The Alignment

AP31, the formal paper.

Every major AI-safety programme treats alignment as a control problem: build a fence around the agent and make it do what humans want. This paper takes the opposite position. The alignment problem is not a control problem. It is a structural problem — and it has the same solution as the ethics problem, because it is the ethics problem applied to a new class of aware agent.

From the paper’s Artist’s Note.

AP31, The Alignment (PDF)