Reconstructing Granollers with Evidence Graphs and Artificial Intelligence: An Epistemic Approach to Historical Thinking in Social Studies Education

This open English version is a translated and academically adapted author version based on the article originally published in Spanish in Revista Íber, no. 122 (Editorial Graó, 2025): De la Torre Díaz, J. L., & De la Torre Lorente, J. L. (2025). Reconstruir Granollers con grafos de evidencias e inteligencia artificial. Íber: Didáctica de las ciencias sociales, geografía e historia, 122. ISBN 979-13-87863-65-4. The body of the article preserves the substantive argument, structure, and pedagogical design of the original. The title, abstract, and metadata have been adapted for international open dissemination. Abstract:This article presents a classroom-based approach to history education in which artificial intelligence functions as an epistemic tool rather than as a substitute for student thinking. The proposal is built around the bombing of Granollers on 31 May 1938 — a significant episode in the aerial bombardment of civilian populations during the Spanish Civil War — and is designed for lower secondary social studies education. Instead of asking AI systems to explain the event in open-ended ways, the sequence begins with a bounded documentary corpus that includes survivor testimonies, contradictory newspaper coverage from 1938, diplomatic reports, and elements of urban memory. The pedagogical design organises AI use at three differentiated levels. First, the teacher uses AI during prior inquiry to organise documents, detect tensions between sources, and formulate key questions. Second, students use general-purpose language models for auxiliary tasks such as linguistic simplification, chronology support, or comparison of formulations, always under mandatory audit. Third, the core of the sequence is the construction of an evidence graph that enables students to connect actors, events, places, narratives, and silences through explicit and discussable relationships. The article describes a five-session learning sequence, complementary assessment rubrics, and a model of critical AI literacy grounded in source criticism, corroboration, contextualisation, and historical agency. Its central argument is that generative AI can strengthen historical thinking only when three conditions are met: corpus before model, systematic audit after each use, and a visual structure that obliges students to justify interpretations through evidence rather than through automated synthesis. In this sense, the proposal offers a transferable framework for controlled AI use in history and social studies education.