Clarifying some key hypotheses in AI alignment

Abstract

This work argues that AI alignment is a crucial issue that requires careful consideration, and it presents a diagram that maps out important and controversial hypotheses related to this topic. The diagram includes key ideas such as the concept of rational agency, goal-directedness, comprehensive AI services, and modularity over integration. The authors aim to help researchers identify and discuss their disagreements, potentially enhancing the productivity of discourse on AI alignment. – AI-generated abstract.