AI is rapidly integrating into education, but its effectiveness depends on how well it aligns with real pedagogical needs. Too often, AI systems are optimized for efficiency rather than meaningful learning. To ensure these tools truly support educators and students, we need rigorous evaluation grounded in learning theory and human-centered design.
This talk explores how educational data science can drive this effort, highlighting three case studies: (1) the National Tutoring Observatory (NTO), which leverages tutoring data to improve AI-driven teaching tools; (2) an AI-powered deliberate practice system for medical students that simulates patient interactions and provides real-time feedback; and (3) AI-driven course assistants that facilitate interactive learning in higher education. Rather than allowing AI to dictate pedagogy, educators must be active in shaping, evaluating, and curating these technologies. By integrating research, transparency, and participatory design, we can ensure AI enhances—not replaces—the craft of teaching.
