The swift growth of digital education requires scalable, high-quality assessment instruments. Conventional exam question creation is arduous and challenging to customize, but current Automated Question Generation systems frequently exhibit deficiencies in pedagogical congruence, openness, and ethical protections. This paper introduces the PXF framework, an innovative AI-driven system for generating exam questions that incorporates Pedagogy, Explainability, and Fairness as core design concepts. The system utilizes a modular architecture that includes a Pedagogy Alignment Module for mapping Bloom's Taxonomy, an Explainability Engine that offers human-interpretable rationales, and a Fairness Module for proactive bias detection, all overseen by a Human-in-the-Loop review interface. Experimental validation on educational datasets indicates that the PXF framework attains a classification accuracy of 91%, an F1-Score of 0.87, and decreases question drafting time by 84% relative to manual authorship, while closely aligning with expert-level pedagogical quality. The results confirm its effectiveness in generating cognitively aligned questions, providing clear insights into AI decision-making, and detecting harmful biases for instructor assessment. This study advances the field of educational AI by presenting a systematic, transparent, and ethically aware framework that enhances assessment scalability while preserving pedagogical integrity and justice, offering a practical model for the future of AI-enhanced education.
Ghanim et al. (Sat,) studied this question.