Why should we ever automate moral decision making?