When responsibility fails to arise: institutional conditions of action in algorithmic governance
Contemporary frameworks of algorithmic accountability typically assume that responsibility persists under conditions of sociotechnical complexity, even when it becomes opaque or distributed. This article builds on a recent argument that under optimization‑driven governance responsibility does not merely become hidden but fails to arise [14], and develops the institutional structure that argument requires. Drawing on an externalist reading of Arendt, I argue that minimal publicly attributable responsibility depends on three jointly necessary conditions: discreteness, authorship, and publicness. I then identify four mechanisms by which optimization‑driven systems erode them: iterative output generation, metric substitution, procedural diffusion, and temporal continuity. The four are derived from the three rather than enumerated as a parallel list. The framework is demonstrated through the Australian Robodebt scheme, a deterministic system in which automated welfare debt notices were issued without identifiable decisions, attributable authors, or contestable forums. A brief extension shows that the framework holds, and strengthens, when applied to probabilistic and generative architectures. The article closes with three concrete design principles, event‑marking, authorship preservation, and deliberative apertures, aimed at preserving the institutional form decisions must take if anyone is to answer for them.
Published in AI and Ethics, 2026
Read the article here