Initial Development and Preliminary Validation of an AI Ethics Sensitivity Scale for Elementary School Students

Artificial intelligence (AI) increasingly confronts children with ethical situations, yet AI ethics education has emphasized awareness, and no instrument measures children’s capacity to identify and prioritise ethically relevant considerations within AI dilemmas. Grounded in Rest’s conception of moral sensitivity, this study reports the development and preliminary validation of an AI Ethics Sensitivity Scale (AIESS) for elementary students. The 24-item instrument, a structured self-report measure using a difference-score format across three dilemmas, was content-validated by experts and administered to 147 fifth-graders, 134 of whom were matched at a second administration. The one-factor model was the only specification that yielded an admissible solution, indicating that AI ethics sensitivity is most defensibly represented as a single dimension rather than four separable sub-elements, which were strongly intercorrelated (r = .51–.64). The total scale was reliable (McDonald’s ω = .81; CR = .83), although average variance extracted was modest (AVE = .31). Longitudinal confirmatory factor analysis showed suboptimal configural fit (CFI = .799); after three loadings and one intercept were freed, partial-metric and partial-scalar models met the prespecified change-in-fit criteria, although absolute fit remained suboptimal. Scores increased following construct-aligned instruction using scenarios distinct from the assessment dilemmas (d_z = 1.06), but this increase is reported descriptively because the single-group design precludes causal claims. These preliminary results position the AIESS as a content-valid, AI-situated measure of structured ethical prioritisation; replication in larger, multi-site samples is a priority.