Operationalizing responsible AI principles through responsible AI capabilities

Abstract

AbstractResponsible artificial intelligence (RAI) has emerged in response to growing concerns about the impact of AI. While high-level principles have been provided, operationalizing these principles poses challenges. This study, grounded in recent RAI literature in organizational contexts and dynamic capability theory, and informed by literature on RAI principles and expert interviews in organizations deploying AI systems, (1) problematizes the high-level principles and low-level requirements and underscores the need for mid-level norms by adopting dynamic capability as a theoretical lens, and (2) develops five themes to capture firms’ RAI capability, including (i) understandable AI model, (ii) bias remediation, (iii) responsiveness, (iv) harmless, and vi) common good. As our contribution to the field of information systems (IS), this study extends the emerging literature on operationalizing RAI and dynamic capabilities, empirically elucidating the capabilities needed by firms. For IS practice, we provide organizations deploying AI with novel insights to aid in the responsible implementation of AI.