Designing for trust: A conjoint experiment on citizens’ support of regulatory regimes for ensuring trustworthy Artificial Intelligence

The use of Artificial Intelligence (AI) systems is becoming increasingly prevalent in many areas based on a vast range of applications. AI systems promise to increase effectiveness and efficiency of private and public services through automation and to improve decision-making processes. However, it has become evident that AI systems can be biased and lead to discriminatory outcomes. In other words, AI systems cannot be blindly trusted to perform well and create output that we deem morally acceptable. At the same time, the trustworthiness of AI systems counts as a crucial precondition for the uptake of these systems. If the public does not deem them trustworthy, they will be reluctant to accept and use them. They may even lose trust in actors that employ them, be they businesses or public organizations. Yet, the vast majority of citizens do not have the capacity – neither the expertise nor the access to information – to evaluate the trustworthiness of AI systems themselves. Instead, they will depend on intermediaries (i.e. regulators) to effectively monitor, evaluate and enforce the trustworthiness of AI systems. It may thus be citizens’ perceptions of such intermediaries rather than of the AI systems themselves that matters for whether citizens trust expect these AI systems to be trustworthy. In this paper, we are examining which type of regulatory regime citizens expect to most effectively regulate AI systems and their uses.