Artificial intelligence (AI) holds both potential benefits and significant risks for organizations, including biases, discrimination, opacity, and reduced human accountability. Technical systems, including AI, must be regulated to safeguard stakeholders' interests and maintain proper functioning over time. However, the problem of designing practical controls for specific AI systems and organizations largely remains unresolved. To address this gap, we propose an initial methodology focusing on identifying and contextualizing stakeholders' values within their local environments. We validate our approach through a case study in the Japanese life insurance industry, aiming to assess its repeatability and potential improvements. Our design method includes 10 steps which AI system developers can use to situate high-level institutions in the local context to control their AI systems. The validation eforts highlight the contextual nature of designing controls for AI systems, emphasizing the need for diverse control mechanisms to comply with stakeholders' values.
governance level, there is currently a gap in literature that aids in conceptualizing and empirically
validating the design of control mechanisms for AI systems in their situational context [
Our paper addresses this gap by providing a repeatable process for situating identified values
into practical organizational controls in the context of specific AI applications. We build our core
contribution on earlier approaches that provide ways to tackle parts of the problem. Building
on Van De Poel [