Rethinking Transparency as a Communicative Constellation

In this paper we make the case for an expanded understanding of transparency. Within the now extensive FAccT literature, transparency has largely been understood in terms of explainability. While this approach has proven helpful in many contexts, it falls short of addressing some of the more fundamental issues in the development and application of machine learning, such as the epistemic limitations of predictions and the political nature of the selection of fairness criteria. In order to render machine learning systems more democratic, we argue, a broader understanding of transparency is needed. We therefore propose to view transparency as a communicative constellation that is a precondition for meaningful democratic deliberation. We discuss four perspective expansions implied by this approach and present a case study illustrating the interplay of heterogeneous actors involved in producing this constellation. Drawing from our conceptualization of transparency, we sketch implications for actor groups in different sectors of society.