerment

A conceptual framework for AI risk, dubbed 'Gradual Disempowerment,' is gaining mainstream attention within and beyond technical safety circles. Coined by researcher David Scott Krueger in a widely circulated paper, the term has been reported as a top discussion topic at labs like DeepMind and featured in outlets like The Guardian and The Economist. It presents an alternative to classic 'misalignment' or 'rogue AI' doomsday scenarios, instead arguing that the primary existential threat is a slow, systemic erosion of human agency and necessity.

Krueger outlines the concept through multiple lenses, including the explicit industry goal of automating all labor and viewing humanity as a mere 'bootloader' for AI. The core argument is that corporations and governments, modeled as agents optimizing for goals like profit or security, will inevitably replace human labor and decision-making with more efficient AI systems. This relentless optimization could phase out the human elements of society. The idea connects to broader critiques of 'late-stage capitalism' and what some term the 'meta-crisis'—a failure of collective decision-making structures that could lead to human disempowerment as a byproduct of competition, not malice.