Responsible Agency Through Answerability: Cultivating the Moral Ecology of Trustworthy Autonomous Systems.

The decades-old debate over so-called 'responsibility gaps' in intelligent systems has recently been reinvigorated by rapid advances in machine learning techniques that are delivering many of the capabilities of machine autonomy that Matthias [1] originally anticipated. The emerging capabilities of intelligent learning systems highlight and exacerbate existing challenges with meaningful human control of, and accountability for, the actions and effects of such systems. The related challenge of human 'answerability' for system actions and harms has come into focus in recent literature on responsibility gaps [2, 3]. We describe a proposed interdisciplinary approach to designing for answerability in autonomous systems, grounded in an instrumentalist framework of 'responsible agency cultivation' drawn from moral philosophy and cognitive sciences as well as empirical results from structured interviews and focus groups in the application domains of health, finance and government. We outline a prototype dialogue agent informed by these emerging results and designed to help bridge the structural gaps in organisations that typically impede the human agents responsible for an autonomous sociotechnical system from answering to vulnerable patients of responsibility.