How Do Algorithmic Decision‐Making Systems Used in Public Benefits Determinations Fail? Insights From Legal Challenges
分析了71起针对残疾、失业和营养援助项目中算法决策的法律挑战，发现算法错误源于法定要求转化为计算逻辑时的数据、设计或系统限制问题，构成跨法律、组织和技术领域的系统性治理问题。
ABSTRACT When algorithmic decision‐making systems fail to function as intended, they become conduits for administrative error and risk producing arbitrary determinations through the very technologies meant to prevent them. Analysis of 71 federal and state court dockets contesting algorithm‐based determinations in disability, unemployment, and nutrition assistance programs shows how this risk manifests in practice. Findings show that deviations from legally prescribed outcomes occur when the translation of statutory requirements into computational logic is compromised by flawed data, problematic design choices, or inherent system limitations. These algorithmic administrative errors are neither isolated glitches nor purely technical problems; they constitute a systemic governance problem that cuts across legal, organizational, and technical domains. Addressing them requires coordinated oversight across all three areas, rather than reliance on post hoc troubleshooting.