This study examines the limitations of efficiency-focused AI education and reinterprets the value of Error-Based Learning (EBL) in multimodal environments. While widespread AI tutoring systems optimize for correctness, this "correctness obsession" constrains learners' metacognitive development and creative thinking. Analysis of four large-scale datasets—OULAD (n=32,593), MIT Open Learning, Stanford AI Index, and UCLA UCUES—using OLS regression (R²=0.542, p<0.001) revealed AI integration increased Virtual Learning Environment interactions by 42.3% and raised course completion by 12.7 percentage points. However, applying a 15% optimal error rate simulation grounded in productive failure theory improved learning outcomes an additional 8.4 points beyond AI-only gains. Notably, creativity indices (strategic diversity, unstructured attempts, retry rates) increased by 58%, while NASA-TLX mental demand paradoxically decreased by 28.4%. Post-error retry rates rose to 67%, suggesting appropriate error experiences simultaneously enhance cognitive efficiency and higher- order thinking. This study proposes the MERGE framework (Metacognition, Engagement, Regulation, Growth, Ethics), integrating pedagogy, HCI, and data science, demonstrating practical applicability with a 121.7% educational ROI. These findings present a paradigm that preserves human creativity and diversity beyond AI-centric uniformity.
Jo et al. (Sat,) studied this question.