Key points are not available for this paper at this time.
While the cutting-edge AI algorithms like ChatGPT are being adopted to government bureaucracy, there is a lack of discussion about which AI systems should be adopted and under what administrative conditions to ensure accountability. This study addresses the issue of AI system adoption in public sector from the perspective of accountability, which derived from institutionalism and democratic governance. Specifically, we borrow two terms from Romzek & Ingraham (2000) 36, "source of control" and "discretion of agency", to present a framework that categorizes "who can hold governments accountable" and "what kind of AI systems should be utilized?" for four types of accountability. we also point out the issue of the states of knowledge for each type of accountability and examines how AI systems can be adopted to address it. AI systems are discussed in terms of Symbolic and Connectionist AI systems, based on two approaches on AI. By doing so, we present a framework to help ensure that AI systems being deployed in the public sector are accountable.
Kim et al. (Sat,) studied this question.