Deontology and Tool AI — LessWrong