Operationalizing AI contestability through technical feasibility analysis

As AI systems increasingly make high-stakes decisions affecting human lives, the question of computational feasibility in implementing contestability mechanisms has emerged as a critical technical challenge. While the field of AI contestability remains in its infancy with limited theoretical frameworks, the practical question persists: which components of proposed contestability mechanisms can actually be implemented with current technical methods? This paper investigates computational feasibility through systematic cross-domain empirical analysis using a structured contestability framework as our testable hypothesis. We examine operational AI systems across security, legal, financial, and medical domains to assess the technical implementability of contestability requirements across diverse architectures and decision contexts. Our analysis reveals a three-tier taxonomy of implementation feasibility: fully computable components, partially computable elements, and non-computable aspects. Through systematic technical evaluation, we demonstrate that significant portions of the contestability framework requirements face substantial implementation challenges with current methods. The findings provide an empirical foundation for understanding technical boundaries of contestability implementation and identify specific areas where current AI system capabilities fall short of theoretical requirements, offering concrete guidance for both framework designers and system implementers.