Ethical generative artificial intelligence and organizational performance: evidence from structural equation modeling and machine learning
Weng Marc Lim, Ahmed Yehia Ebeid, Yara IbrahimPurpose
This study examines how the FATAA ethical principles of fairness, accountability, transparency, accuracy and autonomy relate to the usage of generative artificial intelligence (GenAI) and, through usage, to organizational performance, with ethical leadership as a contextual moderator.
Design/methodology/approach
Drawing on institutional theory and behavioral reasoning theory, the study tests a model linking FATAA principles to GenAI usage and, in turn, to organizational performance. A hybrid analytical design combines partial least squares structural equation modeling (PLS-SEM) with three machine-learning methods and is applied to survey data from 301 AI-literate professionals.
Findings
Fairness and accuracy are most strongly associated with GenAI usage, whereas accountability, autonomy and transparency show weaker or non-significant effects. GenAI usage, in turn, is strongly associated with organizational performance, with bootstrapped indirect effects confirming that only fairness and accuracy carry through usage to performance. Ethical leadership is tested as a moderator at both the adoption stage and the performance stage and, notably, neither conditioning effect is supported.
Originality/value
The study advances the literature in three ways. First, it shows that ethical principles are not equally consequential for GenAI adoption, which calls into question the universalist framing of AI ethics. Second, the hybrid PLS-SEM and machine-learning design surfaces both explanatory effects and feature-level importance, an integration that single-method studies cannot deliver. Third, ethical leadership is examined as a moderator at both the adoption and performance stages, with neither conditioning effect supported, which documents a dual-stage boundary condition for leadership in AI-enabled environments and positions governance-by-design as a candidate explanation for future testing.