Representative Bureaucracy in the age of AI governance

How does the introduction of AI into public sector decision making impact citizens’ desire for representative bureaucracy? It is unrealistic to assume that case workers can completely and fully set aside their own biases in their work, even in professions where objectivity is sought. One way to counteract bias is to ensure that the background of case workers reflects the population they serve. Hence, scholars have argued the case for a representative bureaucracy, where the public servants constitute a cross-section of the populace it serves (Lægreid and Olsen 1978; T. Christensen, Lægreid, and Zuna 2001). We measure the desire for representative bureaucracy in a survey experiment among a representative sample of residents in Norway, and find that citizens care more about the public servants sharing their socio-demographic characteristics when they are told that recommendations based machine learning tools are being utilized in the decision process on who shall receive welfare benefits. We speculate that citizens fear using such tools entails a loss of crucial nuances in individual decisions, highlighting the importance of case workers who understand the citizens’ situations and can intervene and make corrections in specific cases.