From Retrieval to Synthesis: Repair Literacy and the Domestication of Generative AI

AI literacy develops through hands-on practice, not formal training alone. The study tracked over 10,000 real ChatGPT conversations from 36 students over a full academic year, finding that students develop competence through ongoing interaction and experimentation rather than one-time instruction, suggesting educators should design learning environments that encourage sustained engagement with AI tools.

Students learn most when AI fails. The research identifies "repair literacy"—learning that occurs when students encounter AI breakdowns and errors—as a crucial but overlooked dimension of AI competence. This suggests that experiencing and working through AI limitations may be more educationally valuable than only successful interactions, with implications for how we design responsible AI learning experiences.

Students develop sophisticated judgment about when and how to use AI. Rather than using ChatGPT uniformly, students strategically matched different interaction patterns to specific needs (academic work, emotional support, metacognitive reflection, problem-solving, and trust-building), demonstrating that responsible AI use involves critical decision-making about appropriate applications rather than blanket adoption or rejection.

Responsible AI integration requires supporting student agency. The study emphasizes that students exercise "critical judgment about AI limitations" when given space to experiment, suggesting that effective AI literacy education should empower students to evaluate AI capabilities themselves rather than relying solely on external rules or restrictions.

AI literacy is relational and ongoing, not a fixed skill. Competence emerges through "ongoing relational negotiation" between students and the tool, implying that responsible AI use requires continuous learning and adaptation rather than a one-time certification or training program.