Realizing the Accountability of Algorithms in the Public Sector: A Reference Method for Managing Algorithm Registers

Abstract

The government of The Netherlands has mandated public organizations to disclose an algorithm register in the future. This aligns with the upcoming AI Act of the European Union, as a directive that aims to enhance transparency in algorithmic decision-making and use. This research highlights that the practices are rarely mature and not fully understood, which is characterized by their variability. The novelty of the practices in the industry and the limited research to date, currently make this a chaotic domain. In this research, we explore this domain through a method engineering approach. We conduct six case studies to elicit the process that each organization applies or envisions to manage their algorithm register. We model their processes using Process-Deliverable Diagrams, a meta-modeling technique that integrates dynamic and static perspectives. By applying a systematic method comparison approach, we compare and combine the processes into a single reference method for algorithm register management (RM4AR). This paper documents the process and outcomes of developing such a reference method, demonstrating how method engineering techniques aid in bringing structure into an emerging field.