Artificial Morality: Top-down, Bottom-up, and Hybrid Approaches

Reviewer: John M. Artz

The premise of this paper is that as we increasingly rely on artificial software agents, it becomes increasingly important that we can rely on them as artificial moral agents. And this, in turn, requires that morality be designed into the software itself. The paper offers three approaches to achieve this. The top-down approach begins with moral principles and derives moral behavior from those principles. The bottom-up approach provides moral experiences from which the software can learn. The hybrid approach is a combination of the two. Unfortunately, the premises upon which this paper is based are absurd. For example, software does not possess freewill, and hence cannot make moral decisions. Second, software does not possess a moral sense, and hence cannot learn to be moral. Third, the approaches suggested in this paper, if applied to humans, would be of questionable value, since the questions of "What is moral behavior__?__" and "How do you get people to behave morally__?__" have been open questions in philosophy for as long as there has been philosophy. There is a quip among ethicists: "When in doubt, do the right thing." It is funny because it trades one difficult question, "What should I do__?__" for another equally difficult question, "What is the right thing__?__" without making any real progress. This paper attempts exactly the same slight of hand. Question: "How do you teach software agents to behave morally__?__" Answer: "Top down, bottom up, or a hybrid of the two." Online Computing Reviews Service