Ethical boundaries for the use of artificial intelligence in academic assessment

Abstract

Purpose: This study examines educators’ perceptions of the ethical boundaries for using Artificial Intelligence (AI) in teaching and academic assessment, distinguishing AI support from AI delegation using flipped learning as a pedagogical lens. Design/methodology/approach: A questionnaire was administered to 148 faculty members from Universidade Autónoma de Lisboa, Portugal. The paper adopts an exploratory quantitative case-study design. Analysis used descriptive statistics, normalised Shannon entropy, Pearson correlation, and a paired-samples t test. Findings: Respondents were open to AI support for routine, preparatory, and lower-order learning activities, but more cautious when AI was associated with evaluation, fairness, interpretation, and assessment decisions. Teacher judgment was strongly prioritised, while trust in AI-generated fair and unbiased evaluation remained limited. Entropy analysis showed relatively clear boundaries around AI bias and human judgment, but more contested boundaries around AI-supported evaluation, learner independence, and critical reflection. Research limitations/implications: The single-institution convenience sample should be interpreted as exploratory and analytically transferable rather than statistically generalisable. Practical implications: Institutions should clarify where AI may support assessment and where human evaluative authority must remain explicit. Social implications: The findings highlight the need for shared institutional norms. Originality/value: The paper proposes the Dual Approach as an interpretive framework connecting ethical responsibility, cognitive complexity, and flipped learning to distinguish AI support from delegated evaluative authority. Its contribution lies in locating the ethical boundary at the transfer of assessment responsibility and identifying where educators’ perceptions indicate crisp or fuzzy boundaries.