Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (AI) can improve students’ work, but an AI-assisted answer does not show whether they can reason alone. This quasi-experiment examined whether tax reasoning demonstrated with AI assistance was maintained when 41 Business Administration students reconstructed it independently. In a digital economy taxation case, the experimental group (n = 20) used the TaxDigital Tutor, an AI tutor developed for this study that guides reasoning without supplying answers, whereas the control group (n = 21) could use generative AI and other resources. All students immediately reconstructed their reasoning by hand without assistance, and both phases were scored with the same five-criterion rubric. Overall, scores were 14.3 points lower in the unassisted reconstruction (F(1, 39) = 18.39, p < 0.001). Scores declined by 17.9 points in the control group (p < 0.001) and by 10.6 points in the experimental group (p = 0.060); however, the exploratory between-group comparison was not significant (p = 0.282). Using case data as evidence was the lowest-scoring criterion in both phases. These findings suggest that scores obtained with AI do not necessarily reflect performance without it. The study also proposes the From Assisted to Autonomous (A2A) framework for progressively reducing AI assistance and assessing independent performance.
No takes yet. Share an insight, caveat, or question.
Silva et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: