Moloch

A name for the force that pushes competitors to sacrifice what all of them value, because anyone who refuses falls behind. Each actor would prefer that nobody made the sacrifice, but none can stop alone, so everyone ends up worse off. The situation is often called a multipolar trap.

Scott Alexander popularized the term in Meditations on Moloch (Slate Star Codex, 2014), borrowing the figure from Allen Ginsberg’s poem Howl. His examples include arms races, the Malthusian trap, and races to the bottom in which competitors cut wages or safety standards to survive. His point is that no villain is required: the outcome comes from the structure of the competition, not from anyone’s intentions.

Why it matters for AI

Racing is the standard worry about AI development. Each lab or country may prefer to move carefully, yet fear that a less careful rival will get there first. That is the dynamic behind Can AI progress be paused, Can the US and China agree to limit AI, and Should safety-focused labs keep building, and one reason Why people build AGI despite the danger.

It also bears on what happens after powerful AI arrives. A world of many roughly equal AI-empowered actors can sound fair and free, yet slide into the same kind of race. The usual escape is coordination: an agreement, a regulator, or a single dominant power. Each brings its own problems, such as the questions of accountability in Who should control aligned AI.

Related: Goodhart’s law, another way optimization pressure can erode what people actually value.