AI failure, AI success, and AI power dynamics in the public sector

This chapter offers a novel typology of AI failure and AI success in the global context of public administration. First, departing from existing literature that has focused on AI bias and discrimination, this chapter offers a more nuanced perspective on AI-related mistakes. We distinguish three layers of AI failure: (i) AI failure as negligence, which encompasses system imperfections, accidents and other negligent actions; (ii) AI failure as recklessness, or flaws arising from disregarding known risks in AI deployment; (iii) AI failure as intent, which includes deliberate misuse, underpinned by institutional racism or exploitation of power. Second, the chapter presents a more nuanced vision than legal scholarship on law, regulation, governance of AI, underlining the positive impact of AI on government transactions. In doing so, we identify three layers of AI success, illustrating how AI can be harnessed to improve government-citizens interactions and help individuals in vulnerable positions. Third, this chapter draws on examples of AI systems deployed by governments across the globe, thus moving away from a Western-centered perspective, showing the variety in AI failure and the potential for AI success. Ultimately, we argue AI's failure or success is determined by how those in power use digital technology; how they are supported to use power for good, and how they are held accountable for misuse.