Support By Proxy? On the Acceptance of Algorithmic Systems in a Public Context

• In Event: The Politics of Moderation

Abstract

Algorithmic systems increasingly govern into people’s lives. Consumers rely on private sector services that are based on algorithmic systems, e.g. in the form of content filters in online platforms and entertainment recommender systems. However, the capabilities of these systems are also seen as highly valuable in public sector service delivery where they promise to yield major effectiveness and efficiency gains. Many countries over the globe, especially advanced adopters such as the US and New Zealand have taken up algorithmic systems in government to inform decision-making. They are used, for instance, to improve policing activities, to match qualification measures to the unemployed, or to identify the most efficient school bus routes. We aim to shed light on the sources of support of these applications with original survey data, thereby adding to a growing literature that looks at how people perceive algorithmic systems. We do so, however, with a perspective that is more strongly rooted in political science analyses than existing research.
Previous work has mainly looked at people’s perceptions of the technology as such. While some have examined how people generally view widespread applications (such as algorithms in online platforms) in public opinion surveys, other studies are more focused and strongly informed by a literature that cautions about a possible lack of fairness, accountability, and transparency (FAccT, for short) with respect to algorithmic systems. Drawing on this work, previous research on the perceptions of algorithmic systems has investigated the role that perceptions of fairness, transparency and accountability play for the evaluation of such systems, often by focusing on these individual aspects of transparency, accountability, and particularly fairness.
The present paper adds to this emerging literature by taking a broader perspective that stresses the socio-technical nature of algorithmic systems, i.e. the fact that these are embedded in a certain social context. This context involves certain actors and actor relations as well as a concrete purpose of an application which entails context-specific value promises and risks. Against this backdrop, support for the adoption of algorithmic systems could in large part hinge on citizens’ evaluation of aspects that concern the broader context in which those systems are deployed. The thrust of our study is thus that people may show strong support for – or skepticism toward – algorithmic systems by proxy. This would have important implications because it would mean that people may reject well-designed systems for reasons that have nothing to do with the application. Equally, they might accept badly designed systems simply because they trust the organization operating the system or because they desperately want a certain problem to be solved which an application has the potential to deal with.
To probe the role of people’s dispositions toward the larger setting in which an algorithmic system is implemented, we look at two real-world applications that provide a service to citizens: applications used for (1) predictive policing and for (2) skin cancer prediction. In both settings, the values at stake are clearly defined and of a fundamental sort: security and health. Also, both applications are set in public context, for which the acceptance of algorithmic systems is especially important. We test determinants of support for algorithmic systems in the two domains with a survey specifically designed for that purpose. First, the set of variables is theoretically rooted in the socio-technical perspective on algorithmic systems. Second, the survey expressly deals with the issue that respondents have to evaluate unfamiliar and cognitively demanding issues. Third, it also includes an experimental stimulus in both settings by suggesting that a very high versus a very low performance is generally achieved with an application. In this way, we can probe the importance of an important application-specific quality – i.e. effectiveness – in relation to the more general dispositions toward the context of implementation that we are primarily interested in. The sample used for the analysis comprises more than 2400 respondents from a representative online panel that reflects the composition of the German population – with about 1200 cases for each setting.

Authors

• Pascal König, TU Kaiserslautern

• Julia Felfeli