A Public Values Perspective on the Application of Artificial Intelligence in Government Practices: A Synthesis of Case Studies

Introduction

The first wave of technological innovation in governments focussed on digitising back-office operations with the goals of efficiency and cost savings inspired by the New Public Management (NPM) reforms of the 1980s. NPM was driven by the neo-liberal agenda and critique of large bureaucratic structures associated with red tape and cumbersome processes (Bernier et al., 2015; Kamarck, 2004). However, technology took a backseat and was considered simply a tool for achieving managerialism. Succeeding this initial technology implementation which has had mixed results in meeting its innovation goals (Hung et al., 2006), the second wave driven by Artificial Intelligence (AI), however, is transforming the roles and functions of government. Often referred to as the next frontier of digital-era governance (DEG) (Dunleavy et al., 2006), this technologically centred model of governance enabled by AI has the potential for a lean government providing personalised services that are efficient and cost-effective. The use of AI also introduces new risks and ethical challenges such as biased data, fairness, transparency, the surveillance state, and citizen behavioural control (Ashok et al., 2022; Saura et al., 2021A; Ashok, 2018). Maintaining citizen trust and legitimacy of AI-driven governmental services and processes is vital more than ever for sustaining democratic processes (Janssen & van den Hoven, 2015).

The concept of AI, introduced by John McCarthy in 1956, is aimed at developing intelligent machines that can emulate human cognition autonomously (von Krogh, 2018; Washington, 2006). Following an enthusiastic start, progress stalled due to technical limitations; AI was limited to expert systems with specific applications (Haenlein & Kaplan, 2019). At the beginning of the 21^st century, with advances in processing speeds and storage, and decreasing computational costs, interest in AI grew exponentially (Haenlein & Kaplan, 2019; von Krogh, 2018). Brynjolfsson and McAfee (2014, p. 7) claim this renewed interest as the “second machine age” where machines are taking over cognitive human tasks.

Dwivedi et al. (2021) discuss the terminological challenges associated with defining AI. The meaning of artificial vs natural is derived from the epistemological assumptions of objectivist or constructivist ideas and scientists and philosophers still do not have a good grasp of what intelligence entails (Ibid.). Following Dwivedi et al. (2021, p. 24) “institutional hybrid” approach, AI for this chapter is defined as emerging technologies that enable machines to “learn, adapt, be creative and solve problems” autonomously (Rosa et al., 2016, p. 6). Scholars (Raisch & Krakowski, 2020; Sousa et al., 2019; von Krogh, 2018) generally agree on the three components of AI: input, often big data; task processing algorithms; and output, either digital or physical.

The primary applications of AI in government are process automation, virtual agents, predictive analytics, resource management, and threat intelligence and security (Ojo et al., 2019; Wirtz et al., 2018). The associated benefits include efficiencies, accelerated processing of cases, workforce redistribution to productive tasks, and enhancing satisfaction and trust in public authorities (Susar & Aquaro, 2019; Wirtz & Müller, 2018). AI represents radical innovation transforming internal organisational structures and introducing new governance models (Ashok et al., 2016). However, the use of AI for making policy decisions is accompanied by ethical dilemmas of fairness, transparency of black-box algorithms, privacy concerns, and respect for human rights (Ashok et al., 2022; Ribeiro-Navarrete et al., 2021; Wirtz et al., 2018). Kuziemski and Misuraca (2020) and Helbing et al. (2019) discuss externalities from the use of AI leading to the detriment of human dignity and well-being such as mass surveillance, profiling, and nudging for incentivising compliance with government direction akin to programming citizens. Mehr et al. (2017) caution AI should not be used solely for its innovation potential but adapted towards a broader social development goal. Citizens expect responsive governments able to meet their personalised needs with the adoption of AI-driven governmental services. The level of trust and legitimacy of government determines expectations of privacy and a fair, equitable, and secure outcome. Erosion of this trust with mismanagement of ethical issues undermines democratic institutions and impacts adoption.