Transparency’s Role in AI Governance

Abstract The adoption of artificial intelligence in the public sector has generated a lively academic and policy debate about what sort of transparency is needed to govern AI systems. This chapter explores the background to this debate. It explains what transparency means and why it is expected of public organizations. The chapter further highlights the big questions in the debate about AI transparency, such as who is being transparent about what, how public values are influenced by AI, and whether explainability of AI systems is a feasible way of delivering transparency. There are four main ways that scholars have tried to answer how the demands for transparency should be met: constructivist, democratization, legal, and capacity-building. Each of these approaches views the matter of transparency from the point of view of a different institutional locus. Another area of divergence is between approaches centered on audiences as citizens or audiences as experts. But, far from being contradictory, each of these points of view can play a part in the effort to develop AI transparency solutions. By integrating the approaches, a diversity of possible transparency instruments is mapped out for policymakers and their application is demonstrated using the case of smart energy meters.