The Dilemma of Accountability Fragmentation in Algorithmic Governance: Constructing an Ethical Accountability Framework Based on Multi-Agent Collaboration

Abstract

Accountability fragmentation is investigated as ethical responsibility is dispersed to such an extent that responsibility vacuums are formed in algorithmic governance systems. Based on interviews with 22 algorithm engineers from different Chinese technology companies, this study explores how technical practitioners perceive, distribute, and fulfill ethical responsibility in complex algorithmic ecosystems. Empirical results show that although engineers show awareness of ethical risks, they diffuse accountability into four areas: autonomous technological artifacts, organizations, users, and regulators. We discover three interconnected factors that sustain accountability fragmentation: responsibility gaps caused by autonomous and opaque algorithmic systems that endow engineers with low accountability, power differences within organizations that reduce engineers' autonomy, and diverse external expectations that further undermine ethical standardization. Based on these findings, we propose a comprehensive multi-agent ethical accountability framework including three components: clarified role responsibilities that specify concrete ethical obligations for different actor categories, collaborative governance mechanisms that connect organizational silos via cross-organizational ethics committees, and embedded value-sensitive design methods that convert ethical requirements into technical specifications. The framework simultaneously contributes to theory and practice of algorithmic governance. For practice, it provides an explicit guide for organizations to maintain effective ethical oversight in increasingly technological environments. For theory, this study extends the literature on technology ethics by bridging theoretical concepts of distributed responsibility with empirical records from technical practitioners. In addition, our work provides actionable suggestions for constructing accountability systems in algorithmic governance.