The mainstreaming of Artificial Intelligence (AI) in technologies of various application fields offers many opportunities, but also comes with a number of risks for data subjects, users, developers, intermediaries and other stakeholders. This includes, among other things, risks posed by technical unreliability, epistemic opacity, privacy intrusions and algorithmic discrimination. As a consequence, scholars of various academic disciplines have proposed contestability as a concept and design principle that supports stakeholders in challenging and possibly correcting the adverse effects caused by certain AI systems. Most of the academic work in this context has been carried out by scholars of Human–Computer-Interaction, Design Research and Public Law. Building on this important work, we discuss a number of exemplary practices and aspects that should, in our opinion, be considered within the scientific study of AI contestation. Moreover, we propose a shift in attention from system-specific features and mechanisms toward a stronger consideration of motivations for contestations, of critical practices in organizations and of a contextualization within legal regulatory regimes. This includes anticipating risks and effective means for contestation, co-creating among stakeholders, formalizing the translation from principles to practices for AI contestability, governing contestation in deployed systems, facilitating a culture for contestation, rejecting AI as well as enabling external oversight and critical monitoring. These conceptual considerations will hopefully facilitate further research and empirical investigations on AI contestability in social science fields such as organizational studies, communication studies, science technology studies and law.
Hirsbrunner et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: