Abstract

A rule-based ethical theory is a good candidate for the practical reasoning of machine ethics because it generates duties or rules for action, and rules are computationally tractable. Among principle- or rule-based theories, the first formulation of Kant's categorical imperative offers a formalizable procedure. We explore a version of machine ethics along the lines of Kantian formalist ethics, both to suggest what computational structures such a view would require and to see what challenges remain for its successful implementation. In reformulating Kant for the purposes of machine ethics, we consider three views of how the categorical imperative works: mere consistency, commonsense practical reasoning, and coherency. The first view envisions straightforward deductions of actions from facts. The second view incorporates recent work in nonmonotonic logic and commonsense reasoning. The last view takes ethical deliberation to follow a logic similar to that of belief revision.