Author ORCID Identifier
Maciel M. Queiroz: 0000-0002-6025-9191
Isis Domingues: 0009-0001-5197-0539
Abstract
As AI systems become embedded in organizational processes, they are widely celebrated for their precision and objectivity. Yet real-world deployments often reveal a different reality: algorithmic systems produce biased outputs, exhibit performance drift, and generate socially misaligned decisions, exposing their inherent instability. This paper introduces the concept of algorithmic fragility to capture this persistent instability as a structural condition, not a temporary technical defect, arising from the interaction between machine learning properties and the socially complex environments in which these systems operate. To explain how organizations cope, we advance the concept of stabilization work: the distributed sociomaterial labor through which actors absorb, reinterpret, and manage algorithmic breakdowns to sustain the appearance of reliability. We identify three micro-level practices — buffering, reframing, and patching — and show how these become institutionalized into meso-level routines and macro-level legitimacy. The framework conceptualizes algorithmic fragility as constitutive rather than exceptional, advances stabilization work as a form of invisible organizational labor, and reframes AI governance as a continuous, practice-based accomplishment instead of ex ante technical control.
Recommended Citation
Queiroz, M. M., & Domingues, I. (In press). Algorithmic Fragility: How Organizations Stabilize Unstable Machines. Communications of the Association for Information Systems, 59, pp-pp. Retrieved from https://aisel.aisnet.org/cais/vol59/iss1/9
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
