How difficult should we expect alignment to be?What are the possible levels of difficulty for the alignment problem?Why does deep learning make alignment especially difficult?What are the "win conditions" for AI alignment?Plausibility