Dr Bjorn Kleizen, University of Antwerp, 'Can we foster citizen trust in governmental AI projects Experimental evidence on the limits of project-level measures?'

AI-related scandals have stimulated the development of various techniques and guidelines to implement trustworthy AI. However, scant evidence is available regarding the conditions under which citizens trust and support governmental AI projects. Although extant research has focused on user trust, citizens rarely use public sector algorithms directly. This article takes a novel approach, presenting evidence of two survey experiments in which citizens were given information on the trustworthiness of governmental AI projects. Interventions consisted of information on 1) legal compliance, 2) ethical AI measures, 3) limitations placed on data-gathering, 4) a human-in-the-loop, 5) ensuring non-discrimination and 6) enhancing technical robustness. The results provide scant evidence that information on trustworthiness enhances trust or policy support. Instead, perceived trustworthiness is predicted by prior attitudes and experiences, particularly privacy concerns, trust in government and trust in AI. These results suggest that the short-term efficacy of communicating trustworthiness is minimal.