Date of Original Version

Winter 2-2018



Published In

AAAI & ACM Conference on Artificial Intelligene, Ethics and Society 2018

Abstract or Description

We propose a deontological approach to machine ethics that avoids some weaknesses of an intuition-based system, such as that of Anderson and Anderson. In particular, it has no need to deal with conflicting intuitions, and it yields a more satisfactory account of when autonomy should be respected. We begin with a “dual standpoint” theory of action that regards actions as grounded in reasons and therefore as having a conditional form that is suited to machine instructions. We then derive ethical principles based on formal properties that the reasons must exhibit to be coherent, and formulate the principles using quantified modal logic. We conclude that deontology not only provides a more satisfactory basis for machine ethics but endows the machine with an ability to explain its actions, thus contributing to transparency in AI.



Published In

Toward Non-Intuition-Based Machine Ethics.