abstract
- This study examines integrating generative AI-combining ChatGPT with Wolfram Language-to support simulation of electric and magnetic fields in first-year engineering at Tecnológico de Monterrey. Grounded in constructivism and project-/challenge-based learning, the intervention situates vector-field modeling in authentic contexts, using Coulomb's law for electric fields and Biot-Savart/Ampère's laws for magnetic fields across common geometries. In a quasi-experimental, non-equivalent groups design (n=44; six weeks; same instructor), the experimental group used AI-assisted modeling while the comparison group relied on traditional resources. Outcomes included conceptual understanding (pre/post concept test), quality of simulations (programming rubric; inter-rater ICC =0.82), and perceived usefulness of AI (Likert survey). Analysis showed an 7.3-point post-test advantage for the AI group (95 % CI 3.5-11.1), a large standardized effect (Hedges' g 0.90), and a positive perception- performance association (ß 0. 7 8; 95% CI 0.52-1.04). Students using AI produced more accurate, interpretable simulations and improved spatial reasoning. Findings support careful integration of generative AI to strengthen conceptual reasoning and modeling in early electromagnetism. Limitations include the quasi-experimental design and single-context implementation; future work will test longitudinal retention and transferability.