Artificial intelligence as the More Knowledgeable Other (MKO) 2.0: metadata-driven adaptive scaffolding for narrative comprehension

Introduction Recent advances in large language models (LLMs) have expanded the capabilities of artificial intelligence (AI) in education, yet much of this progress remains focused on answer generation rather than theoretically grounded instructional support. This study introduces AI as the More Knowledgeable Other (MKO) 2.0 , an MKO-inspired framework for generating task-sensitive scaffolds from structured question metadata. Methods Using the Quest Dataset of 6,989 narrative-comprehension questions, a rule-based system operationalized three task dimensions, comprehension attribute, explicit/implicit evidence status, and local/summary scope, into seven primary scaffold families and 28 surface realizations. An attribute-stratified sample of 140 questions was subsequently used to compare rule-based scaffolds with generic narrative-grounded and MKO-conditioned LLM scaffolds. Findings Across paired LLM outputs, MKO-conditioned scaffolds were more elaborated than generic scaffolds, averaging 5.18 additional words per instance ( W = 356.5, p < 0.001), and showed slightly higher corpus-level lexical variation (MATTR-50 = 0.785 vs. 0.765). Narrative-evidence cues occurred more frequently under MKO conditioning (40.0% vs. 10.0%, p < 0.001), while, among implicit questions, inference-oriented cues occurred in 36.1% of MKO-conditioned scaffolds compared with 2.8% of generic scaffolds ( p =0.002). In contrast, question-specific reference did not differ significantly, and MKO conditioning did not reduce lexical answer overlap. Discussion These findings demonstrate the feasibility of using structured task metadata to organize AI-generated scaffolding and show that metadata conditioning is associated with measurable differences in scaffold realization. The study contributes a transparent, task-sensitive framework for shifting AI-supported learning from answer provision toward structured guidance, while recognizing that learner-level pedagogical effectiveness requires direct human evaluation.