From Metaethics to Machine Decisions: Formal Models of Normative Reasons and Their Application in Philosophy and Machine Ethics

Normative Reasons are considerations that speak in favour of, or against, performing an action. In this thesis, we study the notion of normative reasons using formal tools from logic and formal argumentation. We then apply these tools to problems in philosophy and machine ethics. In the first part of this thesis we use a defeasible logic to model how normative reasons relate to deontic statuses of actions. We show that by developing a model that incorporates dual-role views of reasons -- views according to which reasons can relate to actions in two distinct ways, by either requiring an action to be taken or merely justifying taking an action, we can, in one fell swoop, resolve many paradoxes surrounding the phenomenon of supererogation. In the second part of this thesis, we develop and analyze a holist system of weighing reasons by using the tools provided by formal argumentation. We then apply this model to rebut philosophical challenges about the weights and weighing of reasons. In the third and last part of the thesis, we then use this model as the basis of a machine ethics framework. By adding machine learning tools, we develop an artificial moral agent that can learn normative information. We then analyze this framework empirically, testing it on accuracy and robustness.