A Three-Facet Framework for AI Alignment

What it is

Essay proposing three distinct facets of AI alignment: control (the system cannot cause unacceptably bad outcomes even if trying), intent alignment (the system tries to do what the user wants), and values alignment (the system refuses widely unacceptable actions). Uses a stock-trading agent scenario to show how the facets can conflict.

Topics