Active Inference for Ethical Decision-Making in Socially Assistive Robotics

摘要

The aim of this contribution is to present the ethical aspects of decision making (DM) in socially assistive robotics (SAR) and discuss the model of active inference (AIF) within this context. The literature has extensively addressed the ethical issues related to SAR, especially concerning trust and justification in the choices and actions made by the robotic agent towards the user and the external environment. However, it is worth considering at a deeper level the technical implementation of DM. Models of AIF recently have been considered attractive for implementing explainable artificial intelligence for DM, with potentially groundbreaking implications for robotics. Indeed, AIF is particularly effective for complex tasks where the dynamics of the robot are uncertain, as these occur in human-robot interaction, and especially in SAR. This contribution tries to provide an encompassing account of the ethical implications of DM for SAR, by focusing on the often overlooked issue of the ethics of technical implementation. We first present the topic of DM for SAR, then discuss relevant ethical implications, and in a third step present the AIF model for DM and its applications in robotics. Finally, we consider whether AIF could serve as a model for implementing ethically sound decision-making processes in socially assistive robotics. (Paper presented at Workshop 8, Mind (and) the robot: on the epistemology and ethics of the attribution of mental states to robots, organized by Silvia Larghi.)