Is Moral Particularism a Problem for Rule-Based Artificial Normative Systems?

Abstract

One application of logics of moral reasons is to form the backbone of potential future normative AI systems. However, this application may be challenged by meta-ethical views that question the centrality of principles in moral reasoning, such as Dancy’s moral particularism. This article explores to what extent particularism is a problem for rule-based artificial normative systems. I first consider Horty’s argument that nonmonotonic logics can accommodate a core argument for particularism – the argument from holism – in a manner that employs default rules. I conclude that Horty’s argument is successful, but that particularism is motivated also by other arguments that non-monotonic logics can less readily accommodate. Specifically, it is plausible that the range of potential defeaters of moral reasons is such that these defeaters cannot be specified antecedently of the situations in which they occur, and therefore cannot be codified in moral rules. I then explore whether this situation can be mended. The proposal is that any sufficiently flexible reason-based normative system must be supplemented with a connectionist module. A family of formal systems is suggested, in which the connectionist module can identify different types of exceptions to moral default rules.