Coherence-driven Reflective Equilibrium model of ethical decision-making

Abstract

Instilling trust in autonomous systems requires having confidence that agents have decision-making mechanisms governed by ethical principles. The ability to resolve conflicts among moral principles, duties, obligations, and the consequences is a critical challenge in developing such mechanisms. We demonstrate how the Reflective Equilibrium Method and its implementation in terms of a parallel constraint satisfaction mechanism can help manage conflicts and simultaneously assess multiple principles in a context-sensitive manner. The proposed domain architecture and its implementation are used toward developing an Ethical Advisor for training purposes as well as for designing cognitive decision-making models. A high-level Domain-Specific Language is introduced to specify coherence-governed models of ethical decision-making, so that domain experts express concepts relevant to modeling ethical behavior instead of using general-purpose programming language constructs.