Robustness to fundamental uncertainty in AGI alignment — LessWrong