II. Anthropic reasoning with duplication is not consistent with probability properties

LessWrong is reporting on a fundamental impossibility in anthropic probability — the field of how agents should reason about their place and identity. Researcher Stuart Armstrong has identified a mathematical contradiction that emerges when an agent is duplicated. His thought experiment: an agent goes to sleep, a coin flip determines whether she awakens alone in one room or is duplicated across two rooms. Bayesian probability works consistently before the coin flip and after the agent learns which room she's in, but duplication itself — the moment the single agent becomes two — creates an inescapable mathematical conflict. The problem is unique to duplication; creating new agents or death present no such issue. Armstrong identifies Self-Indicating Assumption, or SIA, as the strongest available anthropic probability framework, though he notes it breaks down with infinity and proposes a distributional refinement to fix that problem.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton