Beyond the "Black Box": Enabling Meaningful Transparency of Algorithmic Decision-Making Systems through Public Registers

Abstract

Deployments of algorithmic decision-making systems (ADMs) by the public sector have been plagued with opacity. There is a baseline lack of visibility of the context and purpose of the ADM system as well as its potential risks to individuals and collective goods. In many cases, citizens are unaware of the very existence of algorithmic systems that they interact with or that help decide their access to benefits or influence policing. Moreover, disclosures concerning algorithmic systems often take place when their shortcomings (potential harms) are inadvertently exposed, often through the work of public interest groups.

Given the increasing adoption of algorithmic systems to automate decisions and services in the public sector, there is a need to operationalize transparency requirements to enable better accountability. While algorithmic transparency can take on many forms, this thesis mainly focuses on the role of public ADM registers in enabling meaningful transparency to the public. In the past year, at least five local governments have launched their very first ADM registers. Drawing upon these early experiences, relevant stakeholder interviews and specifically considering Amsterdam as a case study, we attempt to formalize the concept of a register as both a standardized and interpretable ADM disclosure mechanism, as well as a governance framework that enables coordination between a number of stakeholders to provide of transparency to the public. We also propose models through which public interest groups and civilians can be engaged in the creation, development and launch of public ADM systems through the governance of a register, and outline key benefits and limitations of such models.