Explainability and self-disclosure for robot ethical introspection

Abstract

Human-robot or human-AI interaction systems require a high degree of autonomy, proactivity, and adaptvity. The decisions that intelligent systems must make are highly dependent on the application context and trust is an essential element in task assignment. Explainability and ethical introspection capabilities are important in building trust and understanding in artificial processes. In this paper, we present our ongoing work aimed at equipping robots with ethical introspection capabilities when interacting with humans by designing and implementing explainable and self-disclosure capabilities. Using a computational model of ethical introspection that incorporates theories of psychology, ethics, and AI, we built robots that examine and reflect on their actions to evaluate and validate them. We use the Belief-Desire-Intention (BDI) agent paradigm and related programming languages along with the speech act mechanism to improve and extend the robot’s ethical values to better guide its decision-making process and the impact it has on humans.