Responsible Management for Dynamic Black Box AI: a Cybernetic Approach

Black box AI systems, characterized by their opaque internal decision-making processes, remain relatively unexplored within the IS field, often leading to unintended consequences from AI adoption in practice. With the recent hype in AI and technological advancements in Machine Learning (ML) and Deep Learning (DL), this has catalyzed research interest in Responsible AI (RAI) by emphasizing managerial oversight and control to ensure accountable, transparent, and ethical outcomes. Traditional approaches like eXplainable AI (XAI) methods and constraint methods may prove ineffective in managing ML-based AI systems, particularly for dynamic learning AI models. This study employs empirical inquiry from three social media companies to investigate effective control implementation. Our findings develop a Cybernetic control framework, integrating buffering control, feedforward control and feedback controls, to achieve responsible AI use for organizational decision-making.