Algorithms are starting to play an increasingly prominent role in government organizations. The argument is that algorithms can make more objective and efficient decisions than humans. At the same time, recent scandals have highlighted that there are still many problems connected to algorithms in the public sector. There is an increasing emphasis on ethical requirements for algorithms and we aim to connect these requirements to insights from public administration on the use of technologies in the public sector. We stress the need for responsible algorithmization – responsible organizational practices around the use of algorithms – and argue that this is needed to maintain the trust of citizens. We present two key components of responsible algorithmization – value-sensitivity and transparency – and show how these components connect to algorithmization and can contribute to citizen trust. We end the article with an agenda for research into the relation between responsible algorithmization and trust.

The police use all sorts of information to fulfil their tasks. Whereas collection and interpretation of information traditionally could only be done by humans, the emergence of ‘Big Data’ creates new opportunities and dilemmas. On the one hand, large amounts of data can be used to train algorithms. This allows them to ‘predict’ offenses such as bicycle theft, burglary, or even serious crimes such as murder and terrorist attacks. On the other hand, highly relevant questions on purpose, effectiveness, and legitimacy of the application of machine learning/‘artificial intelligence’ drown all too often in the ocean of Big Data. This is particularly problematic if such systems are used in the public sector in democracies, where the rule of law applies, and where accountability, as well as the possibility for judicial review, are guaranteed. In this article, we explore the role transparency could play in reconciling these opportunities and dilemmas. While some propose making the systems and data they use themselves transparent, we submit that an open and broad discussion on purpose and objectives should be held during the design process. This might be a more effective way of embedding ethical and legal principles in the technology, and of ensuring legitimacy during application.

Predictive tools as instruments for understanding and responding to risky behaviour as early as possible are increasingly becoming a normal feature in local and state agencies. A risk that arises from the implementation of these predictive tools is the problem of dirty data. The input of incorrect or illegally obtained information (‘dirty data’) can influence the quality of the predictions used by local and state agencies, such as the police. The article focuses on the risks of dirty data in predictive policing by the Dutch Police. It describes the possibilities to prevent dirty data from being used in predictive policing tools, such as the Criminality Anticipation System (CAS). It concludes by emphasizing the importance of transparency for any serious solution looking to eliminate the use of dirty data in predictive policing.