Algorithmic profiling of the unemployed: A case study and a framework for understanding legitimization processes
研究丹麦一个用于预测失业风险的机器学习系统，分析其七年间因公众批评而终止的过程，识别出影响算法系统合法化的因素和主题。
Machine learning-based algorithms (MLAs) are often criticized for being biased or unfair when used by public institutions to make predictions or decisions about people's lives, particularly when marginalized or disadvantaged groups are involved. In this context, an important topic is the role of public discourses on the legitimacy of these tools, as these can significantly impact such projects. However, research on this topic is sparse. To add to this knowledge, the present study conducts a longitudinal case study of a public Danish project involving the deployment of an MLA system for profiling unemployed citizens with respect to their risk of long-term unemployment. The system was used for around seven years, but after significant public criticism, it was eventually terminated. Through a grounded theory-based document analysis of the public discourse on the MLA system, a set of legitimization factors and themes is identified. Using these elements, it is shown that public criticism and justification do not only appeal to different principles of legitimacy but may concern different aspects of phenomena. Furthermore, the study demonstrates that legitimization aspects should not be considered in isolation, as these can be used to develop arguments related to other aspects, some of which are inclined to counter each other. • The use of machine learning algorithms (MLAs) in public administration is discussed • A case study of an MLA project for profiling unemployed citizens is presented • The public discourse associated with the MLA project is analyzed • Themes in the associated public legitimization processes are identified • Factors affecting the legitimization of public algorithmic systems are identified