Teaching Computational Thinking: A Tangible Card-Based Approach with Multimodal Large Language Models and TiniScript

Guzman Ramos G.G.; Calderon M. · Proceedings - International Conference of the Chilean Computer Science Society, SCCC (2025) · DOI: 10.1109/sccc67219.2025.11420641

Resumen

Introducing programming into low-resource schools requires solutions that work with minimal hardware while sustaining engagement. We present a tangible programming approach based on printed cards that encode basic commands (e.g., move forward, turn). Students arrange cards on a board to build algorithms; a single teacher smartphone captures the sequence, which a Multimodal Large Language Model (MLLM) recognizes and translates into TiniScript code executable on an educational robot. Technically, the image→code pipeline attains high fidelity (CodeBLEU = 0.9671, CodeBERTScore F1 = 0.9988).We complement this with a classroom pilot in Lima, Peru (six 5th-6th grade classes; N=60), using a quasi-experimental design: robot+MLLM (30 students) vs. unplugged without robot (30 students). A five-item post-test on algorithmic thinking shows that, with the robot, 94% scored 5/5 and 6% scored 4/5; without the robot, 65% scored 5/5, 30% scored 4/5, and 5% scored 3/5. The perfect-score rate increases by ≈27 percentage points (two-proportion z ≈ 2.58, p ≈ 0.01). Beyond its explicit focus on algorithmic thinking, classroom observations indicate the approach also promotes collaborative work. By avoiding one-to-one devices, this AI-assisted tangible method offers a scalable pathway to bring computational thinking to under-resourced schools.

Instituciones: Universidad ESAN