Paper Type
Complete
Paper Number
PACIS2026-2017
Description
Artificial Intelligence has become a core driver of industrial transformation but introduces significant risks as it integrates with critical services. While ISO/IEC 42001 and ISO/IEC 42005 establish procedural requirements for AI impact assessments, they lack technical quantitative methodologies. This study designs, implements, and demonstrates an integrated technical assessment layer intended to support broader AI impact assessment processes across seven dimensions: explainability, privacy, fairness, reliability, robustness, data quality, and environmental impact. An initial empirical demonstration using a natural language processing classification task shows how high-level governance principles can be operationalized into measurable, reproducible metrics. By bridging the gap between macroscopic digital governance and engineering practice, this research contributes to IS discourse on AI accountability and the organizational embedding of risk assessment, offering a replicable foundation for institutions seeking to operationalize trustworthy AI mandates within their development workflows.
Recommended Citation
WEI, YU-CHIH and CHEN, YING-XUAN, "Automated Quantitative Impact Assessment Framework for Artificial Intelligence Systems: Bridging Governance and Engineering Practices" (2026). PACIS 2026 Proceedings. 16.
https://aisel.aisnet.org/pacis2026/ai_ethic/ai_ethic/16
Automated Quantitative Impact Assessment Framework for Artificial Intelligence Systems: Bridging Governance and Engineering Practices
Artificial Intelligence has become a core driver of industrial transformation but introduces significant risks as it integrates with critical services. While ISO/IEC 42001 and ISO/IEC 42005 establish procedural requirements for AI impact assessments, they lack technical quantitative methodologies. This study designs, implements, and demonstrates an integrated technical assessment layer intended to support broader AI impact assessment processes across seven dimensions: explainability, privacy, fairness, reliability, robustness, data quality, and environmental impact. An initial empirical demonstration using a natural language processing classification task shows how high-level governance principles can be operationalized into measurable, reproducible metrics. By bridging the gap between macroscopic digital governance and engineering practice, this research contributes to IS discourse on AI accountability and the organizational embedding of risk assessment, offering a replicable foundation for institutions seeking to operationalize trustworthy AI mandates within their development workflows.
Comments
03-EthicsSocietalImpact