Causality, Transformative AI and alignment - part I — LessWrong