Understanding Transparency in Algorithmic Accountability

Abstract

Transparency has been in the crosshairs of recent writing about accountable algorithms. Its critics argue that releasing data can be harmful, and releasing source code won’t be useful. They claim individualized explanations of artificial intelligence (AI) decisions don’t empower people, and instead distract from more effective ways of governing. While criticizing transparency’s efficacy with one breath, with the next they defang it, claiming corporate secrecy exceptions will prevent useful information from getting out.

This chapter bucks the tide. Transparency is necessary, if not sufficient, for building and governing accountable algorithms. But for transparency to be effective, it has to be designed. It can’t be sprinkled on like seasoning; it has to be built into a regulatory system from the onset. And determining the who, what, when, and how of transparency requires first addressing the question of why.

Building on my work elsewhere, I thus begin by discussing the rationales behind regulating algorithmic decision-making, or decision-making by AI. I discuss the growing awareness in the literature that the object of regulation is not the technology of the algorithm in isolation, but includes the human systems around it. I then outline a taxonomy of transparency for accountable algorithms, building on the work of earlier authors and my own research on the European Union’s General Data Protection Regulation (GDPR).

Keywords: AI, Algorithmic Accountability, GDPR