Gradations of Agency — AI Alignment Forum