An Overlooked problem in AI Alignment Goals — LessWrong