How Stakeholders Operationalize Responsible Artificial Intelligence (AI) in Data Sensitive Contexts

Operationalizing the responsible use of AI in data-sensitive, multi-stakeholder contexts is challenging. We studied how six AI tools were operationalized in a humanitarian crisis context, which involved aid agency decision makers, private technology firms and vulnerable populations. From the insights gained, we identify five types of “AI responsibility rifts” (AIRRs - the differences in subjective expectations, value sand perceived impacts of stakeholders when operationalizing an AI tool in data-sensitive contexts). We propose the self-assessment SHARE framework to mitigate these rifts and provide recommendations for closing the identified gaps.