Two Means to an End Goal: Connecting Explainability and Contestability in the Regulation of Public Sector AI

Explainability and contestability mean different things across technical, legal, and organizational contexts, creating confusion about how to actually implement these principles in practice. The paper's interview study with 14 experts reveals that these terms lack consistent definitions, which hampers effective regulation of public sector AI systems.

There's a critical gap between top-down regulation (laws and policies) and bottom-up implementation (how organizations actually build and deploy AI), making it difficult to translate regulatory requirements into real-world practice. This misalignment is identified as a major friction point in realizing both explainability and contestability.

Contestability requires multiple pathways beyond courts, including non-judicial channels where citizens can challenge AI decisions. The paper distinguishes between individual contestation (one person challenging a decision) and collective action (groups challenging systemic issues), suggesting that effective contestability needs diverse mechanisms.

Responsibility for ensuring explainability and contestability remains unclear across organizations, creating accountability gaps. The paper identifies the assignment of responsibility as a key friction point, suggesting that without clear ownership, these principles may not be adequately implemented.

Solving these challenges requires genuine collaboration across technical experts, legal scholars, and organizational practitioners, not just technical fixes. The paper recommends a "Regulation by Design" approach that integrates these perspectives from the start, rather than treating regulation as an afterthought to AI development.